xref: /freebsd/sys/kern/kern_descrip.c (revision 3b216bfb6cce24aa84519315138be8d23ac5d613)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98     "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 
102 MALLOC_DECLARE(M_FADVISE);
103 
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
107 VFS_SMR_DECLARE;
108 
109 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
110 		    struct thread *td, bool holdleaders, bool audit);
111 static int	fd_first_free(struct filedesc *fdp, int low, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the last used fd.
217  *
218  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219  * Otherwise use fdlastfile.
220  */
221 int
222 fdlastfile_single(struct filedesc *fdp)
223 {
224 	NDSLOTTYPE *map = fdp->fd_map;
225 	int off, minoff;
226 
227 	off = NDSLOT(fdp->fd_nfiles - 1);
228 	for (minoff = NDSLOT(0); off >= minoff; --off)
229 		if (map[off] != 0)
230 			return (off * NDENTRIES + flsl(map[off]) - 1);
231 	return (-1);
232 }
233 
234 int
235 fdlastfile(struct filedesc *fdp)
236 {
237 
238 	FILEDESC_LOCK_ASSERT(fdp);
239 	return (fdlastfile_single(fdp));
240 }
241 
242 static int
243 fdisused(struct filedesc *fdp, int fd)
244 {
245 
246 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 
249 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
250 }
251 
252 /*
253  * Mark a file descriptor as used.
254  */
255 static void
256 fdused_init(struct filedesc *fdp, int fd)
257 {
258 
259 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 
261 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
262 }
263 
264 static void
265 fdused(struct filedesc *fdp, int fd)
266 {
267 
268 	FILEDESC_XLOCK_ASSERT(fdp);
269 
270 	fdused_init(fdp, fd);
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile++;
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	FILEDESC_XLOCK_ASSERT(fdp);
311 	fde = &fdp->fd_ofiles[fd];
312 #ifdef CAPABILITIES
313 	seqc_write_begin(&fde->fde_seqc);
314 #endif
315 	fde->fde_file = NULL;
316 #ifdef CAPABILITIES
317 	seqc_write_end(&fde->fde_seqc);
318 #endif
319 	fdefree_last(fde);
320 	fdunused(fdp, fd);
321 }
322 
323 /*
324  * System calls on descriptors.
325  */
326 #ifndef _SYS_SYSPROTO_H_
327 struct getdtablesize_args {
328 	int	dummy;
329 };
330 #endif
331 /* ARGSUSED */
332 int
333 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
334 {
335 #ifdef	RACCT
336 	uint64_t lim;
337 #endif
338 
339 	td->td_retval[0] = getmaxfd(td);
340 #ifdef	RACCT
341 	PROC_LOCK(td->td_proc);
342 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
343 	PROC_UNLOCK(td->td_proc);
344 	if (lim < td->td_retval[0])
345 		td->td_retval[0] = lim;
346 #endif
347 	return (0);
348 }
349 
350 /*
351  * Duplicate a file descriptor to a particular value.
352  *
353  * Note: keep in mind that a potential race condition exists when closing
354  * descriptors from a shared descriptor table (via rfork).
355  */
356 #ifndef _SYS_SYSPROTO_H_
357 struct dup2_args {
358 	u_int	from;
359 	u_int	to;
360 };
361 #endif
362 /* ARGSUSED */
363 int
364 sys_dup2(struct thread *td, struct dup2_args *uap)
365 {
366 
367 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
368 }
369 
370 /*
371  * Duplicate a file descriptor.
372  */
373 #ifndef _SYS_SYSPROTO_H_
374 struct dup_args {
375 	u_int	fd;
376 };
377 #endif
378 /* ARGSUSED */
379 int
380 sys_dup(struct thread *td, struct dup_args *uap)
381 {
382 
383 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
384 }
385 
386 /*
387  * The file control system call.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct fcntl_args {
391 	int	fd;
392 	int	cmd;
393 	long	arg;
394 };
395 #endif
396 /* ARGSUSED */
397 int
398 sys_fcntl(struct thread *td, struct fcntl_args *uap)
399 {
400 
401 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
402 }
403 
404 int
405 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
406 {
407 	struct flock fl;
408 	struct __oflock ofl;
409 	intptr_t arg1;
410 	int error, newcmd;
411 
412 	error = 0;
413 	newcmd = cmd;
414 	switch (cmd) {
415 	case F_OGETLK:
416 	case F_OSETLK:
417 	case F_OSETLKW:
418 		/*
419 		 * Convert old flock structure to new.
420 		 */
421 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
422 		fl.l_start = ofl.l_start;
423 		fl.l_len = ofl.l_len;
424 		fl.l_pid = ofl.l_pid;
425 		fl.l_type = ofl.l_type;
426 		fl.l_whence = ofl.l_whence;
427 		fl.l_sysid = 0;
428 
429 		switch (cmd) {
430 		case F_OGETLK:
431 			newcmd = F_GETLK;
432 			break;
433 		case F_OSETLK:
434 			newcmd = F_SETLK;
435 			break;
436 		case F_OSETLKW:
437 			newcmd = F_SETLKW;
438 			break;
439 		}
440 		arg1 = (intptr_t)&fl;
441 		break;
442 	case F_GETLK:
443 	case F_SETLK:
444 	case F_SETLKW:
445 	case F_SETLK_REMOTE:
446 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
447 		arg1 = (intptr_t)&fl;
448 		break;
449 	default:
450 		arg1 = arg;
451 		break;
452 	}
453 	if (error)
454 		return (error);
455 	error = kern_fcntl(td, fd, newcmd, arg1);
456 	if (error)
457 		return (error);
458 	if (cmd == F_OGETLK) {
459 		ofl.l_start = fl.l_start;
460 		ofl.l_len = fl.l_len;
461 		ofl.l_pid = fl.l_pid;
462 		ofl.l_type = fl.l_type;
463 		ofl.l_whence = fl.l_whence;
464 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
465 	} else if (cmd == F_GETLK) {
466 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
467 	}
468 	return (error);
469 }
470 
471 int
472 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 {
474 	struct filedesc *fdp;
475 	struct flock *flp;
476 	struct file *fp, *fp2;
477 	struct filedescent *fde;
478 	struct proc *p;
479 	struct vnode *vp;
480 	struct mount *mp;
481 	int error, flg, seals, tmp;
482 	uint64_t bsize;
483 	off_t foffset;
484 
485 	error = 0;
486 	flg = F_POSIX;
487 	p = td->td_proc;
488 	fdp = p->p_fd;
489 
490 	AUDIT_ARG_FD(cmd);
491 	AUDIT_ARG_CMD(cmd);
492 	switch (cmd) {
493 	case F_DUPFD:
494 		tmp = arg;
495 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
496 		break;
497 
498 	case F_DUPFD_CLOEXEC:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
501 		break;
502 
503 	case F_DUP2FD:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD_CLOEXEC:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
511 		break;
512 
513 	case F_GETFD:
514 		error = EBADF;
515 		FILEDESC_SLOCK(fdp);
516 		fde = fdeget_locked(fdp, fd);
517 		if (fde != NULL) {
518 			td->td_retval[0] =
519 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
520 			error = 0;
521 		}
522 		FILEDESC_SUNLOCK(fdp);
523 		break;
524 
525 	case F_SETFD:
526 		error = EBADF;
527 		FILEDESC_XLOCK(fdp);
528 		fde = fdeget_locked(fdp, fd);
529 		if (fde != NULL) {
530 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
531 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
532 			error = 0;
533 		}
534 		FILEDESC_XUNLOCK(fdp);
535 		break;
536 
537 	case F_GETFL:
538 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
539 		if (error != 0)
540 			break;
541 		td->td_retval[0] = OFLAGS(fp->f_flag);
542 		fdrop(fp, td);
543 		break;
544 
545 	case F_SETFL:
546 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
547 		if (error != 0)
548 			break;
549 		do {
550 			tmp = flg = fp->f_flag;
551 			tmp &= ~FCNTLFLAGS;
552 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
553 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
554 		tmp = fp->f_flag & FNONBLOCK;
555 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
556 		if (error != 0) {
557 			fdrop(fp, td);
558 			break;
559 		}
560 		tmp = fp->f_flag & FASYNC;
561 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
562 		if (error == 0) {
563 			fdrop(fp, td);
564 			break;
565 		}
566 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
567 		tmp = 0;
568 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
569 		fdrop(fp, td);
570 		break;
571 
572 	case F_GETOWN:
573 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
574 		if (error != 0)
575 			break;
576 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 		if (error == 0)
578 			td->td_retval[0] = tmp;
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_SETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		tmp = arg;
587 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
588 		fdrop(fp, td);
589 		break;
590 
591 	case F_SETLK_REMOTE:
592 		error = priv_check(td, PRIV_NFS_LOCKD);
593 		if (error != 0)
594 			return (error);
595 		flg = F_REMOTE;
596 		goto do_setlk;
597 
598 	case F_SETLKW:
599 		flg |= F_WAIT;
600 		/* FALLTHROUGH F_SETLK */
601 
602 	case F_SETLK:
603 	do_setlk:
604 		flp = (struct flock *)arg;
605 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
606 			error = EINVAL;
607 			break;
608 		}
609 
610 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
611 		if (error != 0)
612 			break;
613 		if (fp->f_type != DTYPE_VNODE) {
614 			error = EBADF;
615 			fdrop(fp, td);
616 			break;
617 		}
618 
619 		if (flp->l_whence == SEEK_CUR) {
620 			foffset = foffset_get(fp);
621 			if (foffset < 0 ||
622 			    (flp->l_start > 0 &&
623 			     foffset > OFF_MAX - flp->l_start)) {
624 				error = EOVERFLOW;
625 				fdrop(fp, td);
626 				break;
627 			}
628 			flp->l_start += foffset;
629 		}
630 
631 		vp = fp->f_vnode;
632 		switch (flp->l_type) {
633 		case F_RDLCK:
634 			if ((fp->f_flag & FREAD) == 0) {
635 				error = EBADF;
636 				break;
637 			}
638 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
639 				PROC_LOCK(p->p_leader);
640 				p->p_leader->p_flag |= P_ADVLOCK;
641 				PROC_UNLOCK(p->p_leader);
642 			}
643 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
644 			    flp, flg);
645 			break;
646 		case F_WRLCK:
647 			if ((fp->f_flag & FWRITE) == 0) {
648 				error = EBADF;
649 				break;
650 			}
651 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
652 				PROC_LOCK(p->p_leader);
653 				p->p_leader->p_flag |= P_ADVLOCK;
654 				PROC_UNLOCK(p->p_leader);
655 			}
656 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
657 			    flp, flg);
658 			break;
659 		case F_UNLCK:
660 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
661 			    flp, flg);
662 			break;
663 		case F_UNLCKSYS:
664 			if (flg != F_REMOTE) {
665 				error = EINVAL;
666 				break;
667 			}
668 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
669 			    F_UNLCKSYS, flp, flg);
670 			break;
671 		default:
672 			error = EINVAL;
673 			break;
674 		}
675 		if (error != 0 || flp->l_type == F_UNLCK ||
676 		    flp->l_type == F_UNLCKSYS) {
677 			fdrop(fp, td);
678 			break;
679 		}
680 
681 		/*
682 		 * Check for a race with close.
683 		 *
684 		 * The vnode is now advisory locked (or unlocked, but this case
685 		 * is not really important) as the caller requested.
686 		 * We had to drop the filedesc lock, so we need to recheck if
687 		 * the descriptor is still valid, because if it was closed
688 		 * in the meantime we need to remove advisory lock from the
689 		 * vnode - close on any descriptor leading to an advisory
690 		 * locked vnode, removes that lock.
691 		 * We will return 0 on purpose in that case, as the result of
692 		 * successful advisory lock might have been externally visible
693 		 * already. This is fine - effectively we pretend to the caller
694 		 * that the closing thread was a bit slower and that the
695 		 * advisory lock succeeded before the close.
696 		 */
697 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
698 		if (error != 0) {
699 			fdrop(fp, td);
700 			break;
701 		}
702 		if (fp != fp2) {
703 			flp->l_whence = SEEK_SET;
704 			flp->l_start = 0;
705 			flp->l_len = 0;
706 			flp->l_type = F_UNLCK;
707 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
708 			    F_UNLCK, flp, F_POSIX);
709 		}
710 		fdrop(fp, td);
711 		fdrop(fp2, td);
712 		break;
713 
714 	case F_GETLK:
715 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
716 		if (error != 0)
717 			break;
718 		if (fp->f_type != DTYPE_VNODE) {
719 			error = EBADF;
720 			fdrop(fp, td);
721 			break;
722 		}
723 		flp = (struct flock *)arg;
724 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
725 		    flp->l_type != F_UNLCK) {
726 			error = EINVAL;
727 			fdrop(fp, td);
728 			break;
729 		}
730 		if (flp->l_whence == SEEK_CUR) {
731 			foffset = foffset_get(fp);
732 			if ((flp->l_start > 0 &&
733 			    foffset > OFF_MAX - flp->l_start) ||
734 			    (flp->l_start < 0 &&
735 			    foffset < OFF_MIN - flp->l_start)) {
736 				error = EOVERFLOW;
737 				fdrop(fp, td);
738 				break;
739 			}
740 			flp->l_start += foffset;
741 		}
742 		vp = fp->f_vnode;
743 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
744 		    F_POSIX);
745 		fdrop(fp, td);
746 		break;
747 
748 	case F_ADD_SEALS:
749 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
750 		if (error != 0)
751 			break;
752 		error = fo_add_seals(fp, arg);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_GET_SEALS:
757 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
758 		if (error != 0)
759 			break;
760 		if (fo_get_seals(fp, &seals) == 0)
761 			td->td_retval[0] = seals;
762 		else
763 			error = EINVAL;
764 		fdrop(fp, td);
765 		break;
766 
767 	case F_RDAHEAD:
768 		arg = arg ? 128 * 1024: 0;
769 		/* FALLTHROUGH */
770 	case F_READAHEAD:
771 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
772 		if (error != 0)
773 			break;
774 		if (fp->f_type != DTYPE_VNODE) {
775 			fdrop(fp, td);
776 			error = EBADF;
777 			break;
778 		}
779 		vp = fp->f_vnode;
780 		if (vp->v_type != VREG) {
781 			fdrop(fp, td);
782 			error = ENOTTY;
783 			break;
784 		}
785 
786 		/*
787 		 * Exclusive lock synchronizes against f_seqcount reads and
788 		 * writes in sequential_heuristic().
789 		 */
790 		error = vn_lock(vp, LK_EXCLUSIVE);
791 		if (error != 0) {
792 			fdrop(fp, td);
793 			break;
794 		}
795 		if (arg >= 0) {
796 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
797 			arg = MIN(arg, INT_MAX - bsize + 1);
798 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
799 			    (arg + bsize - 1) / bsize);
800 			atomic_set_int(&fp->f_flag, FRDAHEAD);
801 		} else {
802 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
803 		}
804 		VOP_UNLOCK(vp);
805 		fdrop(fp, td);
806 		break;
807 
808 	case F_ISUNIONSTACK:
809 		/*
810 		 * Check if the vnode is part of a union stack (either the
811 		 * "union" flag from mount(2) or unionfs).
812 		 *
813 		 * Prior to introduction of this op libc's readdir would call
814 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
815 		 * data just to check fs name and a mount flag.
816 		 *
817 		 * Fixing the code to handle everything in the kernel instead
818 		 * is a non-trivial endeavor and has low priority, thus this
819 		 * horrible kludge facilitates the current behavior in a much
820 		 * cheaper manner until someone(tm) sorts this out.
821 		 */
822 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
823 		if (error != 0)
824 			break;
825 		if (fp->f_type != DTYPE_VNODE) {
826 			fdrop(fp, td);
827 			error = EBADF;
828 			break;
829 		}
830 		vp = fp->f_vnode;
831 		/*
832 		 * Since we don't prevent dooming the vnode even non-null mp
833 		 * found can become immediately stale. This is tolerable since
834 		 * mount points are type-stable (providing safe memory access)
835 		 * and any vfs op on this vnode going forward will return an
836 		 * error (meaning return value in this case is meaningless).
837 		 */
838 		mp = atomic_load_ptr(&vp->v_mount);
839 		if (__predict_false(mp == NULL)) {
840 			fdrop(fp, td);
841 			error = EBADF;
842 			break;
843 		}
844 		td->td_retval[0] = 0;
845 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
846 		    mp->mnt_flag & MNT_UNION)
847 			td->td_retval[0] = 1;
848 		fdrop(fp, td);
849 		break;
850 
851 	default:
852 		error = EINVAL;
853 		break;
854 	}
855 	return (error);
856 }
857 
858 static int
859 getmaxfd(struct thread *td)
860 {
861 
862 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
863 }
864 
865 /*
866  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
867  */
868 int
869 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
870 {
871 	struct filedesc *fdp;
872 	struct filedescent *oldfde, *newfde;
873 	struct proc *p;
874 	struct file *delfp, *oldfp;
875 	u_long *oioctls, *nioctls;
876 	int error, maxfd;
877 
878 	p = td->td_proc;
879 	fdp = p->p_fd;
880 	oioctls = NULL;
881 
882 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
883 	MPASS(mode < FDDUP_LASTMODE);
884 
885 	AUDIT_ARG_FD(old);
886 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
887 
888 	/*
889 	 * Verify we have a valid descriptor to dup from and possibly to
890 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
891 	 * return EINVAL when the new descriptor is out of bounds.
892 	 */
893 	if (old < 0)
894 		return (EBADF);
895 	if (new < 0)
896 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
897 	maxfd = getmaxfd(td);
898 	if (new >= maxfd)
899 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
900 
901 	error = EBADF;
902 	FILEDESC_XLOCK(fdp);
903 	if (fget_locked(fdp, old) == NULL)
904 		goto unlock;
905 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
906 		td->td_retval[0] = new;
907 		if (flags & FDDUP_FLAG_CLOEXEC)
908 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
909 		error = 0;
910 		goto unlock;
911 	}
912 
913 	oldfde = &fdp->fd_ofiles[old];
914 	oldfp = oldfde->fde_file;
915 	if (!fhold(oldfp))
916 		goto unlock;
917 
918 	/*
919 	 * If the caller specified a file descriptor, make sure the file
920 	 * table is large enough to hold it, and grab it.  Otherwise, just
921 	 * allocate a new descriptor the usual way.
922 	 */
923 	switch (mode) {
924 	case FDDUP_NORMAL:
925 	case FDDUP_FCNTL:
926 		if ((error = fdalloc(td, new, &new)) != 0) {
927 			fdrop(oldfp, td);
928 			goto unlock;
929 		}
930 		break;
931 	case FDDUP_MUSTREPLACE:
932 		/* Target file descriptor must exist. */
933 		if (fget_locked(fdp, new) == NULL) {
934 			fdrop(oldfp, td);
935 			goto unlock;
936 		}
937 		break;
938 	case FDDUP_FIXED:
939 		if (new >= fdp->fd_nfiles) {
940 			/*
941 			 * The resource limits are here instead of e.g.
942 			 * fdalloc(), because the file descriptor table may be
943 			 * shared between processes, so we can't really use
944 			 * racct_add()/racct_sub().  Instead of counting the
945 			 * number of actually allocated descriptors, just put
946 			 * the limit on the size of the file descriptor table.
947 			 */
948 #ifdef RACCT
949 			if (RACCT_ENABLED()) {
950 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
951 				if (error != 0) {
952 					error = EMFILE;
953 					fdrop(oldfp, td);
954 					goto unlock;
955 				}
956 			}
957 #endif
958 			fdgrowtable_exp(fdp, new + 1);
959 		}
960 		if (!fdisused(fdp, new))
961 			fdused(fdp, new);
962 		break;
963 	default:
964 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
965 	}
966 
967 	KASSERT(old != new, ("new fd is same as old"));
968 
969 	/* Refetch oldfde because the table may have grown and old one freed. */
970 	oldfde = &fdp->fd_ofiles[old];
971 	KASSERT(oldfp == oldfde->fde_file,
972 	    ("fdt_ofiles shift from growth observed at fd %d",
973 	    old));
974 
975 	newfde = &fdp->fd_ofiles[new];
976 	delfp = newfde->fde_file;
977 
978 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
979 
980 	/*
981 	 * Duplicate the source descriptor.
982 	 */
983 #ifdef CAPABILITIES
984 	seqc_write_begin(&newfde->fde_seqc);
985 #endif
986 	oioctls = filecaps_free_prep(&newfde->fde_caps);
987 	memcpy(newfde, oldfde, fde_change_size);
988 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
989 	    nioctls);
990 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
991 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
992 	else
993 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
994 #ifdef CAPABILITIES
995 	seqc_write_end(&newfde->fde_seqc);
996 #endif
997 	td->td_retval[0] = new;
998 
999 	error = 0;
1000 
1001 	if (delfp != NULL) {
1002 		(void) closefp(fdp, new, delfp, td, true, false);
1003 		FILEDESC_UNLOCK_ASSERT(fdp);
1004 	} else {
1005 unlock:
1006 		FILEDESC_XUNLOCK(fdp);
1007 	}
1008 
1009 	filecaps_free_finish(oioctls);
1010 	return (error);
1011 }
1012 
1013 static void
1014 sigiofree(struct sigio *sigio)
1015 {
1016 	crfree(sigio->sio_ucred);
1017 	free(sigio, M_SIGIO);
1018 }
1019 
1020 static struct sigio *
1021 funsetown_locked(struct sigio *sigio)
1022 {
1023 	struct proc *p;
1024 	struct pgrp *pg;
1025 
1026 	SIGIO_ASSERT_LOCKED();
1027 
1028 	if (sigio == NULL)
1029 		return (NULL);
1030 	*(sigio->sio_myref) = NULL;
1031 	if (sigio->sio_pgid < 0) {
1032 		pg = sigio->sio_pgrp;
1033 		PGRP_LOCK(pg);
1034 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1035 		    sigio, sio_pgsigio);
1036 		PGRP_UNLOCK(pg);
1037 	} else {
1038 		p = sigio->sio_proc;
1039 		PROC_LOCK(p);
1040 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1041 		    sigio, sio_pgsigio);
1042 		PROC_UNLOCK(p);
1043 	}
1044 	return (sigio);
1045 }
1046 
1047 /*
1048  * If sigio is on the list associated with a process or process group,
1049  * disable signalling from the device, remove sigio from the list and
1050  * free sigio.
1051  */
1052 void
1053 funsetown(struct sigio **sigiop)
1054 {
1055 	struct sigio *sigio;
1056 
1057 	/* Racy check, consumers must provide synchronization. */
1058 	if (*sigiop == NULL)
1059 		return;
1060 
1061 	SIGIO_LOCK();
1062 	sigio = funsetown_locked(*sigiop);
1063 	SIGIO_UNLOCK();
1064 	if (sigio != NULL)
1065 		sigiofree(sigio);
1066 }
1067 
1068 /*
1069  * Free a list of sigio structures.  The caller must ensure that new sigio
1070  * structures cannot be added after this point.  For process groups this is
1071  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1072  * as an interlock.
1073  */
1074 void
1075 funsetownlst(struct sigiolst *sigiolst)
1076 {
1077 	struct proc *p;
1078 	struct pgrp *pg;
1079 	struct sigio *sigio, *tmp;
1080 
1081 	/* Racy check. */
1082 	sigio = SLIST_FIRST(sigiolst);
1083 	if (sigio == NULL)
1084 		return;
1085 
1086 	p = NULL;
1087 	pg = NULL;
1088 
1089 	SIGIO_LOCK();
1090 	sigio = SLIST_FIRST(sigiolst);
1091 	if (sigio == NULL) {
1092 		SIGIO_UNLOCK();
1093 		return;
1094 	}
1095 
1096 	/*
1097 	 * Every entry of the list should belong to a single proc or pgrp.
1098 	 */
1099 	if (sigio->sio_pgid < 0) {
1100 		pg = sigio->sio_pgrp;
1101 		sx_assert(&proctree_lock, SX_XLOCKED);
1102 		PGRP_LOCK(pg);
1103 	} else /* if (sigio->sio_pgid > 0) */ {
1104 		p = sigio->sio_proc;
1105 		PROC_LOCK(p);
1106 		KASSERT((p->p_flag & P_WEXIT) != 0,
1107 		    ("%s: process %p is not exiting", __func__, p));
1108 	}
1109 
1110 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1111 		*sigio->sio_myref = NULL;
1112 		if (pg != NULL) {
1113 			KASSERT(sigio->sio_pgid < 0,
1114 			    ("Proc sigio in pgrp sigio list"));
1115 			KASSERT(sigio->sio_pgrp == pg,
1116 			    ("Bogus pgrp in sigio list"));
1117 		} else /* if (p != NULL) */ {
1118 			KASSERT(sigio->sio_pgid > 0,
1119 			    ("Pgrp sigio in proc sigio list"));
1120 			KASSERT(sigio->sio_proc == p,
1121 			    ("Bogus proc in sigio list"));
1122 		}
1123 	}
1124 
1125 	if (pg != NULL)
1126 		PGRP_UNLOCK(pg);
1127 	else
1128 		PROC_UNLOCK(p);
1129 	SIGIO_UNLOCK();
1130 
1131 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1132 		sigiofree(sigio);
1133 }
1134 
1135 /*
1136  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1137  *
1138  * After permission checking, add a sigio structure to the sigio list for
1139  * the process or process group.
1140  */
1141 int
1142 fsetown(pid_t pgid, struct sigio **sigiop)
1143 {
1144 	struct proc *proc;
1145 	struct pgrp *pgrp;
1146 	struct sigio *osigio, *sigio;
1147 	int ret;
1148 
1149 	if (pgid == 0) {
1150 		funsetown(sigiop);
1151 		return (0);
1152 	}
1153 
1154 	ret = 0;
1155 
1156 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1157 	sigio->sio_pgid = pgid;
1158 	sigio->sio_ucred = crhold(curthread->td_ucred);
1159 	sigio->sio_myref = sigiop;
1160 
1161 	sx_slock(&proctree_lock);
1162 	SIGIO_LOCK();
1163 	osigio = funsetown_locked(*sigiop);
1164 	if (pgid > 0) {
1165 		proc = pfind(pgid);
1166 		if (proc == NULL) {
1167 			ret = ESRCH;
1168 			goto fail;
1169 		}
1170 
1171 		/*
1172 		 * Policy - Don't allow a process to FSETOWN a process
1173 		 * in another session.
1174 		 *
1175 		 * Remove this test to allow maximum flexibility or
1176 		 * restrict FSETOWN to the current process or process
1177 		 * group for maximum safety.
1178 		 */
1179 		if (proc->p_session != curthread->td_proc->p_session) {
1180 			PROC_UNLOCK(proc);
1181 			ret = EPERM;
1182 			goto fail;
1183 		}
1184 
1185 		sigio->sio_proc = proc;
1186 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1187 		PROC_UNLOCK(proc);
1188 	} else /* if (pgid < 0) */ {
1189 		pgrp = pgfind(-pgid);
1190 		if (pgrp == NULL) {
1191 			ret = ESRCH;
1192 			goto fail;
1193 		}
1194 
1195 		/*
1196 		 * Policy - Don't allow a process to FSETOWN a process
1197 		 * in another session.
1198 		 *
1199 		 * Remove this test to allow maximum flexibility or
1200 		 * restrict FSETOWN to the current process or process
1201 		 * group for maximum safety.
1202 		 */
1203 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1204 			PGRP_UNLOCK(pgrp);
1205 			ret = EPERM;
1206 			goto fail;
1207 		}
1208 
1209 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1210 		sigio->sio_pgrp = pgrp;
1211 		PGRP_UNLOCK(pgrp);
1212 	}
1213 	sx_sunlock(&proctree_lock);
1214 	*sigiop = sigio;
1215 	SIGIO_UNLOCK();
1216 	if (osigio != NULL)
1217 		sigiofree(osigio);
1218 	return (0);
1219 
1220 fail:
1221 	SIGIO_UNLOCK();
1222 	sx_sunlock(&proctree_lock);
1223 	sigiofree(sigio);
1224 	if (osigio != NULL)
1225 		sigiofree(osigio);
1226 	return (ret);
1227 }
1228 
1229 /*
1230  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1231  */
1232 pid_t
1233 fgetown(struct sigio **sigiop)
1234 {
1235 	pid_t pgid;
1236 
1237 	SIGIO_LOCK();
1238 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1239 	SIGIO_UNLOCK();
1240 	return (pgid);
1241 }
1242 
1243 static int
1244 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1245     bool audit)
1246 {
1247 	int error;
1248 
1249 	FILEDESC_XLOCK_ASSERT(fdp);
1250 
1251 	/*
1252 	 * We now hold the fp reference that used to be owned by the
1253 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1254 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1255 	 * added, and deleteing a knote for the new fd.
1256 	 */
1257 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1258 		knote_fdclose(td, fd);
1259 
1260 	/*
1261 	 * We need to notify mqueue if the object is of type mqueue.
1262 	 */
1263 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1264 		mq_fdclose(td, fd, fp);
1265 	FILEDESC_XUNLOCK(fdp);
1266 
1267 #ifdef AUDIT
1268 	if (AUDITING_TD(td) && audit)
1269 		audit_sysclose(td, fd, fp);
1270 #endif
1271 	error = closef(fp, td);
1272 
1273 	/*
1274 	 * All paths leading up to closefp() will have already removed or
1275 	 * replaced the fd in the filedesc table, so a restart would not
1276 	 * operate on the same file.
1277 	 */
1278 	if (error == ERESTART)
1279 		error = EINTR;
1280 
1281 	return (error);
1282 }
1283 
1284 static int
1285 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1286     bool holdleaders, bool audit)
1287 {
1288 	int error;
1289 
1290 	FILEDESC_XLOCK_ASSERT(fdp);
1291 
1292 	if (holdleaders) {
1293 		if (td->td_proc->p_fdtol != NULL) {
1294 			/*
1295 			 * Ask fdfree() to sleep to ensure that all relevant
1296 			 * process leaders can be traversed in closef().
1297 			 */
1298 			fdp->fd_holdleaderscount++;
1299 		} else {
1300 			holdleaders = false;
1301 		}
1302 	}
1303 
1304 	error = closefp_impl(fdp, fd, fp, td, audit);
1305 	if (holdleaders) {
1306 		FILEDESC_XLOCK(fdp);
1307 		fdp->fd_holdleaderscount--;
1308 		if (fdp->fd_holdleaderscount == 0 &&
1309 		    fdp->fd_holdleaderswakeup != 0) {
1310 			fdp->fd_holdleaderswakeup = 0;
1311 			wakeup(&fdp->fd_holdleaderscount);
1312 		}
1313 		FILEDESC_XUNLOCK(fdp);
1314 	}
1315 	return (error);
1316 }
1317 
1318 static int
1319 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1320     bool holdleaders, bool audit)
1321 {
1322 
1323 	FILEDESC_XLOCK_ASSERT(fdp);
1324 
1325 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1326 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1327 	} else {
1328 		return (closefp_impl(fdp, fd, fp, td, audit));
1329 	}
1330 }
1331 
1332 /*
1333  * Close a file descriptor.
1334  */
1335 #ifndef _SYS_SYSPROTO_H_
1336 struct close_args {
1337 	int     fd;
1338 };
1339 #endif
1340 /* ARGSUSED */
1341 int
1342 sys_close(struct thread *td, struct close_args *uap)
1343 {
1344 
1345 	return (kern_close(td, uap->fd));
1346 }
1347 
1348 int
1349 kern_close(struct thread *td, int fd)
1350 {
1351 	struct filedesc *fdp;
1352 	struct file *fp;
1353 
1354 	fdp = td->td_proc->p_fd;
1355 
1356 	FILEDESC_XLOCK(fdp);
1357 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1358 		FILEDESC_XUNLOCK(fdp);
1359 		return (EBADF);
1360 	}
1361 	fdfree(fdp, fd);
1362 
1363 	/* closefp() drops the FILEDESC lock for us. */
1364 	return (closefp(fdp, fd, fp, td, true, true));
1365 }
1366 
1367 int
1368 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1369 {
1370 	struct filedesc *fdp;
1371 	const struct fdescenttbl *fdt;
1372 	struct file *fp;
1373 	int fd;
1374 
1375 	/*
1376 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1377 	 * open should not be a usage error.  From a close_range() perspective,
1378 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1379 	 * be a usage error as all fd above 3 are in-fact already closed.
1380 	 */
1381 	if (highfd < lowfd) {
1382 		return (EINVAL);
1383 	}
1384 
1385 	fdp = td->td_proc->p_fd;
1386 	FILEDESC_XLOCK(fdp);
1387 	fdt = atomic_load_ptr(&fdp->fd_files);
1388 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1389 	fd = lowfd;
1390 	if (__predict_false(fd > highfd)) {
1391 		goto out_locked;
1392 	}
1393 	for (;;) {
1394 		fp = fdt->fdt_ofiles[fd].fde_file;
1395 		if (fp == NULL) {
1396 			if (fd == highfd)
1397 				goto out_locked;
1398 		} else {
1399 			fdfree(fdp, fd);
1400 			(void) closefp(fdp, fd, fp, td, true, true);
1401 			if (fd == highfd)
1402 				goto out_unlocked;
1403 			FILEDESC_XLOCK(fdp);
1404 			fdt = atomic_load_ptr(&fdp->fd_files);
1405 		}
1406 		fd++;
1407 	}
1408 out_locked:
1409 	FILEDESC_XUNLOCK(fdp);
1410 out_unlocked:
1411 	return (0);
1412 }
1413 
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct close_range_args {
1416 	u_int	lowfd;
1417 	u_int	highfd;
1418 	int	flags;
1419 };
1420 #endif
1421 int
1422 sys_close_range(struct thread *td, struct close_range_args *uap)
1423 {
1424 
1425 	/* No flags currently defined */
1426 	if (uap->flags != 0)
1427 		return (EINVAL);
1428 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1429 }
1430 
1431 #ifdef COMPAT_FREEBSD12
1432 /*
1433  * Close open file descriptors.
1434  */
1435 #ifndef _SYS_SYSPROTO_H_
1436 struct freebsd12_closefrom_args {
1437 	int	lowfd;
1438 };
1439 #endif
1440 /* ARGSUSED */
1441 int
1442 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1443 {
1444 	u_int lowfd;
1445 
1446 	AUDIT_ARG_FD(uap->lowfd);
1447 
1448 	/*
1449 	 * Treat negative starting file descriptor values identical to
1450 	 * closefrom(0) which closes all files.
1451 	 */
1452 	lowfd = MAX(0, uap->lowfd);
1453 	return (kern_close_range(td, lowfd, ~0U));
1454 }
1455 #endif	/* COMPAT_FREEBSD12 */
1456 
1457 #if defined(COMPAT_43)
1458 /*
1459  * Return status information about a file descriptor.
1460  */
1461 #ifndef _SYS_SYSPROTO_H_
1462 struct ofstat_args {
1463 	int	fd;
1464 	struct	ostat *sb;
1465 };
1466 #endif
1467 /* ARGSUSED */
1468 int
1469 ofstat(struct thread *td, struct ofstat_args *uap)
1470 {
1471 	struct ostat oub;
1472 	struct stat ub;
1473 	int error;
1474 
1475 	error = kern_fstat(td, uap->fd, &ub);
1476 	if (error == 0) {
1477 		cvtstat(&ub, &oub);
1478 		error = copyout(&oub, uap->sb, sizeof(oub));
1479 	}
1480 	return (error);
1481 }
1482 #endif /* COMPAT_43 */
1483 
1484 #if defined(COMPAT_FREEBSD11)
1485 int
1486 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1487 {
1488 	struct stat sb;
1489 	struct freebsd11_stat osb;
1490 	int error;
1491 
1492 	error = kern_fstat(td, uap->fd, &sb);
1493 	if (error != 0)
1494 		return (error);
1495 	error = freebsd11_cvtstat(&sb, &osb);
1496 	if (error == 0)
1497 		error = copyout(&osb, uap->sb, sizeof(osb));
1498 	return (error);
1499 }
1500 #endif	/* COMPAT_FREEBSD11 */
1501 
1502 /*
1503  * Return status information about a file descriptor.
1504  */
1505 #ifndef _SYS_SYSPROTO_H_
1506 struct fstat_args {
1507 	int	fd;
1508 	struct	stat *sb;
1509 };
1510 #endif
1511 /* ARGSUSED */
1512 int
1513 sys_fstat(struct thread *td, struct fstat_args *uap)
1514 {
1515 	struct stat ub;
1516 	int error;
1517 
1518 	error = kern_fstat(td, uap->fd, &ub);
1519 	if (error == 0)
1520 		error = copyout(&ub, uap->sb, sizeof(ub));
1521 	return (error);
1522 }
1523 
1524 int
1525 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1526 {
1527 	struct file *fp;
1528 	int error;
1529 
1530 	AUDIT_ARG_FD(fd);
1531 
1532 	error = fget(td, fd, &cap_fstat_rights, &fp);
1533 	if (__predict_false(error != 0))
1534 		return (error);
1535 
1536 	AUDIT_ARG_FILE(td->td_proc, fp);
1537 
1538 	error = fo_stat(fp, sbp, td->td_ucred, td);
1539 	fdrop(fp, td);
1540 #ifdef __STAT_TIME_T_EXT
1541 	sbp->st_atim_ext = 0;
1542 	sbp->st_mtim_ext = 0;
1543 	sbp->st_ctim_ext = 0;
1544 	sbp->st_btim_ext = 0;
1545 #endif
1546 #ifdef KTRACE
1547 	if (KTRPOINT(td, KTR_STRUCT))
1548 		ktrstat_error(sbp, error);
1549 #endif
1550 	return (error);
1551 }
1552 
1553 #if defined(COMPAT_FREEBSD11)
1554 /*
1555  * Return status information about a file descriptor.
1556  */
1557 #ifndef _SYS_SYSPROTO_H_
1558 struct freebsd11_nfstat_args {
1559 	int	fd;
1560 	struct	nstat *sb;
1561 };
1562 #endif
1563 /* ARGSUSED */
1564 int
1565 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1566 {
1567 	struct nstat nub;
1568 	struct stat ub;
1569 	int error;
1570 
1571 	error = kern_fstat(td, uap->fd, &ub);
1572 	if (error == 0) {
1573 		freebsd11_cvtnstat(&ub, &nub);
1574 		error = copyout(&nub, uap->sb, sizeof(nub));
1575 	}
1576 	return (error);
1577 }
1578 #endif /* COMPAT_FREEBSD11 */
1579 
1580 /*
1581  * Return pathconf information about a file descriptor.
1582  */
1583 #ifndef _SYS_SYSPROTO_H_
1584 struct fpathconf_args {
1585 	int	fd;
1586 	int	name;
1587 };
1588 #endif
1589 /* ARGSUSED */
1590 int
1591 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1592 {
1593 	long value;
1594 	int error;
1595 
1596 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1597 	if (error == 0)
1598 		td->td_retval[0] = value;
1599 	return (error);
1600 }
1601 
1602 int
1603 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1604 {
1605 	struct file *fp;
1606 	struct vnode *vp;
1607 	int error;
1608 
1609 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1610 	if (error != 0)
1611 		return (error);
1612 
1613 	if (name == _PC_ASYNC_IO) {
1614 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1615 		goto out;
1616 	}
1617 	vp = fp->f_vnode;
1618 	if (vp != NULL) {
1619 		vn_lock(vp, LK_SHARED | LK_RETRY);
1620 		error = VOP_PATHCONF(vp, name, valuep);
1621 		VOP_UNLOCK(vp);
1622 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1623 		if (name != _PC_PIPE_BUF) {
1624 			error = EINVAL;
1625 		} else {
1626 			*valuep = PIPE_BUF;
1627 			error = 0;
1628 		}
1629 	} else {
1630 		error = EOPNOTSUPP;
1631 	}
1632 out:
1633 	fdrop(fp, td);
1634 	return (error);
1635 }
1636 
1637 /*
1638  * Copy filecaps structure allocating memory for ioctls array if needed.
1639  *
1640  * The last parameter indicates whether the fdtable is locked. If it is not and
1641  * ioctls are encountered, copying fails and the caller must lock the table.
1642  *
1643  * Note that if the table was not locked, the caller has to check the relevant
1644  * sequence counter to determine whether the operation was successful.
1645  */
1646 bool
1647 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1648 {
1649 	size_t size;
1650 
1651 	if (src->fc_ioctls != NULL && !locked)
1652 		return (false);
1653 	memcpy(dst, src, sizeof(*src));
1654 	if (src->fc_ioctls == NULL)
1655 		return (true);
1656 
1657 	KASSERT(src->fc_nioctls > 0,
1658 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1659 
1660 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1661 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1662 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1663 	return (true);
1664 }
1665 
1666 static u_long *
1667 filecaps_copy_prep(const struct filecaps *src)
1668 {
1669 	u_long *ioctls;
1670 	size_t size;
1671 
1672 	if (__predict_true(src->fc_ioctls == NULL))
1673 		return (NULL);
1674 
1675 	KASSERT(src->fc_nioctls > 0,
1676 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1677 
1678 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1679 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1680 	return (ioctls);
1681 }
1682 
1683 static void
1684 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1685     u_long *ioctls)
1686 {
1687 	size_t size;
1688 
1689 	*dst = *src;
1690 	if (__predict_true(src->fc_ioctls == NULL)) {
1691 		MPASS(ioctls == NULL);
1692 		return;
1693 	}
1694 
1695 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1696 	dst->fc_ioctls = ioctls;
1697 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1698 }
1699 
1700 /*
1701  * Move filecaps structure to the new place and clear the old place.
1702  */
1703 void
1704 filecaps_move(struct filecaps *src, struct filecaps *dst)
1705 {
1706 
1707 	*dst = *src;
1708 	bzero(src, sizeof(*src));
1709 }
1710 
1711 /*
1712  * Fill the given filecaps structure with full rights.
1713  */
1714 static void
1715 filecaps_fill(struct filecaps *fcaps)
1716 {
1717 
1718 	CAP_ALL(&fcaps->fc_rights);
1719 	fcaps->fc_ioctls = NULL;
1720 	fcaps->fc_nioctls = -1;
1721 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1722 }
1723 
1724 /*
1725  * Free memory allocated within filecaps structure.
1726  */
1727 void
1728 filecaps_free(struct filecaps *fcaps)
1729 {
1730 
1731 	free(fcaps->fc_ioctls, M_FILECAPS);
1732 	bzero(fcaps, sizeof(*fcaps));
1733 }
1734 
1735 static u_long *
1736 filecaps_free_prep(struct filecaps *fcaps)
1737 {
1738 	u_long *ioctls;
1739 
1740 	ioctls = fcaps->fc_ioctls;
1741 	bzero(fcaps, sizeof(*fcaps));
1742 	return (ioctls);
1743 }
1744 
1745 static void
1746 filecaps_free_finish(u_long *ioctls)
1747 {
1748 
1749 	free(ioctls, M_FILECAPS);
1750 }
1751 
1752 /*
1753  * Validate the given filecaps structure.
1754  */
1755 static void
1756 filecaps_validate(const struct filecaps *fcaps, const char *func)
1757 {
1758 
1759 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1760 	    ("%s: invalid rights", func));
1761 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1762 	    ("%s: invalid fcntls", func));
1763 	KASSERT(fcaps->fc_fcntls == 0 ||
1764 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1765 	    ("%s: fcntls without CAP_FCNTL", func));
1766 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1767 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1768 	    ("%s: invalid ioctls", func));
1769 	KASSERT(fcaps->fc_nioctls == 0 ||
1770 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1771 	    ("%s: ioctls without CAP_IOCTL", func));
1772 }
1773 
1774 static void
1775 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1776 {
1777 	int nfd1;
1778 
1779 	FILEDESC_XLOCK_ASSERT(fdp);
1780 
1781 	nfd1 = fdp->fd_nfiles * 2;
1782 	if (nfd1 < nfd)
1783 		nfd1 = nfd;
1784 	fdgrowtable(fdp, nfd1);
1785 }
1786 
1787 /*
1788  * Grow the file table to accommodate (at least) nfd descriptors.
1789  */
1790 static void
1791 fdgrowtable(struct filedesc *fdp, int nfd)
1792 {
1793 	struct filedesc0 *fdp0;
1794 	struct freetable *ft;
1795 	struct fdescenttbl *ntable;
1796 	struct fdescenttbl *otable;
1797 	int nnfiles, onfiles;
1798 	NDSLOTTYPE *nmap, *omap;
1799 
1800 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1801 
1802 	/* save old values */
1803 	onfiles = fdp->fd_nfiles;
1804 	otable = fdp->fd_files;
1805 	omap = fdp->fd_map;
1806 
1807 	/* compute the size of the new table */
1808 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1809 	if (nnfiles <= onfiles)
1810 		/* the table is already large enough */
1811 		return;
1812 
1813 	/*
1814 	 * Allocate a new table.  We need enough space for the number of
1815 	 * entries, file entries themselves and the struct freetable we will use
1816 	 * when we decommission the table and place it on the freelist.
1817 	 * We place the struct freetable in the middle so we don't have
1818 	 * to worry about padding.
1819 	 */
1820 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1821 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1822 	    sizeof(struct freetable),
1823 	    M_FILEDESC, M_ZERO | M_WAITOK);
1824 	/* copy the old data */
1825 	ntable->fdt_nfiles = nnfiles;
1826 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1827 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1828 
1829 	/*
1830 	 * Allocate a new map only if the old is not large enough.  It will
1831 	 * grow at a slower rate than the table as it can map more
1832 	 * entries than the table can hold.
1833 	 */
1834 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1835 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1836 		    M_ZERO | M_WAITOK);
1837 		/* copy over the old data and update the pointer */
1838 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1839 		fdp->fd_map = nmap;
1840 	}
1841 
1842 	/*
1843 	 * Make sure that ntable is correctly initialized before we replace
1844 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1845 	 * data.
1846 	 */
1847 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1848 
1849 	/*
1850 	 * Free the old file table when not shared by other threads or processes.
1851 	 * The old file table is considered to be shared when either are true:
1852 	 * - The process has more than one thread.
1853 	 * - The file descriptor table has been shared via fdshare().
1854 	 *
1855 	 * When shared, the old file table will be placed on a freelist
1856 	 * which will be processed when the struct filedesc is released.
1857 	 *
1858 	 * Note that if onfiles == NDFILE, we're dealing with the original
1859 	 * static allocation contained within (struct filedesc0 *)fdp,
1860 	 * which must not be freed.
1861 	 */
1862 	if (onfiles > NDFILE) {
1863 		if (curproc->p_numthreads == 1 &&
1864 		    refcount_load(&fdp->fd_refcnt) == 1)
1865 			free(otable, M_FILEDESC);
1866 		else {
1867 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1868 			fdp0 = (struct filedesc0 *)fdp;
1869 			ft->ft_table = otable;
1870 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1871 		}
1872 	}
1873 	/*
1874 	 * The map does not have the same possibility of threads still
1875 	 * holding references to it.  So always free it as long as it
1876 	 * does not reference the original static allocation.
1877 	 */
1878 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1879 		free(omap, M_FILEDESC);
1880 }
1881 
1882 /*
1883  * Allocate a file descriptor for the process.
1884  */
1885 int
1886 fdalloc(struct thread *td, int minfd, int *result)
1887 {
1888 	struct proc *p = td->td_proc;
1889 	struct filedesc *fdp = p->p_fd;
1890 	int fd, maxfd, allocfd;
1891 #ifdef RACCT
1892 	int error;
1893 #endif
1894 
1895 	FILEDESC_XLOCK_ASSERT(fdp);
1896 
1897 	if (fdp->fd_freefile > minfd)
1898 		minfd = fdp->fd_freefile;
1899 
1900 	maxfd = getmaxfd(td);
1901 
1902 	/*
1903 	 * Search the bitmap for a free descriptor starting at minfd.
1904 	 * If none is found, grow the file table.
1905 	 */
1906 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1907 	if (__predict_false(fd >= maxfd))
1908 		return (EMFILE);
1909 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1910 		allocfd = min(fd * 2, maxfd);
1911 #ifdef RACCT
1912 		if (RACCT_ENABLED()) {
1913 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1914 			if (error != 0)
1915 				return (EMFILE);
1916 		}
1917 #endif
1918 		/*
1919 		 * fd is already equal to first free descriptor >= minfd, so
1920 		 * we only need to grow the table and we are done.
1921 		 */
1922 		fdgrowtable_exp(fdp, allocfd);
1923 	}
1924 
1925 	/*
1926 	 * Perform some sanity checks, then mark the file descriptor as
1927 	 * used and return it to the caller.
1928 	 */
1929 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1930 	    ("invalid descriptor %d", fd));
1931 	KASSERT(!fdisused(fdp, fd),
1932 	    ("fd_first_free() returned non-free descriptor"));
1933 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1934 	    ("file descriptor isn't free"));
1935 	fdused(fdp, fd);
1936 	*result = fd;
1937 	return (0);
1938 }
1939 
1940 /*
1941  * Allocate n file descriptors for the process.
1942  */
1943 int
1944 fdallocn(struct thread *td, int minfd, int *fds, int n)
1945 {
1946 	struct proc *p = td->td_proc;
1947 	struct filedesc *fdp = p->p_fd;
1948 	int i;
1949 
1950 	FILEDESC_XLOCK_ASSERT(fdp);
1951 
1952 	for (i = 0; i < n; i++)
1953 		if (fdalloc(td, 0, &fds[i]) != 0)
1954 			break;
1955 
1956 	if (i < n) {
1957 		for (i--; i >= 0; i--)
1958 			fdunused(fdp, fds[i]);
1959 		return (EMFILE);
1960 	}
1961 
1962 	return (0);
1963 }
1964 
1965 /*
1966  * Create a new open file structure and allocate a file descriptor for the
1967  * process that refers to it.  We add one reference to the file for the
1968  * descriptor table and one reference for resultfp. This is to prevent us
1969  * being preempted and the entry in the descriptor table closed after we
1970  * release the FILEDESC lock.
1971  */
1972 int
1973 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1974     struct filecaps *fcaps)
1975 {
1976 	struct file *fp;
1977 	int error, fd;
1978 
1979 	error = falloc_noinstall(td, &fp);
1980 	if (error)
1981 		return (error);		/* no reference held on error */
1982 
1983 	error = finstall(td, fp, &fd, flags, fcaps);
1984 	if (error) {
1985 		fdrop(fp, td);		/* one reference (fp only) */
1986 		return (error);
1987 	}
1988 
1989 	if (resultfp != NULL)
1990 		*resultfp = fp;		/* copy out result */
1991 	else
1992 		fdrop(fp, td);		/* release local reference */
1993 
1994 	if (resultfd != NULL)
1995 		*resultfd = fd;
1996 
1997 	return (0);
1998 }
1999 
2000 /*
2001  * Create a new open file structure without allocating a file descriptor.
2002  */
2003 int
2004 falloc_noinstall(struct thread *td, struct file **resultfp)
2005 {
2006 	struct file *fp;
2007 	int maxuserfiles = maxfiles - (maxfiles / 20);
2008 	int openfiles_new;
2009 	static struct timeval lastfail;
2010 	static int curfail;
2011 
2012 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2013 
2014 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2015 	if ((openfiles_new >= maxuserfiles &&
2016 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2017 	    openfiles_new >= maxfiles) {
2018 		atomic_subtract_int(&openfiles, 1);
2019 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2020 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2021 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2022 		}
2023 		return (ENFILE);
2024 	}
2025 	fp = uma_zalloc(file_zone, M_WAITOK);
2026 	bzero(fp, sizeof(*fp));
2027 	refcount_init(&fp->f_count, 1);
2028 	fp->f_cred = crhold(td->td_ucred);
2029 	fp->f_ops = &badfileops;
2030 	*resultfp = fp;
2031 	return (0);
2032 }
2033 
2034 /*
2035  * Install a file in a file descriptor table.
2036  */
2037 void
2038 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2039     struct filecaps *fcaps)
2040 {
2041 	struct filedescent *fde;
2042 
2043 	MPASS(fp != NULL);
2044 	if (fcaps != NULL)
2045 		filecaps_validate(fcaps, __func__);
2046 	FILEDESC_XLOCK_ASSERT(fdp);
2047 
2048 	fde = &fdp->fd_ofiles[fd];
2049 #ifdef CAPABILITIES
2050 	seqc_write_begin(&fde->fde_seqc);
2051 #endif
2052 	fde->fde_file = fp;
2053 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2054 	if (fcaps != NULL)
2055 		filecaps_move(fcaps, &fde->fde_caps);
2056 	else
2057 		filecaps_fill(&fde->fde_caps);
2058 #ifdef CAPABILITIES
2059 	seqc_write_end(&fde->fde_seqc);
2060 #endif
2061 }
2062 
2063 int
2064 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2065     struct filecaps *fcaps)
2066 {
2067 	struct filedesc *fdp = td->td_proc->p_fd;
2068 	int error;
2069 
2070 	MPASS(fd != NULL);
2071 
2072 	if (!fhold(fp))
2073 		return (EBADF);
2074 	FILEDESC_XLOCK(fdp);
2075 	error = fdalloc(td, 0, fd);
2076 	if (__predict_false(error != 0)) {
2077 		FILEDESC_XUNLOCK(fdp);
2078 		fdrop(fp, td);
2079 		return (error);
2080 	}
2081 	_finstall(fdp, fp, *fd, flags, fcaps);
2082 	FILEDESC_XUNLOCK(fdp);
2083 	return (0);
2084 }
2085 
2086 /*
2087  * Build a new filedesc structure from another.
2088  *
2089  * If fdp is not NULL, return with it shared locked.
2090  */
2091 struct filedesc *
2092 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2093 {
2094 	struct filedesc0 *newfdp0;
2095 	struct filedesc *newfdp;
2096 
2097 	if (prepfiles)
2098 		MPASS(lastfile != NULL);
2099 	else
2100 		MPASS(lastfile == NULL);
2101 
2102 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2103 	newfdp = &newfdp0->fd_fd;
2104 
2105 	/* Create the file descriptor table. */
2106 	FILEDESC_LOCK_INIT(newfdp);
2107 	refcount_init(&newfdp->fd_refcnt, 1);
2108 	refcount_init(&newfdp->fd_holdcnt, 1);
2109 	newfdp->fd_map = newfdp0->fd_dmap;
2110 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2111 	newfdp->fd_files->fdt_nfiles = NDFILE;
2112 
2113 	if (fdp == NULL)
2114 		return (newfdp);
2115 
2116 	FILEDESC_SLOCK(fdp);
2117 	if (!prepfiles) {
2118 		FILEDESC_SUNLOCK(fdp);
2119 		return (newfdp);
2120 	}
2121 
2122 	for (;;) {
2123 		*lastfile = fdlastfile(fdp);
2124 		if (*lastfile < newfdp->fd_nfiles)
2125 			break;
2126 		FILEDESC_SUNLOCK(fdp);
2127 		fdgrowtable(newfdp, *lastfile + 1);
2128 		FILEDESC_SLOCK(fdp);
2129 	}
2130 
2131 	return (newfdp);
2132 }
2133 
2134 /*
2135  * Build a pwddesc structure from another.
2136  * Copy the current, root, and jail root vnode references.
2137  *
2138  * If pdp is not NULL, return with it shared locked.
2139  */
2140 struct pwddesc *
2141 pdinit(struct pwddesc *pdp, bool keeplock)
2142 {
2143 	struct pwddesc *newpdp;
2144 	struct pwd *newpwd;
2145 
2146 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2147 
2148 	PWDDESC_LOCK_INIT(newpdp);
2149 	refcount_init(&newpdp->pd_refcount, 1);
2150 	newpdp->pd_cmask = CMASK;
2151 
2152 	if (pdp == NULL) {
2153 		newpwd = pwd_alloc();
2154 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2155 		return (newpdp);
2156 	}
2157 
2158 	PWDDESC_XLOCK(pdp);
2159 	newpwd = pwd_hold_pwddesc(pdp);
2160 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2161 	if (!keeplock)
2162 		PWDDESC_XUNLOCK(pdp);
2163 	return (newpdp);
2164 }
2165 
2166 static struct filedesc *
2167 fdhold(struct proc *p)
2168 {
2169 	struct filedesc *fdp;
2170 
2171 	PROC_LOCK_ASSERT(p, MA_OWNED);
2172 	fdp = p->p_fd;
2173 	if (fdp != NULL)
2174 		refcount_acquire(&fdp->fd_holdcnt);
2175 	return (fdp);
2176 }
2177 
2178 static struct pwddesc *
2179 pdhold(struct proc *p)
2180 {
2181 	struct pwddesc *pdp;
2182 
2183 	PROC_LOCK_ASSERT(p, MA_OWNED);
2184 	pdp = p->p_pd;
2185 	if (pdp != NULL)
2186 		refcount_acquire(&pdp->pd_refcount);
2187 	return (pdp);
2188 }
2189 
2190 static void
2191 fddrop(struct filedesc *fdp)
2192 {
2193 
2194 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2195 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2196 			return;
2197 	}
2198 
2199 	FILEDESC_LOCK_DESTROY(fdp);
2200 	uma_zfree(filedesc0_zone, fdp);
2201 }
2202 
2203 static void
2204 pddrop(struct pwddesc *pdp)
2205 {
2206 	struct pwd *pwd;
2207 
2208 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2209 		return;
2210 
2211 	PWDDESC_XLOCK(pdp);
2212 	if (refcount_release(&pdp->pd_refcount) == 0) {
2213 		PWDDESC_XUNLOCK(pdp);
2214 		return;
2215 	}
2216 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2217 	pwd_set(pdp, NULL);
2218 	PWDDESC_XUNLOCK(pdp);
2219 	pwd_drop(pwd);
2220 
2221 	PWDDESC_LOCK_DESTROY(pdp);
2222 	free(pdp, M_PWDDESC);
2223 }
2224 
2225 /*
2226  * Share a filedesc structure.
2227  */
2228 struct filedesc *
2229 fdshare(struct filedesc *fdp)
2230 {
2231 
2232 	refcount_acquire(&fdp->fd_refcnt);
2233 	return (fdp);
2234 }
2235 
2236 /*
2237  * Share a pwddesc structure.
2238  */
2239 struct pwddesc *
2240 pdshare(struct pwddesc *pdp)
2241 {
2242 	refcount_acquire(&pdp->pd_refcount);
2243 	return (pdp);
2244 }
2245 
2246 /*
2247  * Unshare a filedesc structure, if necessary by making a copy
2248  */
2249 void
2250 fdunshare(struct thread *td)
2251 {
2252 	struct filedesc *tmp;
2253 	struct proc *p = td->td_proc;
2254 
2255 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2256 		return;
2257 
2258 	tmp = fdcopy(p->p_fd);
2259 	fdescfree(td);
2260 	p->p_fd = tmp;
2261 }
2262 
2263 /*
2264  * Unshare a pwddesc structure.
2265  */
2266 void
2267 pdunshare(struct thread *td)
2268 {
2269 	struct pwddesc *pdp;
2270 	struct proc *p;
2271 
2272 	p = td->td_proc;
2273 	/* Not shared. */
2274 	if (p->p_pd->pd_refcount == 1)
2275 		return;
2276 
2277 	pdp = pdcopy(p->p_pd);
2278 	pdescfree(td);
2279 	p->p_pd = pdp;
2280 }
2281 
2282 void
2283 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2284 {
2285 
2286 	fdescfree(td);
2287 	td->td_proc->p_fd = fdp;
2288 }
2289 
2290 /*
2291  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2292  * this is to ease callers, not catch errors.
2293  */
2294 struct filedesc *
2295 fdcopy(struct filedesc *fdp)
2296 {
2297 	struct filedesc *newfdp;
2298 	struct filedescent *nfde, *ofde;
2299 	int i, lastfile;
2300 
2301 	MPASS(fdp != NULL);
2302 
2303 	newfdp = fdinit(fdp, true, &lastfile);
2304 	/* copy all passable descriptors (i.e. not kqueue) */
2305 	newfdp->fd_freefile = -1;
2306 	for (i = 0; i <= lastfile; ++i) {
2307 		ofde = &fdp->fd_ofiles[i];
2308 		if (ofde->fde_file == NULL ||
2309 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2310 		    !fhold(ofde->fde_file)) {
2311 			if (newfdp->fd_freefile == -1)
2312 				newfdp->fd_freefile = i;
2313 			continue;
2314 		}
2315 		nfde = &newfdp->fd_ofiles[i];
2316 		*nfde = *ofde;
2317 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2318 		fdused_init(newfdp, i);
2319 	}
2320 	if (newfdp->fd_freefile == -1)
2321 		newfdp->fd_freefile = i;
2322 	FILEDESC_SUNLOCK(fdp);
2323 	return (newfdp);
2324 }
2325 
2326 /*
2327  * Copy a pwddesc structure.
2328  */
2329 struct pwddesc *
2330 pdcopy(struct pwddesc *pdp)
2331 {
2332 	struct pwddesc *newpdp;
2333 
2334 	MPASS(pdp != NULL);
2335 
2336 	newpdp = pdinit(pdp, true);
2337 	newpdp->pd_cmask = pdp->pd_cmask;
2338 	PWDDESC_XUNLOCK(pdp);
2339 	return (newpdp);
2340 }
2341 
2342 /*
2343  * Copies a filedesc structure, while remapping all file descriptors
2344  * stored inside using a translation table.
2345  *
2346  * File descriptors are copied over to the new file descriptor table,
2347  * regardless of whether the close-on-exec flag is set.
2348  */
2349 int
2350 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2351     struct filedesc **ret)
2352 {
2353 	struct filedesc *newfdp;
2354 	struct filedescent *nfde, *ofde;
2355 	int error, i, lastfile;
2356 
2357 	MPASS(fdp != NULL);
2358 
2359 	newfdp = fdinit(fdp, true, &lastfile);
2360 	if (nfds > lastfile + 1) {
2361 		/* New table cannot be larger than the old one. */
2362 		error = E2BIG;
2363 		goto bad;
2364 	}
2365 	/* Copy all passable descriptors (i.e. not kqueue). */
2366 	newfdp->fd_freefile = nfds;
2367 	for (i = 0; i < nfds; ++i) {
2368 		if (fds[i] < 0 || fds[i] > lastfile) {
2369 			/* File descriptor out of bounds. */
2370 			error = EBADF;
2371 			goto bad;
2372 		}
2373 		ofde = &fdp->fd_ofiles[fds[i]];
2374 		if (ofde->fde_file == NULL) {
2375 			/* Unused file descriptor. */
2376 			error = EBADF;
2377 			goto bad;
2378 		}
2379 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2380 			/* File descriptor cannot be passed. */
2381 			error = EINVAL;
2382 			goto bad;
2383 		}
2384 		if (!fhold(ofde->fde_file)) {
2385 			error = EBADF;
2386 			goto bad;
2387 		}
2388 		nfde = &newfdp->fd_ofiles[i];
2389 		*nfde = *ofde;
2390 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2391 		fdused_init(newfdp, i);
2392 	}
2393 	FILEDESC_SUNLOCK(fdp);
2394 	*ret = newfdp;
2395 	return (0);
2396 bad:
2397 	FILEDESC_SUNLOCK(fdp);
2398 	fdescfree_remapped(newfdp);
2399 	return (error);
2400 }
2401 
2402 /*
2403  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2404  * one of processes using it exits) and the table used to be shared.
2405  */
2406 static void
2407 fdclearlocks(struct thread *td)
2408 {
2409 	struct filedesc *fdp;
2410 	struct filedesc_to_leader *fdtol;
2411 	struct flock lf;
2412 	struct file *fp;
2413 	struct proc *p;
2414 	struct vnode *vp;
2415 	int i, lastfile;
2416 
2417 	p = td->td_proc;
2418 	fdp = p->p_fd;
2419 	fdtol = p->p_fdtol;
2420 	MPASS(fdtol != NULL);
2421 
2422 	FILEDESC_XLOCK(fdp);
2423 	KASSERT(fdtol->fdl_refcount > 0,
2424 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2425 	    fdtol->fdl_refcount));
2426 	if (fdtol->fdl_refcount == 1 &&
2427 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2428 		lastfile = fdlastfile(fdp);
2429 		for (i = 0; i <= lastfile; i++) {
2430 			fp = fdp->fd_ofiles[i].fde_file;
2431 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2432 			    !fhold(fp))
2433 				continue;
2434 			FILEDESC_XUNLOCK(fdp);
2435 			lf.l_whence = SEEK_SET;
2436 			lf.l_start = 0;
2437 			lf.l_len = 0;
2438 			lf.l_type = F_UNLCK;
2439 			vp = fp->f_vnode;
2440 			(void) VOP_ADVLOCK(vp,
2441 			    (caddr_t)p->p_leader, F_UNLCK,
2442 			    &lf, F_POSIX);
2443 			FILEDESC_XLOCK(fdp);
2444 			fdrop(fp, td);
2445 		}
2446 	}
2447 retry:
2448 	if (fdtol->fdl_refcount == 1) {
2449 		if (fdp->fd_holdleaderscount > 0 &&
2450 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2451 			/*
2452 			 * close() or kern_dup() has cleared a reference
2453 			 * in a shared file descriptor table.
2454 			 */
2455 			fdp->fd_holdleaderswakeup = 1;
2456 			sx_sleep(&fdp->fd_holdleaderscount,
2457 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2458 			goto retry;
2459 		}
2460 		if (fdtol->fdl_holdcount > 0) {
2461 			/*
2462 			 * Ensure that fdtol->fdl_leader remains
2463 			 * valid in closef().
2464 			 */
2465 			fdtol->fdl_wakeup = 1;
2466 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2467 			    "fdlhold", 0);
2468 			goto retry;
2469 		}
2470 	}
2471 	fdtol->fdl_refcount--;
2472 	if (fdtol->fdl_refcount == 0 &&
2473 	    fdtol->fdl_holdcount == 0) {
2474 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2475 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2476 	} else
2477 		fdtol = NULL;
2478 	p->p_fdtol = NULL;
2479 	FILEDESC_XUNLOCK(fdp);
2480 	if (fdtol != NULL)
2481 		free(fdtol, M_FILEDESC_TO_LEADER);
2482 }
2483 
2484 /*
2485  * Release a filedesc structure.
2486  */
2487 static void
2488 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2489 {
2490 	struct filedesc0 *fdp0;
2491 	struct freetable *ft, *tft;
2492 	struct filedescent *fde;
2493 	struct file *fp;
2494 	int i, lastfile;
2495 
2496 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2497 	    ("%s: fd table %p carries references", __func__, fdp));
2498 
2499 	/*
2500 	 * Serialize with threads iterating over the table, if any.
2501 	 */
2502 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2503 		FILEDESC_XLOCK(fdp);
2504 		FILEDESC_XUNLOCK(fdp);
2505 	}
2506 
2507 	lastfile = fdlastfile_single(fdp);
2508 	for (i = 0; i <= lastfile; i++) {
2509 		fde = &fdp->fd_ofiles[i];
2510 		fp = fde->fde_file;
2511 		if (fp != NULL) {
2512 			fdefree_last(fde);
2513 			if (needclose)
2514 				(void) closef(fp, td);
2515 			else
2516 				fdrop(fp, td);
2517 		}
2518 	}
2519 
2520 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2521 		free(fdp->fd_map, M_FILEDESC);
2522 	if (fdp->fd_nfiles > NDFILE)
2523 		free(fdp->fd_files, M_FILEDESC);
2524 
2525 	fdp0 = (struct filedesc0 *)fdp;
2526 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2527 		free(ft->ft_table, M_FILEDESC);
2528 
2529 	fddrop(fdp);
2530 }
2531 
2532 void
2533 fdescfree(struct thread *td)
2534 {
2535 	struct proc *p;
2536 	struct filedesc *fdp;
2537 
2538 	p = td->td_proc;
2539 	fdp = p->p_fd;
2540 	MPASS(fdp != NULL);
2541 
2542 #ifdef RACCT
2543 	if (RACCT_ENABLED())
2544 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2545 #endif
2546 
2547 	if (p->p_fdtol != NULL)
2548 		fdclearlocks(td);
2549 
2550 	PROC_LOCK(p);
2551 	p->p_fd = NULL;
2552 	PROC_UNLOCK(p);
2553 
2554 	if (refcount_release(&fdp->fd_refcnt) == 0)
2555 		return;
2556 
2557 	fdescfree_fds(td, fdp, 1);
2558 }
2559 
2560 void
2561 pdescfree(struct thread *td)
2562 {
2563 	struct proc *p;
2564 	struct pwddesc *pdp;
2565 
2566 	p = td->td_proc;
2567 	pdp = p->p_pd;
2568 	MPASS(pdp != NULL);
2569 
2570 	PROC_LOCK(p);
2571 	p->p_pd = NULL;
2572 	PROC_UNLOCK(p);
2573 
2574 	pddrop(pdp);
2575 }
2576 
2577 void
2578 fdescfree_remapped(struct filedesc *fdp)
2579 {
2580 #ifdef INVARIANTS
2581 	/* fdescfree_fds() asserts that fd_refcnt == 0. */
2582 	if (!refcount_release(&fdp->fd_refcnt))
2583 		panic("%s: fd table %p has extra references", __func__, fdp);
2584 #endif
2585 	fdescfree_fds(curthread, fdp, 0);
2586 }
2587 
2588 /*
2589  * For setugid programs, we don't want to people to use that setugidness
2590  * to generate error messages which write to a file which otherwise would
2591  * otherwise be off-limits to the process.  We check for filesystems where
2592  * the vnode can change out from under us after execve (like [lin]procfs).
2593  *
2594  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2595  * sufficient.  We also don't check for setugidness since we know we are.
2596  */
2597 static bool
2598 is_unsafe(struct file *fp)
2599 {
2600 	struct vnode *vp;
2601 
2602 	if (fp->f_type != DTYPE_VNODE)
2603 		return (false);
2604 
2605 	vp = fp->f_vnode;
2606 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2607 }
2608 
2609 /*
2610  * Make this setguid thing safe, if at all possible.
2611  */
2612 void
2613 fdsetugidsafety(struct thread *td)
2614 {
2615 	struct filedesc *fdp;
2616 	struct file *fp;
2617 	int i;
2618 
2619 	fdp = td->td_proc->p_fd;
2620 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2621 	    ("the fdtable should not be shared"));
2622 	MPASS(fdp->fd_nfiles >= 3);
2623 	for (i = 0; i <= 2; i++) {
2624 		fp = fdp->fd_ofiles[i].fde_file;
2625 		if (fp != NULL && is_unsafe(fp)) {
2626 			FILEDESC_XLOCK(fdp);
2627 			knote_fdclose(td, i);
2628 			/*
2629 			 * NULL-out descriptor prior to close to avoid
2630 			 * a race while close blocks.
2631 			 */
2632 			fdfree(fdp, i);
2633 			FILEDESC_XUNLOCK(fdp);
2634 			(void) closef(fp, td);
2635 		}
2636 	}
2637 }
2638 
2639 /*
2640  * If a specific file object occupies a specific file descriptor, close the
2641  * file descriptor entry and drop a reference on the file object.  This is a
2642  * convenience function to handle a subsequent error in a function that calls
2643  * falloc() that handles the race that another thread might have closed the
2644  * file descriptor out from under the thread creating the file object.
2645  */
2646 void
2647 fdclose(struct thread *td, struct file *fp, int idx)
2648 {
2649 	struct filedesc *fdp = td->td_proc->p_fd;
2650 
2651 	FILEDESC_XLOCK(fdp);
2652 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2653 		fdfree(fdp, idx);
2654 		FILEDESC_XUNLOCK(fdp);
2655 		fdrop(fp, td);
2656 	} else
2657 		FILEDESC_XUNLOCK(fdp);
2658 }
2659 
2660 /*
2661  * Close any files on exec?
2662  */
2663 void
2664 fdcloseexec(struct thread *td)
2665 {
2666 	struct filedesc *fdp;
2667 	struct filedescent *fde;
2668 	struct file *fp;
2669 	int i, lastfile;
2670 
2671 	fdp = td->td_proc->p_fd;
2672 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2673 	    ("the fdtable should not be shared"));
2674 	lastfile = fdlastfile_single(fdp);
2675 	for (i = 0; i <= lastfile; i++) {
2676 		fde = &fdp->fd_ofiles[i];
2677 		fp = fde->fde_file;
2678 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2679 		    (fde->fde_flags & UF_EXCLOSE))) {
2680 			FILEDESC_XLOCK(fdp);
2681 			fdfree(fdp, i);
2682 			(void) closefp(fdp, i, fp, td, false, false);
2683 			FILEDESC_UNLOCK_ASSERT(fdp);
2684 		}
2685 	}
2686 }
2687 
2688 /*
2689  * It is unsafe for set[ug]id processes to be started with file
2690  * descriptors 0..2 closed, as these descriptors are given implicit
2691  * significance in the Standard C library.  fdcheckstd() will create a
2692  * descriptor referencing /dev/null for each of stdin, stdout, and
2693  * stderr that is not already open.
2694  */
2695 int
2696 fdcheckstd(struct thread *td)
2697 {
2698 	struct filedesc *fdp;
2699 	register_t save;
2700 	int i, error, devnull;
2701 
2702 	fdp = td->td_proc->p_fd;
2703 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2704 	    ("the fdtable should not be shared"));
2705 	MPASS(fdp->fd_nfiles >= 3);
2706 	devnull = -1;
2707 	for (i = 0; i <= 2; i++) {
2708 		if (fdp->fd_ofiles[i].fde_file != NULL)
2709 			continue;
2710 
2711 		save = td->td_retval[0];
2712 		if (devnull != -1) {
2713 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2714 		} else {
2715 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2716 			    UIO_SYSSPACE, O_RDWR, 0);
2717 			if (error == 0) {
2718 				devnull = td->td_retval[0];
2719 				KASSERT(devnull == i, ("we didn't get our fd"));
2720 			}
2721 		}
2722 		td->td_retval[0] = save;
2723 		if (error != 0)
2724 			return (error);
2725 	}
2726 	return (0);
2727 }
2728 
2729 /*
2730  * Internal form of close.  Decrement reference count on file structure.
2731  * Note: td may be NULL when closing a file that was being passed in a
2732  * message.
2733  */
2734 int
2735 closef(struct file *fp, struct thread *td)
2736 {
2737 	struct vnode *vp;
2738 	struct flock lf;
2739 	struct filedesc_to_leader *fdtol;
2740 	struct filedesc *fdp;
2741 
2742 	/*
2743 	 * POSIX record locking dictates that any close releases ALL
2744 	 * locks owned by this process.  This is handled by setting
2745 	 * a flag in the unlock to free ONLY locks obeying POSIX
2746 	 * semantics, and not to free BSD-style file locks.
2747 	 * If the descriptor was in a message, POSIX-style locks
2748 	 * aren't passed with the descriptor, and the thread pointer
2749 	 * will be NULL.  Callers should be careful only to pass a
2750 	 * NULL thread pointer when there really is no owning
2751 	 * context that might have locks, or the locks will be
2752 	 * leaked.
2753 	 */
2754 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2755 		vp = fp->f_vnode;
2756 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2757 			lf.l_whence = SEEK_SET;
2758 			lf.l_start = 0;
2759 			lf.l_len = 0;
2760 			lf.l_type = F_UNLCK;
2761 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2762 			    F_UNLCK, &lf, F_POSIX);
2763 		}
2764 		fdtol = td->td_proc->p_fdtol;
2765 		if (fdtol != NULL) {
2766 			/*
2767 			 * Handle special case where file descriptor table is
2768 			 * shared between multiple process leaders.
2769 			 */
2770 			fdp = td->td_proc->p_fd;
2771 			FILEDESC_XLOCK(fdp);
2772 			for (fdtol = fdtol->fdl_next;
2773 			    fdtol != td->td_proc->p_fdtol;
2774 			    fdtol = fdtol->fdl_next) {
2775 				if ((fdtol->fdl_leader->p_flag &
2776 				    P_ADVLOCK) == 0)
2777 					continue;
2778 				fdtol->fdl_holdcount++;
2779 				FILEDESC_XUNLOCK(fdp);
2780 				lf.l_whence = SEEK_SET;
2781 				lf.l_start = 0;
2782 				lf.l_len = 0;
2783 				lf.l_type = F_UNLCK;
2784 				vp = fp->f_vnode;
2785 				(void) VOP_ADVLOCK(vp,
2786 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2787 				    F_POSIX);
2788 				FILEDESC_XLOCK(fdp);
2789 				fdtol->fdl_holdcount--;
2790 				if (fdtol->fdl_holdcount == 0 &&
2791 				    fdtol->fdl_wakeup != 0) {
2792 					fdtol->fdl_wakeup = 0;
2793 					wakeup(fdtol);
2794 				}
2795 			}
2796 			FILEDESC_XUNLOCK(fdp);
2797 		}
2798 	}
2799 	return (fdrop_close(fp, td));
2800 }
2801 
2802 /*
2803  * Initialize the file pointer with the specified properties.
2804  *
2805  * The ops are set with release semantics to be certain that the flags, type,
2806  * and data are visible when ops is.  This is to prevent ops methods from being
2807  * called with bad data.
2808  */
2809 void
2810 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2811 {
2812 	fp->f_data = data;
2813 	fp->f_flag = flag;
2814 	fp->f_type = type;
2815 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2816 }
2817 
2818 void
2819 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2820 {
2821 	fp->f_seqcount[UIO_READ] = 1;
2822 	fp->f_seqcount[UIO_WRITE] = 1;
2823 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2824 	    data, ops);
2825 }
2826 
2827 int
2828 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2829     struct file **fpp, struct filecaps *havecapsp)
2830 {
2831 	struct filedescent *fde;
2832 	int error;
2833 
2834 	FILEDESC_LOCK_ASSERT(fdp);
2835 
2836 	fde = fdeget_locked(fdp, fd);
2837 	if (fde == NULL) {
2838 		error = EBADF;
2839 		goto out;
2840 	}
2841 
2842 #ifdef CAPABILITIES
2843 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2844 	if (error != 0)
2845 		goto out;
2846 #endif
2847 
2848 	if (havecapsp != NULL)
2849 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2850 
2851 	*fpp = fde->fde_file;
2852 
2853 	error = 0;
2854 out:
2855 	return (error);
2856 }
2857 
2858 int
2859 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2860     struct file **fpp, struct filecaps *havecapsp)
2861 {
2862 	struct filedesc *fdp = td->td_proc->p_fd;
2863 	int error;
2864 #ifndef CAPABILITIES
2865 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2866 	if (havecapsp != NULL && error == 0)
2867 		filecaps_fill(havecapsp);
2868 #else
2869 	struct file *fp;
2870 	seqc_t seq;
2871 
2872 	*fpp = NULL;
2873 	for (;;) {
2874 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2875 		if (error != 0)
2876 			return (error);
2877 
2878 		if (havecapsp != NULL) {
2879 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2880 			    havecapsp, false)) {
2881 				fdrop(fp, td);
2882 				goto get_locked;
2883 			}
2884 		}
2885 
2886 		if (!fd_modified(fdp, fd, seq))
2887 			break;
2888 		fdrop(fp, td);
2889 	}
2890 
2891 	*fpp = fp;
2892 	return (0);
2893 
2894 get_locked:
2895 	FILEDESC_SLOCK(fdp);
2896 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2897 	if (error == 0 && !fhold(*fpp))
2898 		error = EBADF;
2899 	FILEDESC_SUNLOCK(fdp);
2900 #endif
2901 	return (error);
2902 }
2903 
2904 #ifdef CAPABILITIES
2905 int
2906 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2907 {
2908 	const struct filedescent *fde;
2909 	const struct fdescenttbl *fdt;
2910 	struct filedesc *fdp;
2911 	struct file *fp;
2912 	struct vnode *vp;
2913 	const cap_rights_t *haverights;
2914 	cap_rights_t rights;
2915 	seqc_t seq;
2916 
2917 	VFS_SMR_ASSERT_ENTERED();
2918 
2919 	rights = *ndp->ni_rightsneeded;
2920 	cap_rights_set_one(&rights, CAP_LOOKUP);
2921 
2922 	fdp = curproc->p_fd;
2923 	fdt = fdp->fd_files;
2924 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2925 		return (EBADF);
2926 	seq = seqc_read_any(fd_seqc(fdt, fd));
2927 	if (__predict_false(seqc_in_modify(seq)))
2928 		return (EAGAIN);
2929 	fde = &fdt->fdt_ofiles[fd];
2930 	haverights = cap_rights_fde_inline(fde);
2931 	fp = fde->fde_file;
2932 	if (__predict_false(fp == NULL))
2933 		return (EAGAIN);
2934 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2935 		return (EAGAIN);
2936 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2937 	vp = fp->f_vnode;
2938 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2939 		return (EAGAIN);
2940 	}
2941 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2942 		return (EAGAIN);
2943 	}
2944 	/*
2945 	 * Use an acquire barrier to force re-reading of fdt so it is
2946 	 * refreshed for verification.
2947 	 */
2948 	atomic_thread_fence_acq();
2949 	fdt = fdp->fd_files;
2950 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2951 		return (EAGAIN);
2952 	/*
2953 	 * If file descriptor doesn't have all rights,
2954 	 * all lookups relative to it must also be
2955 	 * strictly relative.
2956 	 *
2957 	 * Not yet supported by fast path.
2958 	 */
2959 	CAP_ALL(&rights);
2960 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2961 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2962 	    ndp->ni_filecaps.fc_nioctls != -1) {
2963 #ifdef notyet
2964 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2965 #else
2966 		return (EAGAIN);
2967 #endif
2968 	}
2969 	*vpp = vp;
2970 	return (0);
2971 }
2972 #else
2973 int
2974 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2975 {
2976 	const struct fdescenttbl *fdt;
2977 	struct filedesc *fdp;
2978 	struct file *fp;
2979 	struct vnode *vp;
2980 
2981 	VFS_SMR_ASSERT_ENTERED();
2982 
2983 	fdp = curproc->p_fd;
2984 	fdt = fdp->fd_files;
2985 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2986 		return (EBADF);
2987 	fp = fdt->fdt_ofiles[fd].fde_file;
2988 	if (__predict_false(fp == NULL))
2989 		return (EAGAIN);
2990 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2991 	vp = fp->f_vnode;
2992 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2993 		return (EAGAIN);
2994 	}
2995 	/*
2996 	 * Use an acquire barrier to force re-reading of fdt so it is
2997 	 * refreshed for verification.
2998 	 */
2999 	atomic_thread_fence_acq();
3000 	fdt = fdp->fd_files;
3001 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3002 		return (EAGAIN);
3003 	filecaps_fill(&ndp->ni_filecaps);
3004 	*vpp = vp;
3005 	return (0);
3006 }
3007 #endif
3008 
3009 int
3010 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3011     struct file **fpp, seqc_t *seqp)
3012 {
3013 #ifdef CAPABILITIES
3014 	const struct filedescent *fde;
3015 #endif
3016 	const struct fdescenttbl *fdt;
3017 	struct file *fp;
3018 #ifdef CAPABILITIES
3019 	seqc_t seq;
3020 	cap_rights_t haverights;
3021 	int error;
3022 #endif
3023 
3024 	fdt = fdp->fd_files;
3025 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3026 		return (EBADF);
3027 	/*
3028 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
3029 	 * never raising a refcount above 0.  To accomplish this we have
3030 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3031 	 * must be re-verified once we acquire a reference to be certain
3032 	 * that the identity is still correct and we did not lose a race
3033 	 * due to preemption.
3034 	 */
3035 	for (;;) {
3036 #ifdef CAPABILITIES
3037 		seq = seqc_read(fd_seqc(fdt, fd));
3038 		fde = &fdt->fdt_ofiles[fd];
3039 		haverights = *cap_rights_fde_inline(fde);
3040 		fp = fde->fde_file;
3041 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3042 			continue;
3043 #else
3044 		fp = fdt->fdt_ofiles[fd].fde_file;
3045 #endif
3046 		if (fp == NULL)
3047 			return (EBADF);
3048 #ifdef CAPABILITIES
3049 		error = cap_check_inline(&haverights, needrightsp);
3050 		if (error != 0)
3051 			return (error);
3052 #endif
3053 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3054 			/*
3055 			 * Force a reload. Other thread could reallocate the
3056 			 * table before this fd was closed, so it is possible
3057 			 * that there is a stale fp pointer in cached version.
3058 			 */
3059 			fdt = atomic_load_ptr(&fdp->fd_files);
3060 			continue;
3061 		}
3062 		/*
3063 		 * Use an acquire barrier to force re-reading of fdt so it is
3064 		 * refreshed for verification.
3065 		 */
3066 		atomic_thread_fence_acq();
3067 		fdt = fdp->fd_files;
3068 #ifdef	CAPABILITIES
3069 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3070 #else
3071 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3072 #endif
3073 			break;
3074 		fdrop(fp, curthread);
3075 	}
3076 	*fpp = fp;
3077 	if (seqp != NULL) {
3078 #ifdef CAPABILITIES
3079 		*seqp = seq;
3080 #endif
3081 	}
3082 	return (0);
3083 }
3084 
3085 /*
3086  * See the comments in fget_unlocked_seq for an explanation of how this works.
3087  *
3088  * This is a simplified variant which bails out to the aforementioned routine
3089  * if anything goes wrong. In practice this only happens when userspace is
3090  * racing with itself.
3091  */
3092 int
3093 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3094     struct file **fpp)
3095 {
3096 #ifdef CAPABILITIES
3097 	const struct filedescent *fde;
3098 #endif
3099 	const struct fdescenttbl *fdt;
3100 	struct file *fp;
3101 #ifdef CAPABILITIES
3102 	seqc_t seq;
3103 	const cap_rights_t *haverights;
3104 #endif
3105 
3106 	fdt = fdp->fd_files;
3107 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3108 		return (EBADF);
3109 #ifdef CAPABILITIES
3110 	seq = seqc_read_any(fd_seqc(fdt, fd));
3111 	if (__predict_false(seqc_in_modify(seq)))
3112 		goto out_fallback;
3113 	fde = &fdt->fdt_ofiles[fd];
3114 	haverights = cap_rights_fde_inline(fde);
3115 	fp = fde->fde_file;
3116 #else
3117 	fp = fdt->fdt_ofiles[fd].fde_file;
3118 #endif
3119 	if (__predict_false(fp == NULL))
3120 		goto out_fallback;
3121 #ifdef CAPABILITIES
3122 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3123 		goto out_fallback;
3124 #endif
3125 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3126 		goto out_fallback;
3127 
3128 	/*
3129 	 * Use an acquire barrier to force re-reading of fdt so it is
3130 	 * refreshed for verification.
3131 	 */
3132 	atomic_thread_fence_acq();
3133 	fdt = fdp->fd_files;
3134 #ifdef	CAPABILITIES
3135 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3136 #else
3137 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3138 #endif
3139 		goto out_fdrop;
3140 	*fpp = fp;
3141 	return (0);
3142 out_fdrop:
3143 	fdrop(fp, curthread);
3144 out_fallback:
3145 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3146 }
3147 
3148 /*
3149  * Extract the file pointer associated with the specified descriptor for the
3150  * current user process.
3151  *
3152  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3153  * returned.
3154  *
3155  * File's rights will be checked against the capability rights mask.
3156  *
3157  * If an error occurred the non-zero error is returned and *fpp is set to
3158  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3159  * responsible for fdrop().
3160  */
3161 static __inline int
3162 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3163     cap_rights_t *needrightsp)
3164 {
3165 	struct filedesc *fdp;
3166 	struct file *fp;
3167 	int error;
3168 
3169 	*fpp = NULL;
3170 	fdp = td->td_proc->p_fd;
3171 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3172 	if (__predict_false(error != 0))
3173 		return (error);
3174 	if (__predict_false(fp->f_ops == &badfileops)) {
3175 		fdrop(fp, td);
3176 		return (EBADF);
3177 	}
3178 
3179 	/*
3180 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3181 	 */
3182 	error = 0;
3183 	switch (flags) {
3184 	case FREAD:
3185 	case FWRITE:
3186 		if ((fp->f_flag & flags) == 0)
3187 			error = EBADF;
3188 		break;
3189 	case FEXEC:
3190 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3191 		    ((fp->f_flag & FWRITE) != 0))
3192 			error = EBADF;
3193 		break;
3194 	case 0:
3195 		break;
3196 	default:
3197 		KASSERT(0, ("wrong flags"));
3198 	}
3199 
3200 	if (error != 0) {
3201 		fdrop(fp, td);
3202 		return (error);
3203 	}
3204 
3205 	*fpp = fp;
3206 	return (0);
3207 }
3208 
3209 int
3210 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3211 {
3212 
3213 	return (_fget(td, fd, fpp, 0, rightsp));
3214 }
3215 
3216 int
3217 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3218     struct file **fpp)
3219 {
3220 	int error;
3221 #ifndef CAPABILITIES
3222 	error = _fget(td, fd, fpp, 0, rightsp);
3223 	if (maxprotp != NULL)
3224 		*maxprotp = VM_PROT_ALL;
3225 	return (error);
3226 #else
3227 	cap_rights_t fdrights;
3228 	struct filedesc *fdp;
3229 	struct file *fp;
3230 	seqc_t seq;
3231 
3232 	*fpp = NULL;
3233 	fdp = td->td_proc->p_fd;
3234 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3235 	for (;;) {
3236 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3237 		if (__predict_false(error != 0))
3238 			return (error);
3239 		if (__predict_false(fp->f_ops == &badfileops)) {
3240 			fdrop(fp, td);
3241 			return (EBADF);
3242 		}
3243 		if (maxprotp != NULL)
3244 			fdrights = *cap_rights(fdp, fd);
3245 		if (!fd_modified(fdp, fd, seq))
3246 			break;
3247 		fdrop(fp, td);
3248 	}
3249 
3250 	/*
3251 	 * If requested, convert capability rights to access flags.
3252 	 */
3253 	if (maxprotp != NULL)
3254 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3255 	*fpp = fp;
3256 	return (0);
3257 #endif
3258 }
3259 
3260 int
3261 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3262 {
3263 
3264 	return (_fget(td, fd, fpp, FREAD, rightsp));
3265 }
3266 
3267 int
3268 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3269 {
3270 
3271 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3272 }
3273 
3274 int
3275 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3276     struct file **fpp)
3277 {
3278 	struct filedesc *fdp = td->td_proc->p_fd;
3279 #ifndef CAPABILITIES
3280 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3281 #else
3282 	struct file *fp;
3283 	int error;
3284 	seqc_t seq;
3285 
3286 	*fpp = NULL;
3287 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3288 	for (;;) {
3289 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3290 		if (error != 0)
3291 			return (error);
3292 		error = cap_fcntl_check(fdp, fd, needfcntl);
3293 		if (!fd_modified(fdp, fd, seq))
3294 			break;
3295 		fdrop(fp, td);
3296 	}
3297 	if (error != 0) {
3298 		fdrop(fp, td);
3299 		return (error);
3300 	}
3301 	*fpp = fp;
3302 	return (0);
3303 #endif
3304 }
3305 
3306 /*
3307  * Like fget() but loads the underlying vnode, or returns an error if the
3308  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3309  * never have VM objects.  The returned vnode will be vref()'d.
3310  *
3311  * XXX: what about the unused flags ?
3312  */
3313 static __inline int
3314 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3315     struct vnode **vpp)
3316 {
3317 	struct file *fp;
3318 	int error;
3319 
3320 	*vpp = NULL;
3321 	error = _fget(td, fd, &fp, flags, needrightsp);
3322 	if (error != 0)
3323 		return (error);
3324 	if (fp->f_vnode == NULL) {
3325 		error = EINVAL;
3326 	} else {
3327 		*vpp = fp->f_vnode;
3328 		vrefact(*vpp);
3329 	}
3330 	fdrop(fp, td);
3331 
3332 	return (error);
3333 }
3334 
3335 int
3336 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3337 {
3338 
3339 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3340 }
3341 
3342 int
3343 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3344     struct filecaps *havecaps, struct vnode **vpp)
3345 {
3346 	struct filecaps caps;
3347 	struct file *fp;
3348 	int error;
3349 
3350 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3351 	if (error != 0)
3352 		return (error);
3353 	if (fp->f_ops == &badfileops) {
3354 		error = EBADF;
3355 		goto out;
3356 	}
3357 	if (fp->f_vnode == NULL) {
3358 		error = EINVAL;
3359 		goto out;
3360 	}
3361 
3362 	*havecaps = caps;
3363 	*vpp = fp->f_vnode;
3364 	vrefact(*vpp);
3365 	fdrop(fp, td);
3366 
3367 	return (0);
3368 out:
3369 	filecaps_free(&caps);
3370 	fdrop(fp, td);
3371 	return (error);
3372 }
3373 
3374 int
3375 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3376 {
3377 
3378 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3379 }
3380 
3381 int
3382 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3383 {
3384 
3385 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3386 }
3387 
3388 #ifdef notyet
3389 int
3390 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3391     struct vnode **vpp)
3392 {
3393 
3394 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3395 }
3396 #endif
3397 
3398 /*
3399  * Handle the last reference to a file being closed.
3400  *
3401  * Without the noinline attribute clang keeps inlining the func thorough this
3402  * file when fdrop is used.
3403  */
3404 int __noinline
3405 _fdrop(struct file *fp, struct thread *td)
3406 {
3407 	int error;
3408 #ifdef INVARIANTS
3409 	int count;
3410 
3411 	count = refcount_load(&fp->f_count);
3412 	if (count != 0)
3413 		panic("fdrop: fp %p count %d", fp, count);
3414 #endif
3415 	error = fo_close(fp, td);
3416 	atomic_subtract_int(&openfiles, 1);
3417 	crfree(fp->f_cred);
3418 	free(fp->f_advice, M_FADVISE);
3419 	uma_zfree(file_zone, fp);
3420 
3421 	return (error);
3422 }
3423 
3424 /*
3425  * Apply an advisory lock on a file descriptor.
3426  *
3427  * Just attempt to get a record lock of the requested type on the entire file
3428  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3429  */
3430 #ifndef _SYS_SYSPROTO_H_
3431 struct flock_args {
3432 	int	fd;
3433 	int	how;
3434 };
3435 #endif
3436 /* ARGSUSED */
3437 int
3438 sys_flock(struct thread *td, struct flock_args *uap)
3439 {
3440 	struct file *fp;
3441 	struct vnode *vp;
3442 	struct flock lf;
3443 	int error;
3444 
3445 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3446 	if (error != 0)
3447 		return (error);
3448 	if (fp->f_type != DTYPE_VNODE) {
3449 		fdrop(fp, td);
3450 		return (EOPNOTSUPP);
3451 	}
3452 
3453 	vp = fp->f_vnode;
3454 	lf.l_whence = SEEK_SET;
3455 	lf.l_start = 0;
3456 	lf.l_len = 0;
3457 	if (uap->how & LOCK_UN) {
3458 		lf.l_type = F_UNLCK;
3459 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3460 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3461 		goto done2;
3462 	}
3463 	if (uap->how & LOCK_EX)
3464 		lf.l_type = F_WRLCK;
3465 	else if (uap->how & LOCK_SH)
3466 		lf.l_type = F_RDLCK;
3467 	else {
3468 		error = EBADF;
3469 		goto done2;
3470 	}
3471 	atomic_set_int(&fp->f_flag, FHASLOCK);
3472 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3473 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3474 done2:
3475 	fdrop(fp, td);
3476 	return (error);
3477 }
3478 /*
3479  * Duplicate the specified descriptor to a free descriptor.
3480  */
3481 int
3482 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3483     int openerror, int *indxp)
3484 {
3485 	struct filedescent *newfde, *oldfde;
3486 	struct file *fp;
3487 	u_long *ioctls;
3488 	int error, indx;
3489 
3490 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3491 	    ("unexpected error %d in %s", openerror, __func__));
3492 
3493 	/*
3494 	 * If the to-be-dup'd fd number is greater than the allowed number
3495 	 * of file descriptors, or the fd to be dup'd has already been
3496 	 * closed, then reject.
3497 	 */
3498 	FILEDESC_XLOCK(fdp);
3499 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3500 		FILEDESC_XUNLOCK(fdp);
3501 		return (EBADF);
3502 	}
3503 
3504 	error = fdalloc(td, 0, &indx);
3505 	if (error != 0) {
3506 		FILEDESC_XUNLOCK(fdp);
3507 		return (error);
3508 	}
3509 
3510 	/*
3511 	 * There are two cases of interest here.
3512 	 *
3513 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3514 	 *
3515 	 * For ENXIO steal away the file structure from (dfd) and store it in
3516 	 * (indx).  (dfd) is effectively closed by this operation.
3517 	 */
3518 	switch (openerror) {
3519 	case ENODEV:
3520 		/*
3521 		 * Check that the mode the file is being opened for is a
3522 		 * subset of the mode of the existing descriptor.
3523 		 */
3524 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3525 			fdunused(fdp, indx);
3526 			FILEDESC_XUNLOCK(fdp);
3527 			return (EACCES);
3528 		}
3529 		if (!fhold(fp)) {
3530 			fdunused(fdp, indx);
3531 			FILEDESC_XUNLOCK(fdp);
3532 			return (EBADF);
3533 		}
3534 		newfde = &fdp->fd_ofiles[indx];
3535 		oldfde = &fdp->fd_ofiles[dfd];
3536 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3537 #ifdef CAPABILITIES
3538 		seqc_write_begin(&newfde->fde_seqc);
3539 #endif
3540 		memcpy(newfde, oldfde, fde_change_size);
3541 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3542 		    ioctls);
3543 #ifdef CAPABILITIES
3544 		seqc_write_end(&newfde->fde_seqc);
3545 #endif
3546 		break;
3547 	case ENXIO:
3548 		/*
3549 		 * Steal away the file pointer from dfd and stuff it into indx.
3550 		 */
3551 		newfde = &fdp->fd_ofiles[indx];
3552 		oldfde = &fdp->fd_ofiles[dfd];
3553 #ifdef CAPABILITIES
3554 		seqc_write_begin(&newfde->fde_seqc);
3555 #endif
3556 		memcpy(newfde, oldfde, fde_change_size);
3557 		oldfde->fde_file = NULL;
3558 		fdunused(fdp, dfd);
3559 #ifdef CAPABILITIES
3560 		seqc_write_end(&newfde->fde_seqc);
3561 #endif
3562 		break;
3563 	}
3564 	FILEDESC_XUNLOCK(fdp);
3565 	*indxp = indx;
3566 	return (0);
3567 }
3568 
3569 /*
3570  * This sysctl determines if we will allow a process to chroot(2) if it
3571  * has a directory open:
3572  *	0: disallowed for all processes.
3573  *	1: allowed for processes that were not already chroot(2)'ed.
3574  *	2: allowed for all processes.
3575  */
3576 
3577 static int chroot_allow_open_directories = 1;
3578 
3579 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3580     &chroot_allow_open_directories, 0,
3581     "Allow a process to chroot(2) if it has a directory open");
3582 
3583 /*
3584  * Helper function for raised chroot(2) security function:  Refuse if
3585  * any filedescriptors are open directories.
3586  */
3587 static int
3588 chroot_refuse_vdir_fds(struct filedesc *fdp)
3589 {
3590 	struct vnode *vp;
3591 	struct file *fp;
3592 	int fd, lastfile;
3593 
3594 	FILEDESC_LOCK_ASSERT(fdp);
3595 
3596 	lastfile = fdlastfile(fdp);
3597 	for (fd = 0; fd <= lastfile; fd++) {
3598 		fp = fget_locked(fdp, fd);
3599 		if (fp == NULL)
3600 			continue;
3601 		if (fp->f_type == DTYPE_VNODE) {
3602 			vp = fp->f_vnode;
3603 			if (vp->v_type == VDIR)
3604 				return (EPERM);
3605 		}
3606 	}
3607 	return (0);
3608 }
3609 
3610 static void
3611 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3612 {
3613 
3614 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3615 		vrefact(oldpwd->pwd_cdir);
3616 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3617 	}
3618 
3619 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3620 		vrefact(oldpwd->pwd_rdir);
3621 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3622 	}
3623 
3624 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3625 		vrefact(oldpwd->pwd_jdir);
3626 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3627 	}
3628 }
3629 
3630 struct pwd *
3631 pwd_hold_pwddesc(struct pwddesc *pdp)
3632 {
3633 	struct pwd *pwd;
3634 
3635 	PWDDESC_ASSERT_XLOCKED(pdp);
3636 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3637 	if (pwd != NULL)
3638 		refcount_acquire(&pwd->pwd_refcount);
3639 	return (pwd);
3640 }
3641 
3642 bool
3643 pwd_hold_smr(struct pwd *pwd)
3644 {
3645 
3646 	MPASS(pwd != NULL);
3647 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3648 		return (true);
3649 	}
3650 	return (false);
3651 }
3652 
3653 struct pwd *
3654 pwd_hold(struct thread *td)
3655 {
3656 	struct pwddesc *pdp;
3657 	struct pwd *pwd;
3658 
3659 	pdp = td->td_proc->p_pd;
3660 
3661 	vfs_smr_enter();
3662 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3663 	if (pwd_hold_smr(pwd)) {
3664 		vfs_smr_exit();
3665 		return (pwd);
3666 	}
3667 	vfs_smr_exit();
3668 	PWDDESC_XLOCK(pdp);
3669 	pwd = pwd_hold_pwddesc(pdp);
3670 	MPASS(pwd != NULL);
3671 	PWDDESC_XUNLOCK(pdp);
3672 	return (pwd);
3673 }
3674 
3675 struct pwd *
3676 pwd_get_smr(void)
3677 {
3678 	struct pwd *pwd;
3679 
3680 	pwd = vfs_smr_entered_load(&curproc->p_pd->pd_pwd);
3681 	MPASS(pwd != NULL);
3682 	return (pwd);
3683 }
3684 
3685 static struct pwd *
3686 pwd_alloc(void)
3687 {
3688 	struct pwd *pwd;
3689 
3690 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3691 	bzero(pwd, sizeof(*pwd));
3692 	refcount_init(&pwd->pwd_refcount, 1);
3693 	return (pwd);
3694 }
3695 
3696 void
3697 pwd_drop(struct pwd *pwd)
3698 {
3699 
3700 	if (!refcount_release(&pwd->pwd_refcount))
3701 		return;
3702 
3703 	if (pwd->pwd_cdir != NULL)
3704 		vrele(pwd->pwd_cdir);
3705 	if (pwd->pwd_rdir != NULL)
3706 		vrele(pwd->pwd_rdir);
3707 	if (pwd->pwd_jdir != NULL)
3708 		vrele(pwd->pwd_jdir);
3709 	uma_zfree_smr(pwd_zone, pwd);
3710 }
3711 
3712 /*
3713 * Common routine for kern_chroot() and jail_attach().  The caller is
3714 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3715 * authorize this operation.
3716 */
3717 int
3718 pwd_chroot(struct thread *td, struct vnode *vp)
3719 {
3720 	struct pwddesc *pdp;
3721 	struct filedesc *fdp;
3722 	struct pwd *newpwd, *oldpwd;
3723 	int error;
3724 
3725 	fdp = td->td_proc->p_fd;
3726 	pdp = td->td_proc->p_pd;
3727 	newpwd = pwd_alloc();
3728 	FILEDESC_SLOCK(fdp);
3729 	PWDDESC_XLOCK(pdp);
3730 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3731 	if (chroot_allow_open_directories == 0 ||
3732 	    (chroot_allow_open_directories == 1 &&
3733 	    oldpwd->pwd_rdir != rootvnode)) {
3734 		error = chroot_refuse_vdir_fds(fdp);
3735 		FILEDESC_SUNLOCK(fdp);
3736 		if (error != 0) {
3737 			PWDDESC_XUNLOCK(pdp);
3738 			pwd_drop(newpwd);
3739 			return (error);
3740 		}
3741 	} else {
3742 		FILEDESC_SUNLOCK(fdp);
3743 	}
3744 
3745 	vrefact(vp);
3746 	newpwd->pwd_rdir = vp;
3747 	if (oldpwd->pwd_jdir == NULL) {
3748 		vrefact(vp);
3749 		newpwd->pwd_jdir = vp;
3750 	}
3751 	pwd_fill(oldpwd, newpwd);
3752 	pwd_set(pdp, newpwd);
3753 	PWDDESC_XUNLOCK(pdp);
3754 	pwd_drop(oldpwd);
3755 	return (0);
3756 }
3757 
3758 void
3759 pwd_chdir(struct thread *td, struct vnode *vp)
3760 {
3761 	struct pwddesc *pdp;
3762 	struct pwd *newpwd, *oldpwd;
3763 
3764 	VNPASS(vp->v_usecount > 0, vp);
3765 
3766 	newpwd = pwd_alloc();
3767 	pdp = td->td_proc->p_pd;
3768 	PWDDESC_XLOCK(pdp);
3769 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3770 	newpwd->pwd_cdir = vp;
3771 	pwd_fill(oldpwd, newpwd);
3772 	pwd_set(pdp, newpwd);
3773 	PWDDESC_XUNLOCK(pdp);
3774 	pwd_drop(oldpwd);
3775 }
3776 
3777 void
3778 pwd_ensure_dirs(void)
3779 {
3780 	struct pwddesc *pdp;
3781 	struct pwd *oldpwd, *newpwd;
3782 
3783 	pdp = curproc->p_pd;
3784 	PWDDESC_XLOCK(pdp);
3785 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3786 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3787 		PWDDESC_XUNLOCK(pdp);
3788 		return;
3789 	}
3790 	PWDDESC_XUNLOCK(pdp);
3791 
3792 	newpwd = pwd_alloc();
3793 	PWDDESC_XLOCK(pdp);
3794 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3795 	pwd_fill(oldpwd, newpwd);
3796 	if (newpwd->pwd_cdir == NULL) {
3797 		vrefact(rootvnode);
3798 		newpwd->pwd_cdir = rootvnode;
3799 	}
3800 	if (newpwd->pwd_rdir == NULL) {
3801 		vrefact(rootvnode);
3802 		newpwd->pwd_rdir = rootvnode;
3803 	}
3804 	pwd_set(pdp, newpwd);
3805 	PWDDESC_XUNLOCK(pdp);
3806 	pwd_drop(oldpwd);
3807 }
3808 
3809 void
3810 pwd_set_rootvnode(void)
3811 {
3812 	struct pwddesc *pdp;
3813 	struct pwd *oldpwd, *newpwd;
3814 
3815 	pdp = curproc->p_pd;
3816 
3817 	newpwd = pwd_alloc();
3818 	PWDDESC_XLOCK(pdp);
3819 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3820 	vrefact(rootvnode);
3821 	newpwd->pwd_cdir = rootvnode;
3822 	vrefact(rootvnode);
3823 	newpwd->pwd_rdir = rootvnode;
3824 	pwd_fill(oldpwd, newpwd);
3825 	pwd_set(pdp, newpwd);
3826 	PWDDESC_XUNLOCK(pdp);
3827 	pwd_drop(oldpwd);
3828 }
3829 
3830 /*
3831  * Scan all active processes and prisons to see if any of them have a current
3832  * or root directory of `olddp'. If so, replace them with the new mount point.
3833  */
3834 void
3835 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3836 {
3837 	struct pwddesc *pdp;
3838 	struct pwd *newpwd, *oldpwd;
3839 	struct prison *pr;
3840 	struct proc *p;
3841 	int nrele;
3842 
3843 	if (vrefcnt(olddp) == 1)
3844 		return;
3845 	nrele = 0;
3846 	newpwd = pwd_alloc();
3847 	sx_slock(&allproc_lock);
3848 	FOREACH_PROC_IN_SYSTEM(p) {
3849 		PROC_LOCK(p);
3850 		pdp = pdhold(p);
3851 		PROC_UNLOCK(p);
3852 		if (pdp == NULL)
3853 			continue;
3854 		PWDDESC_XLOCK(pdp);
3855 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3856 		if (oldpwd == NULL ||
3857 		    (oldpwd->pwd_cdir != olddp &&
3858 		    oldpwd->pwd_rdir != olddp &&
3859 		    oldpwd->pwd_jdir != olddp)) {
3860 			PWDDESC_XUNLOCK(pdp);
3861 			pddrop(pdp);
3862 			continue;
3863 		}
3864 		if (oldpwd->pwd_cdir == olddp) {
3865 			vrefact(newdp);
3866 			newpwd->pwd_cdir = newdp;
3867 		}
3868 		if (oldpwd->pwd_rdir == olddp) {
3869 			vrefact(newdp);
3870 			newpwd->pwd_rdir = newdp;
3871 		}
3872 		if (oldpwd->pwd_jdir == olddp) {
3873 			vrefact(newdp);
3874 			newpwd->pwd_jdir = newdp;
3875 		}
3876 		pwd_fill(oldpwd, newpwd);
3877 		pwd_set(pdp, newpwd);
3878 		PWDDESC_XUNLOCK(pdp);
3879 		pwd_drop(oldpwd);
3880 		pddrop(pdp);
3881 		newpwd = pwd_alloc();
3882 	}
3883 	sx_sunlock(&allproc_lock);
3884 	pwd_drop(newpwd);
3885 	if (rootvnode == olddp) {
3886 		vrefact(newdp);
3887 		rootvnode = newdp;
3888 		nrele++;
3889 	}
3890 	mtx_lock(&prison0.pr_mtx);
3891 	if (prison0.pr_root == olddp) {
3892 		vrefact(newdp);
3893 		prison0.pr_root = newdp;
3894 		nrele++;
3895 	}
3896 	mtx_unlock(&prison0.pr_mtx);
3897 	sx_slock(&allprison_lock);
3898 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3899 		mtx_lock(&pr->pr_mtx);
3900 		if (pr->pr_root == olddp) {
3901 			vrefact(newdp);
3902 			pr->pr_root = newdp;
3903 			nrele++;
3904 		}
3905 		mtx_unlock(&pr->pr_mtx);
3906 	}
3907 	sx_sunlock(&allprison_lock);
3908 	while (nrele--)
3909 		vrele(olddp);
3910 }
3911 
3912 struct filedesc_to_leader *
3913 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3914 {
3915 	struct filedesc_to_leader *fdtol;
3916 
3917 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3918 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3919 	fdtol->fdl_refcount = 1;
3920 	fdtol->fdl_holdcount = 0;
3921 	fdtol->fdl_wakeup = 0;
3922 	fdtol->fdl_leader = leader;
3923 	if (old != NULL) {
3924 		FILEDESC_XLOCK(fdp);
3925 		fdtol->fdl_next = old->fdl_next;
3926 		fdtol->fdl_prev = old;
3927 		old->fdl_next = fdtol;
3928 		fdtol->fdl_next->fdl_prev = fdtol;
3929 		FILEDESC_XUNLOCK(fdp);
3930 	} else {
3931 		fdtol->fdl_next = fdtol;
3932 		fdtol->fdl_prev = fdtol;
3933 	}
3934 	return (fdtol);
3935 }
3936 
3937 static int
3938 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3939 {
3940 	NDSLOTTYPE *map;
3941 	struct filedesc *fdp;
3942 	int count, off, minoff;
3943 
3944 	if (*(int *)arg1 != 0)
3945 		return (EINVAL);
3946 
3947 	fdp = curproc->p_fd;
3948 	count = 0;
3949 	FILEDESC_SLOCK(fdp);
3950 	map = fdp->fd_map;
3951 	off = NDSLOT(fdp->fd_nfiles - 1);
3952 	for (minoff = NDSLOT(0); off >= minoff; --off)
3953 		count += bitcountl(map[off]);
3954 	FILEDESC_SUNLOCK(fdp);
3955 
3956 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3957 }
3958 
3959 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3960     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3961     "Number of open file descriptors");
3962 
3963 /*
3964  * Get file structures globally.
3965  */
3966 static int
3967 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3968 {
3969 	struct xfile xf;
3970 	struct filedesc *fdp;
3971 	struct file *fp;
3972 	struct proc *p;
3973 	int error, n, lastfile;
3974 
3975 	error = sysctl_wire_old_buffer(req, 0);
3976 	if (error != 0)
3977 		return (error);
3978 	if (req->oldptr == NULL) {
3979 		n = 0;
3980 		sx_slock(&allproc_lock);
3981 		FOREACH_PROC_IN_SYSTEM(p) {
3982 			PROC_LOCK(p);
3983 			if (p->p_state == PRS_NEW) {
3984 				PROC_UNLOCK(p);
3985 				continue;
3986 			}
3987 			fdp = fdhold(p);
3988 			PROC_UNLOCK(p);
3989 			if (fdp == NULL)
3990 				continue;
3991 			/* overestimates sparse tables. */
3992 			n += fdp->fd_nfiles;
3993 			fddrop(fdp);
3994 		}
3995 		sx_sunlock(&allproc_lock);
3996 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3997 	}
3998 	error = 0;
3999 	bzero(&xf, sizeof(xf));
4000 	xf.xf_size = sizeof(xf);
4001 	sx_slock(&allproc_lock);
4002 	FOREACH_PROC_IN_SYSTEM(p) {
4003 		PROC_LOCK(p);
4004 		if (p->p_state == PRS_NEW) {
4005 			PROC_UNLOCK(p);
4006 			continue;
4007 		}
4008 		if (p_cansee(req->td, p) != 0) {
4009 			PROC_UNLOCK(p);
4010 			continue;
4011 		}
4012 		xf.xf_pid = p->p_pid;
4013 		xf.xf_uid = p->p_ucred->cr_uid;
4014 		fdp = fdhold(p);
4015 		PROC_UNLOCK(p);
4016 		if (fdp == NULL)
4017 			continue;
4018 		FILEDESC_SLOCK(fdp);
4019 		lastfile = fdlastfile(fdp);
4020 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4021 		    n++) {
4022 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4023 				continue;
4024 			xf.xf_fd = n;
4025 			xf.xf_file = (uintptr_t)fp;
4026 			xf.xf_data = (uintptr_t)fp->f_data;
4027 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4028 			xf.xf_type = (uintptr_t)fp->f_type;
4029 			xf.xf_count = refcount_load(&fp->f_count);
4030 			xf.xf_msgcount = 0;
4031 			xf.xf_offset = foffset_get(fp);
4032 			xf.xf_flag = fp->f_flag;
4033 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4034 			if (error)
4035 				break;
4036 		}
4037 		FILEDESC_SUNLOCK(fdp);
4038 		fddrop(fdp);
4039 		if (error)
4040 			break;
4041 	}
4042 	sx_sunlock(&allproc_lock);
4043 	return (error);
4044 }
4045 
4046 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4047     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4048 
4049 #ifdef KINFO_FILE_SIZE
4050 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4051 #endif
4052 
4053 static int
4054 xlate_fflags(int fflags)
4055 {
4056 	static const struct {
4057 		int	fflag;
4058 		int	kf_fflag;
4059 	} fflags_table[] = {
4060 		{ FAPPEND, KF_FLAG_APPEND },
4061 		{ FASYNC, KF_FLAG_ASYNC },
4062 		{ FFSYNC, KF_FLAG_FSYNC },
4063 		{ FHASLOCK, KF_FLAG_HASLOCK },
4064 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4065 		{ FREAD, KF_FLAG_READ },
4066 		{ FWRITE, KF_FLAG_WRITE },
4067 		{ O_CREAT, KF_FLAG_CREAT },
4068 		{ O_DIRECT, KF_FLAG_DIRECT },
4069 		{ O_EXCL, KF_FLAG_EXCL },
4070 		{ O_EXEC, KF_FLAG_EXEC },
4071 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4072 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4073 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4074 		{ O_TRUNC, KF_FLAG_TRUNC }
4075 	};
4076 	unsigned int i;
4077 	int kflags;
4078 
4079 	kflags = 0;
4080 	for (i = 0; i < nitems(fflags_table); i++)
4081 		if (fflags & fflags_table[i].fflag)
4082 			kflags |=  fflags_table[i].kf_fflag;
4083 	return (kflags);
4084 }
4085 
4086 /* Trim unused data from kf_path by truncating the structure size. */
4087 void
4088 pack_kinfo(struct kinfo_file *kif)
4089 {
4090 
4091 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4092 	    strlen(kif->kf_path) + 1;
4093 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4094 }
4095 
4096 static void
4097 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4098     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4099 {
4100 	int error;
4101 
4102 	bzero(kif, sizeof(*kif));
4103 
4104 	/* Set a default type to allow for empty fill_kinfo() methods. */
4105 	kif->kf_type = KF_TYPE_UNKNOWN;
4106 	kif->kf_flags = xlate_fflags(fp->f_flag);
4107 	if (rightsp != NULL)
4108 		kif->kf_cap_rights = *rightsp;
4109 	else
4110 		cap_rights_init_zero(&kif->kf_cap_rights);
4111 	kif->kf_fd = fd;
4112 	kif->kf_ref_count = refcount_load(&fp->f_count);
4113 	kif->kf_offset = foffset_get(fp);
4114 
4115 	/*
4116 	 * This may drop the filedesc lock, so the 'fp' cannot be
4117 	 * accessed after this call.
4118 	 */
4119 	error = fo_fill_kinfo(fp, kif, fdp);
4120 	if (error == 0)
4121 		kif->kf_status |= KF_ATTR_VALID;
4122 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4123 		pack_kinfo(kif);
4124 	else
4125 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4126 }
4127 
4128 static void
4129 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4130     struct kinfo_file *kif, int flags)
4131 {
4132 	int error;
4133 
4134 	bzero(kif, sizeof(*kif));
4135 
4136 	kif->kf_type = KF_TYPE_VNODE;
4137 	error = vn_fill_kinfo_vnode(vp, kif);
4138 	if (error == 0)
4139 		kif->kf_status |= KF_ATTR_VALID;
4140 	kif->kf_flags = xlate_fflags(fflags);
4141 	cap_rights_init_zero(&kif->kf_cap_rights);
4142 	kif->kf_fd = fd;
4143 	kif->kf_ref_count = -1;
4144 	kif->kf_offset = -1;
4145 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4146 		pack_kinfo(kif);
4147 	else
4148 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4149 	vrele(vp);
4150 }
4151 
4152 struct export_fd_buf {
4153 	struct filedesc		*fdp;
4154 	struct pwddesc	*pdp;
4155 	struct sbuf 		*sb;
4156 	ssize_t			remainder;
4157 	struct kinfo_file	kif;
4158 	int			flags;
4159 };
4160 
4161 static int
4162 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4163 {
4164 	struct kinfo_file *kif;
4165 
4166 	kif = &efbuf->kif;
4167 	if (efbuf->remainder != -1) {
4168 		if (efbuf->remainder < kif->kf_structsize) {
4169 			/* Terminate export. */
4170 			efbuf->remainder = 0;
4171 			return (0);
4172 		}
4173 		efbuf->remainder -= kif->kf_structsize;
4174 	}
4175 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4176 }
4177 
4178 static int
4179 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4180     struct export_fd_buf *efbuf)
4181 {
4182 	int error;
4183 
4184 	if (efbuf->remainder == 0)
4185 		return (0);
4186 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4187 	    efbuf->flags);
4188 	FILEDESC_SUNLOCK(efbuf->fdp);
4189 	error = export_kinfo_to_sb(efbuf);
4190 	FILEDESC_SLOCK(efbuf->fdp);
4191 	return (error);
4192 }
4193 
4194 static int
4195 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4196     struct export_fd_buf *efbuf)
4197 {
4198 	int error;
4199 
4200 	if (efbuf->remainder == 0)
4201 		return (0);
4202 	if (efbuf->pdp != NULL)
4203 		PWDDESC_XUNLOCK(efbuf->pdp);
4204 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4205 	error = export_kinfo_to_sb(efbuf);
4206 	if (efbuf->pdp != NULL)
4207 		PWDDESC_XLOCK(efbuf->pdp);
4208 	return (error);
4209 }
4210 
4211 /*
4212  * Store a process file descriptor information to sbuf.
4213  *
4214  * Takes a locked proc as argument, and returns with the proc unlocked.
4215  */
4216 int
4217 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4218     int flags)
4219 {
4220 	struct file *fp;
4221 	struct filedesc *fdp;
4222 	struct pwddesc *pdp;
4223 	struct export_fd_buf *efbuf;
4224 	struct vnode *cttyvp, *textvp, *tracevp;
4225 	struct pwd *pwd;
4226 	int error, i, lastfile;
4227 	cap_rights_t rights;
4228 
4229 	PROC_LOCK_ASSERT(p, MA_OWNED);
4230 
4231 	/* ktrace vnode */
4232 	tracevp = p->p_tracevp;
4233 	if (tracevp != NULL)
4234 		vrefact(tracevp);
4235 	/* text vnode */
4236 	textvp = p->p_textvp;
4237 	if (textvp != NULL)
4238 		vrefact(textvp);
4239 	/* Controlling tty. */
4240 	cttyvp = NULL;
4241 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4242 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4243 		if (cttyvp != NULL)
4244 			vrefact(cttyvp);
4245 	}
4246 	fdp = fdhold(p);
4247 	pdp = pdhold(p);
4248 	PROC_UNLOCK(p);
4249 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4250 	efbuf->fdp = NULL;
4251 	efbuf->pdp = NULL;
4252 	efbuf->sb = sb;
4253 	efbuf->remainder = maxlen;
4254 	efbuf->flags = flags;
4255 	if (tracevp != NULL)
4256 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4257 		    efbuf);
4258 	if (textvp != NULL)
4259 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4260 	if (cttyvp != NULL)
4261 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4262 		    efbuf);
4263 	error = 0;
4264 	if (pdp == NULL || fdp == NULL)
4265 		goto fail;
4266 	efbuf->fdp = fdp;
4267 	efbuf->pdp = pdp;
4268 	PWDDESC_XLOCK(pdp);
4269 	pwd = pwd_hold_pwddesc(pdp);
4270 	if (pwd != NULL) {
4271 		/* working directory */
4272 		if (pwd->pwd_cdir != NULL) {
4273 			vrefact(pwd->pwd_cdir);
4274 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4275 		}
4276 		/* root directory */
4277 		if (pwd->pwd_rdir != NULL) {
4278 			vrefact(pwd->pwd_rdir);
4279 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4280 		}
4281 		/* jail directory */
4282 		if (pwd->pwd_jdir != NULL) {
4283 			vrefact(pwd->pwd_jdir);
4284 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4285 		}
4286 	}
4287 	PWDDESC_XUNLOCK(pdp);
4288 	if (pwd != NULL)
4289 		pwd_drop(pwd);
4290 	FILEDESC_SLOCK(fdp);
4291 	lastfile = fdlastfile(fdp);
4292 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4293 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4294 			continue;
4295 #ifdef CAPABILITIES
4296 		rights = *cap_rights(fdp, i);
4297 #else /* !CAPABILITIES */
4298 		rights = cap_no_rights;
4299 #endif
4300 		/*
4301 		 * Create sysctl entry.  It is OK to drop the filedesc
4302 		 * lock inside of export_file_to_sb() as we will
4303 		 * re-validate and re-evaluate its properties when the
4304 		 * loop continues.
4305 		 */
4306 		error = export_file_to_sb(fp, i, &rights, efbuf);
4307 		if (error != 0 || efbuf->remainder == 0)
4308 			break;
4309 	}
4310 	FILEDESC_SUNLOCK(fdp);
4311 fail:
4312 	if (fdp != NULL)
4313 		fddrop(fdp);
4314 	if (pdp != NULL)
4315 		pddrop(pdp);
4316 	free(efbuf, M_TEMP);
4317 	return (error);
4318 }
4319 
4320 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4321 
4322 /*
4323  * Get per-process file descriptors for use by procstat(1), et al.
4324  */
4325 static int
4326 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4327 {
4328 	struct sbuf sb;
4329 	struct proc *p;
4330 	ssize_t maxlen;
4331 	int error, error2, *name;
4332 
4333 	name = (int *)arg1;
4334 
4335 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4336 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4337 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4338 	if (error != 0) {
4339 		sbuf_delete(&sb);
4340 		return (error);
4341 	}
4342 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4343 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4344 	    KERN_FILEDESC_PACK_KINFO);
4345 	error2 = sbuf_finish(&sb);
4346 	sbuf_delete(&sb);
4347 	return (error != 0 ? error : error2);
4348 }
4349 
4350 #ifdef COMPAT_FREEBSD7
4351 #ifdef KINFO_OFILE_SIZE
4352 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4353 #endif
4354 
4355 static void
4356 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4357 {
4358 
4359 	okif->kf_structsize = sizeof(*okif);
4360 	okif->kf_type = kif->kf_type;
4361 	okif->kf_fd = kif->kf_fd;
4362 	okif->kf_ref_count = kif->kf_ref_count;
4363 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4364 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4365 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4366 	okif->kf_offset = kif->kf_offset;
4367 	if (kif->kf_type == KF_TYPE_VNODE)
4368 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4369 	else
4370 		okif->kf_vnode_type = KF_VTYPE_VNON;
4371 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4372 	if (kif->kf_type == KF_TYPE_SOCKET) {
4373 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4374 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4375 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4376 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4377 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4378 	} else {
4379 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4380 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4381 	}
4382 }
4383 
4384 static int
4385 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4386     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4387 {
4388 	int error;
4389 
4390 	vrefact(vp);
4391 	PWDDESC_XUNLOCK(pdp);
4392 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4393 	kinfo_to_okinfo(kif, okif);
4394 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4395 	PWDDESC_XLOCK(pdp);
4396 	return (error);
4397 }
4398 
4399 /*
4400  * Get per-process file descriptors for use by procstat(1), et al.
4401  */
4402 static int
4403 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4404 {
4405 	struct kinfo_ofile *okif;
4406 	struct kinfo_file *kif;
4407 	struct filedesc *fdp;
4408 	struct pwddesc *pdp;
4409 	struct pwd *pwd;
4410 	int error, i, lastfile, *name;
4411 	struct file *fp;
4412 	struct proc *p;
4413 
4414 	name = (int *)arg1;
4415 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4416 	if (error != 0)
4417 		return (error);
4418 	fdp = fdhold(p);
4419 	if (fdp != NULL)
4420 		pdp = pdhold(p);
4421 	PROC_UNLOCK(p);
4422 	if (fdp == NULL || pdp == NULL) {
4423 		if (fdp != NULL)
4424 			fddrop(fdp);
4425 		return (ENOENT);
4426 	}
4427 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4428 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4429 	PWDDESC_XLOCK(pdp);
4430 	pwd = pwd_hold_pwddesc(pdp);
4431 	if (pwd != NULL) {
4432 		if (pwd->pwd_cdir != NULL)
4433 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4434 			    okif, pdp, req);
4435 		if (pwd->pwd_rdir != NULL)
4436 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4437 			    okif, pdp, req);
4438 		if (pwd->pwd_jdir != NULL)
4439 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4440 			    okif, pdp, req);
4441 	}
4442 	PWDDESC_XUNLOCK(pdp);
4443 	if (pwd != NULL)
4444 		pwd_drop(pwd);
4445 	FILEDESC_SLOCK(fdp);
4446 	lastfile = fdlastfile(fdp);
4447 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4448 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4449 			continue;
4450 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4451 		    KERN_FILEDESC_PACK_KINFO);
4452 		FILEDESC_SUNLOCK(fdp);
4453 		kinfo_to_okinfo(kif, okif);
4454 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4455 		FILEDESC_SLOCK(fdp);
4456 		if (error)
4457 			break;
4458 	}
4459 	FILEDESC_SUNLOCK(fdp);
4460 	fddrop(fdp);
4461 	pddrop(pdp);
4462 	free(kif, M_TEMP);
4463 	free(okif, M_TEMP);
4464 	return (0);
4465 }
4466 
4467 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4468     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4469     "Process ofiledesc entries");
4470 #endif	/* COMPAT_FREEBSD7 */
4471 
4472 int
4473 vntype_to_kinfo(int vtype)
4474 {
4475 	struct {
4476 		int	vtype;
4477 		int	kf_vtype;
4478 	} vtypes_table[] = {
4479 		{ VBAD, KF_VTYPE_VBAD },
4480 		{ VBLK, KF_VTYPE_VBLK },
4481 		{ VCHR, KF_VTYPE_VCHR },
4482 		{ VDIR, KF_VTYPE_VDIR },
4483 		{ VFIFO, KF_VTYPE_VFIFO },
4484 		{ VLNK, KF_VTYPE_VLNK },
4485 		{ VNON, KF_VTYPE_VNON },
4486 		{ VREG, KF_VTYPE_VREG },
4487 		{ VSOCK, KF_VTYPE_VSOCK }
4488 	};
4489 	unsigned int i;
4490 
4491 	/*
4492 	 * Perform vtype translation.
4493 	 */
4494 	for (i = 0; i < nitems(vtypes_table); i++)
4495 		if (vtypes_table[i].vtype == vtype)
4496 			return (vtypes_table[i].kf_vtype);
4497 
4498 	return (KF_VTYPE_UNKNOWN);
4499 }
4500 
4501 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4502     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4503     "Process filedesc entries");
4504 
4505 /*
4506  * Store a process current working directory information to sbuf.
4507  *
4508  * Takes a locked proc as argument, and returns with the proc unlocked.
4509  */
4510 int
4511 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4512 {
4513 	struct pwddesc *pdp;
4514 	struct pwd *pwd;
4515 	struct export_fd_buf *efbuf;
4516 	struct vnode *cdir;
4517 	int error;
4518 
4519 	PROC_LOCK_ASSERT(p, MA_OWNED);
4520 
4521 	pdp = pdhold(p);
4522 	PROC_UNLOCK(p);
4523 	if (pdp == NULL)
4524 		return (EINVAL);
4525 
4526 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4527 	efbuf->pdp = pdp;
4528 	efbuf->sb = sb;
4529 	efbuf->remainder = maxlen;
4530 
4531 	PWDDESC_XLOCK(pdp);
4532 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4533 	cdir = pwd->pwd_cdir;
4534 	if (cdir == NULL) {
4535 		error = EINVAL;
4536 	} else {
4537 		vrefact(cdir);
4538 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4539 	}
4540 	PWDDESC_XUNLOCK(pdp);
4541 	pddrop(pdp);
4542 	free(efbuf, M_TEMP);
4543 	return (error);
4544 }
4545 
4546 /*
4547  * Get per-process current working directory.
4548  */
4549 static int
4550 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4551 {
4552 	struct sbuf sb;
4553 	struct proc *p;
4554 	ssize_t maxlen;
4555 	int error, error2, *name;
4556 
4557 	name = (int *)arg1;
4558 
4559 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4560 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4561 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4562 	if (error != 0) {
4563 		sbuf_delete(&sb);
4564 		return (error);
4565 	}
4566 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4567 	error = kern_proc_cwd_out(p, &sb, maxlen);
4568 	error2 = sbuf_finish(&sb);
4569 	sbuf_delete(&sb);
4570 	return (error != 0 ? error : error2);
4571 }
4572 
4573 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4574     sysctl_kern_proc_cwd, "Process current working directory");
4575 
4576 #ifdef DDB
4577 /*
4578  * For the purposes of debugging, generate a human-readable string for the
4579  * file type.
4580  */
4581 static const char *
4582 file_type_to_name(short type)
4583 {
4584 
4585 	switch (type) {
4586 	case 0:
4587 		return ("zero");
4588 	case DTYPE_VNODE:
4589 		return ("vnode");
4590 	case DTYPE_SOCKET:
4591 		return ("socket");
4592 	case DTYPE_PIPE:
4593 		return ("pipe");
4594 	case DTYPE_FIFO:
4595 		return ("fifo");
4596 	case DTYPE_KQUEUE:
4597 		return ("kqueue");
4598 	case DTYPE_CRYPTO:
4599 		return ("crypto");
4600 	case DTYPE_MQUEUE:
4601 		return ("mqueue");
4602 	case DTYPE_SHM:
4603 		return ("shm");
4604 	case DTYPE_SEM:
4605 		return ("ksem");
4606 	case DTYPE_PTS:
4607 		return ("pts");
4608 	case DTYPE_DEV:
4609 		return ("dev");
4610 	case DTYPE_PROCDESC:
4611 		return ("proc");
4612 	case DTYPE_LINUXEFD:
4613 		return ("levent");
4614 	case DTYPE_LINUXTFD:
4615 		return ("ltimer");
4616 	default:
4617 		return ("unkn");
4618 	}
4619 }
4620 
4621 /*
4622  * For the purposes of debugging, identify a process (if any, perhaps one of
4623  * many) that references the passed file in its file descriptor array. Return
4624  * NULL if none.
4625  */
4626 static struct proc *
4627 file_to_first_proc(struct file *fp)
4628 {
4629 	struct filedesc *fdp;
4630 	struct proc *p;
4631 	int n;
4632 
4633 	FOREACH_PROC_IN_SYSTEM(p) {
4634 		if (p->p_state == PRS_NEW)
4635 			continue;
4636 		fdp = p->p_fd;
4637 		if (fdp == NULL)
4638 			continue;
4639 		for (n = 0; n < fdp->fd_nfiles; n++) {
4640 			if (fp == fdp->fd_ofiles[n].fde_file)
4641 				return (p);
4642 		}
4643 	}
4644 	return (NULL);
4645 }
4646 
4647 static void
4648 db_print_file(struct file *fp, int header)
4649 {
4650 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4651 	struct proc *p;
4652 
4653 	if (header)
4654 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4655 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4656 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4657 		    "FCmd");
4658 	p = file_to_first_proc(fp);
4659 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4660 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4661 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4662 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4663 
4664 #undef XPTRWIDTH
4665 }
4666 
4667 DB_SHOW_COMMAND(file, db_show_file)
4668 {
4669 	struct file *fp;
4670 
4671 	if (!have_addr) {
4672 		db_printf("usage: show file <addr>\n");
4673 		return;
4674 	}
4675 	fp = (struct file *)addr;
4676 	db_print_file(fp, 1);
4677 }
4678 
4679 DB_SHOW_COMMAND(files, db_show_files)
4680 {
4681 	struct filedesc *fdp;
4682 	struct file *fp;
4683 	struct proc *p;
4684 	int header;
4685 	int n;
4686 
4687 	header = 1;
4688 	FOREACH_PROC_IN_SYSTEM(p) {
4689 		if (p->p_state == PRS_NEW)
4690 			continue;
4691 		if ((fdp = p->p_fd) == NULL)
4692 			continue;
4693 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4694 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4695 				continue;
4696 			db_print_file(fp, header);
4697 			header = 0;
4698 		}
4699 	}
4700 }
4701 #endif
4702 
4703 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4704     &maxfilesperproc, 0, "Maximum files allowed open per process");
4705 
4706 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4707     &maxfiles, 0, "Maximum number of files");
4708 
4709 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4710     &openfiles, 0, "System-wide number of open files");
4711 
4712 /* ARGSUSED*/
4713 static void
4714 filelistinit(void *dummy)
4715 {
4716 
4717 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4718 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4719 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4720 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4721 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4722 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4723 	/*
4724 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4725 	 * the vnode zone.
4726 	 */
4727 	vfs_smr = uma_zone_get_smr(pwd_zone);
4728 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4729 }
4730 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4731 
4732 /*-------------------------------------------------------------------*/
4733 
4734 static int
4735 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4736     int flags, struct thread *td)
4737 {
4738 
4739 	return (EBADF);
4740 }
4741 
4742 static int
4743 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4744     struct thread *td)
4745 {
4746 
4747 	return (EINVAL);
4748 }
4749 
4750 static int
4751 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4752     struct thread *td)
4753 {
4754 
4755 	return (EBADF);
4756 }
4757 
4758 static int
4759 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4760     struct thread *td)
4761 {
4762 
4763 	return (0);
4764 }
4765 
4766 static int
4767 badfo_kqfilter(struct file *fp, struct knote *kn)
4768 {
4769 
4770 	return (EBADF);
4771 }
4772 
4773 static int
4774 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4775     struct thread *td)
4776 {
4777 
4778 	return (EBADF);
4779 }
4780 
4781 static int
4782 badfo_close(struct file *fp, struct thread *td)
4783 {
4784 
4785 	return (0);
4786 }
4787 
4788 static int
4789 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4790     struct thread *td)
4791 {
4792 
4793 	return (EBADF);
4794 }
4795 
4796 static int
4797 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4798     struct thread *td)
4799 {
4800 
4801 	return (EBADF);
4802 }
4803 
4804 static int
4805 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4806     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4807     struct thread *td)
4808 {
4809 
4810 	return (EBADF);
4811 }
4812 
4813 static int
4814 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4815 {
4816 
4817 	return (0);
4818 }
4819 
4820 struct fileops badfileops = {
4821 	.fo_read = badfo_readwrite,
4822 	.fo_write = badfo_readwrite,
4823 	.fo_truncate = badfo_truncate,
4824 	.fo_ioctl = badfo_ioctl,
4825 	.fo_poll = badfo_poll,
4826 	.fo_kqfilter = badfo_kqfilter,
4827 	.fo_stat = badfo_stat,
4828 	.fo_close = badfo_close,
4829 	.fo_chmod = badfo_chmod,
4830 	.fo_chown = badfo_chown,
4831 	.fo_sendfile = badfo_sendfile,
4832 	.fo_fill_kinfo = badfo_fill_kinfo,
4833 };
4834 
4835 int
4836 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4837     int flags, struct thread *td)
4838 {
4839 
4840 	return (EOPNOTSUPP);
4841 }
4842 
4843 int
4844 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4845     struct thread *td)
4846 {
4847 
4848 	return (EINVAL);
4849 }
4850 
4851 int
4852 invfo_ioctl(struct file *fp, u_long com, void *data,
4853     struct ucred *active_cred, struct thread *td)
4854 {
4855 
4856 	return (ENOTTY);
4857 }
4858 
4859 int
4860 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4861     struct thread *td)
4862 {
4863 
4864 	return (poll_no_poll(events));
4865 }
4866 
4867 int
4868 invfo_kqfilter(struct file *fp, struct knote *kn)
4869 {
4870 
4871 	return (EINVAL);
4872 }
4873 
4874 int
4875 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4876     struct thread *td)
4877 {
4878 
4879 	return (EINVAL);
4880 }
4881 
4882 int
4883 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4884     struct thread *td)
4885 {
4886 
4887 	return (EINVAL);
4888 }
4889 
4890 int
4891 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4892     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4893     struct thread *td)
4894 {
4895 
4896 	return (EINVAL);
4897 }
4898 
4899 /*-------------------------------------------------------------------*/
4900 
4901 /*
4902  * File Descriptor pseudo-device driver (/dev/fd/).
4903  *
4904  * Opening minor device N dup()s the file (if any) connected to file
4905  * descriptor N belonging to the calling process.  Note that this driver
4906  * consists of only the ``open()'' routine, because all subsequent
4907  * references to this file will be direct to the other driver.
4908  *
4909  * XXX: we could give this one a cloning event handler if necessary.
4910  */
4911 
4912 /* ARGSUSED */
4913 static int
4914 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4915 {
4916 
4917 	/*
4918 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4919 	 * the file descriptor being sought for duplication. The error
4920 	 * return ensures that the vnode for this device will be released
4921 	 * by vn_open. Open will detect this special error and take the
4922 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4923 	 * will simply report the error.
4924 	 */
4925 	td->td_dupfd = dev2unit(dev);
4926 	return (ENODEV);
4927 }
4928 
4929 static struct cdevsw fildesc_cdevsw = {
4930 	.d_version =	D_VERSION,
4931 	.d_open =	fdopen,
4932 	.d_name =	"FD",
4933 };
4934 
4935 static void
4936 fildesc_drvinit(void *unused)
4937 {
4938 	struct cdev *dev;
4939 
4940 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4941 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4942 	make_dev_alias(dev, "stdin");
4943 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4944 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4945 	make_dev_alias(dev, "stdout");
4946 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4947 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4948 	make_dev_alias(dev, "stderr");
4949 }
4950 
4951 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4952