xref: /freebsd/sys/kern/kern_descrip.c (revision 7e1d3eefd410ca0fbae5a217422821244c3eeee4)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/poll.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sbuf.h>
71 #include <sys/signalvar.h>
72 #include <sys/kdb.h>
73 #include <sys/smr.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
80 #include <sys/user.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97     "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
100 
101 MALLOC_DECLARE(M_FADVISE);
102 
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
106 VFS_SMR_DECLARE;
107 
108 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
109 		    struct thread *td, bool holdleaders, bool audit);
110 static int	fd_first_free(struct filedesc *fdp, int low, int size);
111 static void	fdgrowtable(struct filedesc *fdp, int nfd);
112 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
113 static void	fdunused(struct filedesc *fdp, int fd);
114 static void	fdused(struct filedesc *fdp, int fd);
115 static int	getmaxfd(struct thread *td);
116 static u_long	*filecaps_copy_prep(const struct filecaps *src);
117 static void	filecaps_copy_finish(const struct filecaps *src,
118 		    struct filecaps *dst, u_long *ioctls);
119 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
120 static void	filecaps_free_finish(u_long *ioctls);
121 
122 static struct pwd *pwd_alloc(void);
123 
124 /*
125  * Each process has:
126  *
127  * - An array of open file descriptors (fd_ofiles)
128  * - An array of file flags (fd_ofileflags)
129  * - A bitmap recording which descriptors are in use (fd_map)
130  *
131  * A process starts out with NDFILE descriptors.  The value of NDFILE has
132  * been selected based the historical limit of 20 open files, and an
133  * assumption that the majority of processes, especially short-lived
134  * processes like shells, will never need more.
135  *
136  * If this initial allocation is exhausted, a larger descriptor table and
137  * map are allocated dynamically, and the pointers in the process's struct
138  * filedesc are updated to point to those.  This is repeated every time
139  * the process runs out of file descriptors (provided it hasn't hit its
140  * resource limit).
141  *
142  * Since threads may hold references to individual descriptor table
143  * entries, the tables are never freed.  Instead, they are placed on a
144  * linked list and freed only when the struct filedesc is released.
145  */
146 #define NDFILE		20
147 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
148 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
149 #define NDSLOT(x)	((x) / NDENTRIES)
150 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
151 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
152 
153 /*
154  * SLIST entry used to keep track of ofiles which must be reclaimed when
155  * the process exits.
156  */
157 struct freetable {
158 	struct fdescenttbl *ft_table;
159 	SLIST_ENTRY(freetable) ft_next;
160 };
161 
162 /*
163  * Initial allocation: a filedesc structure + the head of SLIST used to
164  * keep track of old ofiles + enough space for NDFILE descriptors.
165  */
166 
167 struct fdescenttbl0 {
168 	int	fdt_nfiles;
169 	struct	filedescent fdt_ofiles[NDFILE];
170 };
171 
172 struct filedesc0 {
173 	struct filedesc fd_fd;
174 	SLIST_HEAD(, freetable) fd_free;
175 	struct	fdescenttbl0 fd_dfiles;
176 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
177 };
178 
179 /*
180  * Descriptor management.
181  */
182 static int __exclusive_cache_line openfiles; /* actual number of open files */
183 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
184 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
185 
186 /*
187  * If low >= size, just return low. Otherwise find the first zero bit in the
188  * given bitmap, starting at low and not exceeding size - 1. Return size if
189  * not found.
190  */
191 static int
192 fd_first_free(struct filedesc *fdp, int low, int size)
193 {
194 	NDSLOTTYPE *map = fdp->fd_map;
195 	NDSLOTTYPE mask;
196 	int off, maxoff;
197 
198 	if (low >= size)
199 		return (low);
200 
201 	off = NDSLOT(low);
202 	if (low % NDENTRIES) {
203 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
204 		if ((mask &= ~map[off]) != 0UL)
205 			return (off * NDENTRIES + ffsl(mask) - 1);
206 		++off;
207 	}
208 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
209 		if (map[off] != ~0UL)
210 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
211 	return (size);
212 }
213 
214 /*
215  * Find the last used fd.
216  *
217  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
218  * Otherwise use fdlastfile.
219  */
220 int
221 fdlastfile_single(struct filedesc *fdp)
222 {
223 	NDSLOTTYPE *map = fdp->fd_map;
224 	int off, minoff;
225 
226 	off = NDSLOT(fdp->fd_nfiles - 1);
227 	for (minoff = NDSLOT(0); off >= minoff; --off)
228 		if (map[off] != 0)
229 			return (off * NDENTRIES + flsl(map[off]) - 1);
230 	return (-1);
231 }
232 
233 int
234 fdlastfile(struct filedesc *fdp)
235 {
236 
237 	FILEDESC_LOCK_ASSERT(fdp);
238 	return (fdlastfile_single(fdp));
239 }
240 
241 static int
242 fdisused(struct filedesc *fdp, int fd)
243 {
244 
245 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
246 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
247 
248 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
249 }
250 
251 /*
252  * Mark a file descriptor as used.
253  */
254 static void
255 fdused_init(struct filedesc *fdp, int fd)
256 {
257 
258 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
259 
260 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
261 }
262 
263 static void
264 fdused(struct filedesc *fdp, int fd)
265 {
266 
267 	FILEDESC_XLOCK_ASSERT(fdp);
268 
269 	fdused_init(fdp, fd);
270 	if (fd == fdp->fd_freefile)
271 		fdp->fd_freefile++;
272 }
273 
274 /*
275  * Mark a file descriptor as unused.
276  */
277 static void
278 fdunused(struct filedesc *fdp, int fd)
279 {
280 
281 	FILEDESC_XLOCK_ASSERT(fdp);
282 
283 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
284 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
285 	    ("fd=%d is still in use", fd));
286 
287 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
288 	if (fd < fdp->fd_freefile)
289 		fdp->fd_freefile = fd;
290 }
291 
292 /*
293  * Free a file descriptor.
294  *
295  * Avoid some work if fdp is about to be destroyed.
296  */
297 static inline void
298 fdefree_last(struct filedescent *fde)
299 {
300 
301 	filecaps_free(&fde->fde_caps);
302 }
303 
304 static inline void
305 fdfree(struct filedesc *fdp, int fd)
306 {
307 	struct filedescent *fde;
308 
309 	FILEDESC_XLOCK_ASSERT(fdp);
310 	fde = &fdp->fd_ofiles[fd];
311 #ifdef CAPABILITIES
312 	seqc_write_begin(&fde->fde_seqc);
313 #endif
314 	fde->fde_file = NULL;
315 #ifdef CAPABILITIES
316 	seqc_write_end(&fde->fde_seqc);
317 #endif
318 	fdefree_last(fde);
319 	fdunused(fdp, fd);
320 }
321 
322 /*
323  * System calls on descriptors.
324  */
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
327 	int	dummy;
328 };
329 #endif
330 /* ARGSUSED */
331 int
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
333 {
334 #ifdef	RACCT
335 	uint64_t lim;
336 #endif
337 
338 	td->td_retval[0] = getmaxfd(td);
339 #ifdef	RACCT
340 	PROC_LOCK(td->td_proc);
341 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 	PROC_UNLOCK(td->td_proc);
343 	if (lim < td->td_retval[0])
344 		td->td_retval[0] = lim;
345 #endif
346 	return (0);
347 }
348 
349 /*
350  * Duplicate a file descriptor to a particular value.
351  *
352  * Note: keep in mind that a potential race condition exists when closing
353  * descriptors from a shared descriptor table (via rfork).
354  */
355 #ifndef _SYS_SYSPROTO_H_
356 struct dup2_args {
357 	u_int	from;
358 	u_int	to;
359 };
360 #endif
361 /* ARGSUSED */
362 int
363 sys_dup2(struct thread *td, struct dup2_args *uap)
364 {
365 
366 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
367 }
368 
369 /*
370  * Duplicate a file descriptor.
371  */
372 #ifndef _SYS_SYSPROTO_H_
373 struct dup_args {
374 	u_int	fd;
375 };
376 #endif
377 /* ARGSUSED */
378 int
379 sys_dup(struct thread *td, struct dup_args *uap)
380 {
381 
382 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
383 }
384 
385 /*
386  * The file control system call.
387  */
388 #ifndef _SYS_SYSPROTO_H_
389 struct fcntl_args {
390 	int	fd;
391 	int	cmd;
392 	long	arg;
393 };
394 #endif
395 /* ARGSUSED */
396 int
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
398 {
399 
400 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
401 }
402 
403 int
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
405 {
406 	struct flock fl;
407 	struct __oflock ofl;
408 	intptr_t arg1;
409 	int error, newcmd;
410 
411 	error = 0;
412 	newcmd = cmd;
413 	switch (cmd) {
414 	case F_OGETLK:
415 	case F_OSETLK:
416 	case F_OSETLKW:
417 		/*
418 		 * Convert old flock structure to new.
419 		 */
420 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 		fl.l_start = ofl.l_start;
422 		fl.l_len = ofl.l_len;
423 		fl.l_pid = ofl.l_pid;
424 		fl.l_type = ofl.l_type;
425 		fl.l_whence = ofl.l_whence;
426 		fl.l_sysid = 0;
427 
428 		switch (cmd) {
429 		case F_OGETLK:
430 			newcmd = F_GETLK;
431 			break;
432 		case F_OSETLK:
433 			newcmd = F_SETLK;
434 			break;
435 		case F_OSETLKW:
436 			newcmd = F_SETLKW;
437 			break;
438 		}
439 		arg1 = (intptr_t)&fl;
440 		break;
441 	case F_GETLK:
442 	case F_SETLK:
443 	case F_SETLKW:
444 	case F_SETLK_REMOTE:
445 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 		arg1 = (intptr_t)&fl;
447 		break;
448 	default:
449 		arg1 = arg;
450 		break;
451 	}
452 	if (error)
453 		return (error);
454 	error = kern_fcntl(td, fd, newcmd, arg1);
455 	if (error)
456 		return (error);
457 	if (cmd == F_OGETLK) {
458 		ofl.l_start = fl.l_start;
459 		ofl.l_len = fl.l_len;
460 		ofl.l_pid = fl.l_pid;
461 		ofl.l_type = fl.l_type;
462 		ofl.l_whence = fl.l_whence;
463 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 	} else if (cmd == F_GETLK) {
465 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
466 	}
467 	return (error);
468 }
469 
470 int
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
472 {
473 	struct filedesc *fdp;
474 	struct flock *flp;
475 	struct file *fp, *fp2;
476 	struct filedescent *fde;
477 	struct proc *p;
478 	struct vnode *vp;
479 	struct mount *mp;
480 	int error, flg, seals, tmp;
481 	uint64_t bsize;
482 	off_t foffset;
483 
484 	error = 0;
485 	flg = F_POSIX;
486 	p = td->td_proc;
487 	fdp = p->p_fd;
488 
489 	AUDIT_ARG_FD(cmd);
490 	AUDIT_ARG_CMD(cmd);
491 	switch (cmd) {
492 	case F_DUPFD:
493 		tmp = arg;
494 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
495 		break;
496 
497 	case F_DUPFD_CLOEXEC:
498 		tmp = arg;
499 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
500 		break;
501 
502 	case F_DUP2FD:
503 		tmp = arg;
504 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
505 		break;
506 
507 	case F_DUP2FD_CLOEXEC:
508 		tmp = arg;
509 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
510 		break;
511 
512 	case F_GETFD:
513 		error = EBADF;
514 		FILEDESC_SLOCK(fdp);
515 		fde = fdeget_locked(fdp, fd);
516 		if (fde != NULL) {
517 			td->td_retval[0] =
518 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
519 			error = 0;
520 		}
521 		FILEDESC_SUNLOCK(fdp);
522 		break;
523 
524 	case F_SETFD:
525 		error = EBADF;
526 		FILEDESC_XLOCK(fdp);
527 		fde = fdeget_locked(fdp, fd);
528 		if (fde != NULL) {
529 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
531 			error = 0;
532 		}
533 		FILEDESC_XUNLOCK(fdp);
534 		break;
535 
536 	case F_GETFL:
537 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
538 		if (error != 0)
539 			break;
540 		td->td_retval[0] = OFLAGS(fp->f_flag);
541 		fdrop(fp, td);
542 		break;
543 
544 	case F_SETFL:
545 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
546 		if (error != 0)
547 			break;
548 		if (fp->f_ops == &path_fileops) {
549 			fdrop(fp, td);
550 			error = EBADF;
551 			break;
552 		}
553 		do {
554 			tmp = flg = fp->f_flag;
555 			tmp &= ~FCNTLFLAGS;
556 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
557 		} while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
558 		tmp = fp->f_flag & FNONBLOCK;
559 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
560 		if (error != 0) {
561 			fdrop(fp, td);
562 			break;
563 		}
564 		tmp = fp->f_flag & FASYNC;
565 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
566 		if (error == 0) {
567 			fdrop(fp, td);
568 			break;
569 		}
570 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
571 		tmp = 0;
572 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
573 		fdrop(fp, td);
574 		break;
575 
576 	case F_GETOWN:
577 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
578 		if (error != 0)
579 			break;
580 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
581 		if (error == 0)
582 			td->td_retval[0] = tmp;
583 		fdrop(fp, td);
584 		break;
585 
586 	case F_SETOWN:
587 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
588 		if (error != 0)
589 			break;
590 		tmp = arg;
591 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
592 		fdrop(fp, td);
593 		break;
594 
595 	case F_SETLK_REMOTE:
596 		error = priv_check(td, PRIV_NFS_LOCKD);
597 		if (error != 0)
598 			return (error);
599 		flg = F_REMOTE;
600 		goto do_setlk;
601 
602 	case F_SETLKW:
603 		flg |= F_WAIT;
604 		/* FALLTHROUGH F_SETLK */
605 
606 	case F_SETLK:
607 	do_setlk:
608 		flp = (struct flock *)arg;
609 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
610 			error = EINVAL;
611 			break;
612 		}
613 
614 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
615 		if (error != 0)
616 			break;
617 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
618 			error = EBADF;
619 			fdrop(fp, td);
620 			break;
621 		}
622 
623 		if (flp->l_whence == SEEK_CUR) {
624 			foffset = foffset_get(fp);
625 			if (foffset < 0 ||
626 			    (flp->l_start > 0 &&
627 			     foffset > OFF_MAX - flp->l_start)) {
628 				error = EOVERFLOW;
629 				fdrop(fp, td);
630 				break;
631 			}
632 			flp->l_start += foffset;
633 		}
634 
635 		vp = fp->f_vnode;
636 		switch (flp->l_type) {
637 		case F_RDLCK:
638 			if ((fp->f_flag & FREAD) == 0) {
639 				error = EBADF;
640 				break;
641 			}
642 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
643 				PROC_LOCK(p->p_leader);
644 				p->p_leader->p_flag |= P_ADVLOCK;
645 				PROC_UNLOCK(p->p_leader);
646 			}
647 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
648 			    flp, flg);
649 			break;
650 		case F_WRLCK:
651 			if ((fp->f_flag & FWRITE) == 0) {
652 				error = EBADF;
653 				break;
654 			}
655 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
656 				PROC_LOCK(p->p_leader);
657 				p->p_leader->p_flag |= P_ADVLOCK;
658 				PROC_UNLOCK(p->p_leader);
659 			}
660 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
661 			    flp, flg);
662 			break;
663 		case F_UNLCK:
664 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
665 			    flp, flg);
666 			break;
667 		case F_UNLCKSYS:
668 			if (flg != F_REMOTE) {
669 				error = EINVAL;
670 				break;
671 			}
672 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
673 			    F_UNLCKSYS, flp, flg);
674 			break;
675 		default:
676 			error = EINVAL;
677 			break;
678 		}
679 		if (error != 0 || flp->l_type == F_UNLCK ||
680 		    flp->l_type == F_UNLCKSYS) {
681 			fdrop(fp, td);
682 			break;
683 		}
684 
685 		/*
686 		 * Check for a race with close.
687 		 *
688 		 * The vnode is now advisory locked (or unlocked, but this case
689 		 * is not really important) as the caller requested.
690 		 * We had to drop the filedesc lock, so we need to recheck if
691 		 * the descriptor is still valid, because if it was closed
692 		 * in the meantime we need to remove advisory lock from the
693 		 * vnode - close on any descriptor leading to an advisory
694 		 * locked vnode, removes that lock.
695 		 * We will return 0 on purpose in that case, as the result of
696 		 * successful advisory lock might have been externally visible
697 		 * already. This is fine - effectively we pretend to the caller
698 		 * that the closing thread was a bit slower and that the
699 		 * advisory lock succeeded before the close.
700 		 */
701 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
702 		if (error != 0) {
703 			fdrop(fp, td);
704 			break;
705 		}
706 		if (fp != fp2) {
707 			flp->l_whence = SEEK_SET;
708 			flp->l_start = 0;
709 			flp->l_len = 0;
710 			flp->l_type = F_UNLCK;
711 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
712 			    F_UNLCK, flp, F_POSIX);
713 		}
714 		fdrop(fp, td);
715 		fdrop(fp2, td);
716 		break;
717 
718 	case F_GETLK:
719 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
720 		if (error != 0)
721 			break;
722 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
723 			error = EBADF;
724 			fdrop(fp, td);
725 			break;
726 		}
727 		flp = (struct flock *)arg;
728 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
729 		    flp->l_type != F_UNLCK) {
730 			error = EINVAL;
731 			fdrop(fp, td);
732 			break;
733 		}
734 		if (flp->l_whence == SEEK_CUR) {
735 			foffset = foffset_get(fp);
736 			if ((flp->l_start > 0 &&
737 			    foffset > OFF_MAX - flp->l_start) ||
738 			    (flp->l_start < 0 &&
739 			    foffset < OFF_MIN - flp->l_start)) {
740 				error = EOVERFLOW;
741 				fdrop(fp, td);
742 				break;
743 			}
744 			flp->l_start += foffset;
745 		}
746 		vp = fp->f_vnode;
747 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
748 		    F_POSIX);
749 		fdrop(fp, td);
750 		break;
751 
752 	case F_ADD_SEALS:
753 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
754 		if (error != 0)
755 			break;
756 		error = fo_add_seals(fp, arg);
757 		fdrop(fp, td);
758 		break;
759 
760 	case F_GET_SEALS:
761 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
762 		if (error != 0)
763 			break;
764 		if (fo_get_seals(fp, &seals) == 0)
765 			td->td_retval[0] = seals;
766 		else
767 			error = EINVAL;
768 		fdrop(fp, td);
769 		break;
770 
771 	case F_RDAHEAD:
772 		arg = arg ? 128 * 1024: 0;
773 		/* FALLTHROUGH */
774 	case F_READAHEAD:
775 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
776 		if (error != 0)
777 			break;
778 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
779 			fdrop(fp, td);
780 			error = EBADF;
781 			break;
782 		}
783 		vp = fp->f_vnode;
784 		if (vp->v_type != VREG) {
785 			fdrop(fp, td);
786 			error = ENOTTY;
787 			break;
788 		}
789 
790 		/*
791 		 * Exclusive lock synchronizes against f_seqcount reads and
792 		 * writes in sequential_heuristic().
793 		 */
794 		error = vn_lock(vp, LK_EXCLUSIVE);
795 		if (error != 0) {
796 			fdrop(fp, td);
797 			break;
798 		}
799 		if (arg >= 0) {
800 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
801 			arg = MIN(arg, INT_MAX - bsize + 1);
802 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
803 			    (arg + bsize - 1) / bsize);
804 			atomic_set_int(&fp->f_flag, FRDAHEAD);
805 		} else {
806 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
807 		}
808 		VOP_UNLOCK(vp);
809 		fdrop(fp, td);
810 		break;
811 
812 	case F_ISUNIONSTACK:
813 		/*
814 		 * Check if the vnode is part of a union stack (either the
815 		 * "union" flag from mount(2) or unionfs).
816 		 *
817 		 * Prior to introduction of this op libc's readdir would call
818 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
819 		 * data just to check fs name and a mount flag.
820 		 *
821 		 * Fixing the code to handle everything in the kernel instead
822 		 * is a non-trivial endeavor and has low priority, thus this
823 		 * horrible kludge facilitates the current behavior in a much
824 		 * cheaper manner until someone(tm) sorts this out.
825 		 */
826 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
827 		if (error != 0)
828 			break;
829 		if (fp->f_type != DTYPE_VNODE) {
830 			fdrop(fp, td);
831 			error = EBADF;
832 			break;
833 		}
834 		vp = fp->f_vnode;
835 		/*
836 		 * Since we don't prevent dooming the vnode even non-null mp
837 		 * found can become immediately stale. This is tolerable since
838 		 * mount points are type-stable (providing safe memory access)
839 		 * and any vfs op on this vnode going forward will return an
840 		 * error (meaning return value in this case is meaningless).
841 		 */
842 		mp = atomic_load_ptr(&vp->v_mount);
843 		if (__predict_false(mp == NULL)) {
844 			fdrop(fp, td);
845 			error = EBADF;
846 			break;
847 		}
848 		td->td_retval[0] = 0;
849 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
850 		    mp->mnt_flag & MNT_UNION)
851 			td->td_retval[0] = 1;
852 		fdrop(fp, td);
853 		break;
854 
855 	default:
856 		error = EINVAL;
857 		break;
858 	}
859 	return (error);
860 }
861 
862 static int
863 getmaxfd(struct thread *td)
864 {
865 
866 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
867 }
868 
869 /*
870  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
871  */
872 int
873 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
874 {
875 	struct filedesc *fdp;
876 	struct filedescent *oldfde, *newfde;
877 	struct proc *p;
878 	struct file *delfp, *oldfp;
879 	u_long *oioctls, *nioctls;
880 	int error, maxfd;
881 
882 	p = td->td_proc;
883 	fdp = p->p_fd;
884 	oioctls = NULL;
885 
886 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
887 	MPASS(mode < FDDUP_LASTMODE);
888 
889 	AUDIT_ARG_FD(old);
890 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
891 
892 	/*
893 	 * Verify we have a valid descriptor to dup from and possibly to
894 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
895 	 * return EINVAL when the new descriptor is out of bounds.
896 	 */
897 	if (old < 0)
898 		return (EBADF);
899 	if (new < 0)
900 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
901 	maxfd = getmaxfd(td);
902 	if (new >= maxfd)
903 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
904 
905 	error = EBADF;
906 	FILEDESC_XLOCK(fdp);
907 	if (fget_locked(fdp, old) == NULL)
908 		goto unlock;
909 	if (mode == FDDUP_FIXED && old == new) {
910 		td->td_retval[0] = new;
911 		if (flags & FDDUP_FLAG_CLOEXEC)
912 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
913 		error = 0;
914 		goto unlock;
915 	}
916 
917 	oldfde = &fdp->fd_ofiles[old];
918 	oldfp = oldfde->fde_file;
919 	if (!fhold(oldfp))
920 		goto unlock;
921 
922 	/*
923 	 * If the caller specified a file descriptor, make sure the file
924 	 * table is large enough to hold it, and grab it.  Otherwise, just
925 	 * allocate a new descriptor the usual way.
926 	 */
927 	switch (mode) {
928 	case FDDUP_NORMAL:
929 	case FDDUP_FCNTL:
930 		if ((error = fdalloc(td, new, &new)) != 0) {
931 			fdrop(oldfp, td);
932 			goto unlock;
933 		}
934 		break;
935 	case FDDUP_FIXED:
936 		if (new >= fdp->fd_nfiles) {
937 			/*
938 			 * The resource limits are here instead of e.g.
939 			 * fdalloc(), because the file descriptor table may be
940 			 * shared between processes, so we can't really use
941 			 * racct_add()/racct_sub().  Instead of counting the
942 			 * number of actually allocated descriptors, just put
943 			 * the limit on the size of the file descriptor table.
944 			 */
945 #ifdef RACCT
946 			if (RACCT_ENABLED()) {
947 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
948 				if (error != 0) {
949 					error = EMFILE;
950 					fdrop(oldfp, td);
951 					goto unlock;
952 				}
953 			}
954 #endif
955 			fdgrowtable_exp(fdp, new + 1);
956 		}
957 		if (!fdisused(fdp, new))
958 			fdused(fdp, new);
959 		break;
960 	default:
961 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
962 	}
963 
964 	KASSERT(old != new, ("new fd is same as old"));
965 
966 	/* Refetch oldfde because the table may have grown and old one freed. */
967 	oldfde = &fdp->fd_ofiles[old];
968 	KASSERT(oldfp == oldfde->fde_file,
969 	    ("fdt_ofiles shift from growth observed at fd %d",
970 	    old));
971 
972 	newfde = &fdp->fd_ofiles[new];
973 	delfp = newfde->fde_file;
974 
975 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
976 
977 	/*
978 	 * Duplicate the source descriptor.
979 	 */
980 #ifdef CAPABILITIES
981 	seqc_write_begin(&newfde->fde_seqc);
982 #endif
983 	oioctls = filecaps_free_prep(&newfde->fde_caps);
984 	memcpy(newfde, oldfde, fde_change_size);
985 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
986 	    nioctls);
987 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
988 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
989 	else
990 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
991 #ifdef CAPABILITIES
992 	seqc_write_end(&newfde->fde_seqc);
993 #endif
994 	td->td_retval[0] = new;
995 
996 	error = 0;
997 
998 	if (delfp != NULL) {
999 		(void) closefp(fdp, new, delfp, td, true, false);
1000 		FILEDESC_UNLOCK_ASSERT(fdp);
1001 	} else {
1002 unlock:
1003 		FILEDESC_XUNLOCK(fdp);
1004 	}
1005 
1006 	filecaps_free_finish(oioctls);
1007 	return (error);
1008 }
1009 
1010 static void
1011 sigiofree(struct sigio *sigio)
1012 {
1013 	crfree(sigio->sio_ucred);
1014 	free(sigio, M_SIGIO);
1015 }
1016 
1017 static struct sigio *
1018 funsetown_locked(struct sigio *sigio)
1019 {
1020 	struct proc *p;
1021 	struct pgrp *pg;
1022 
1023 	SIGIO_ASSERT_LOCKED();
1024 
1025 	if (sigio == NULL)
1026 		return (NULL);
1027 	*sigio->sio_myref = NULL;
1028 	if (sigio->sio_pgid < 0) {
1029 		pg = sigio->sio_pgrp;
1030 		PGRP_LOCK(pg);
1031 		SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1032 		PGRP_UNLOCK(pg);
1033 	} else {
1034 		p = sigio->sio_proc;
1035 		PROC_LOCK(p);
1036 		SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1037 		PROC_UNLOCK(p);
1038 	}
1039 	return (sigio);
1040 }
1041 
1042 /*
1043  * If sigio is on the list associated with a process or process group,
1044  * disable signalling from the device, remove sigio from the list and
1045  * free sigio.
1046  */
1047 void
1048 funsetown(struct sigio **sigiop)
1049 {
1050 	struct sigio *sigio;
1051 
1052 	/* Racy check, consumers must provide synchronization. */
1053 	if (*sigiop == NULL)
1054 		return;
1055 
1056 	SIGIO_LOCK();
1057 	sigio = funsetown_locked(*sigiop);
1058 	SIGIO_UNLOCK();
1059 	if (sigio != NULL)
1060 		sigiofree(sigio);
1061 }
1062 
1063 /*
1064  * Free a list of sigio structures.  The caller must ensure that new sigio
1065  * structures cannot be added after this point.  For process groups this is
1066  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1067  * as an interlock.
1068  */
1069 void
1070 funsetownlst(struct sigiolst *sigiolst)
1071 {
1072 	struct proc *p;
1073 	struct pgrp *pg;
1074 	struct sigio *sigio, *tmp;
1075 
1076 	/* Racy check. */
1077 	sigio = SLIST_FIRST(sigiolst);
1078 	if (sigio == NULL)
1079 		return;
1080 
1081 	p = NULL;
1082 	pg = NULL;
1083 
1084 	SIGIO_LOCK();
1085 	sigio = SLIST_FIRST(sigiolst);
1086 	if (sigio == NULL) {
1087 		SIGIO_UNLOCK();
1088 		return;
1089 	}
1090 
1091 	/*
1092 	 * Every entry of the list should belong to a single proc or pgrp.
1093 	 */
1094 	if (sigio->sio_pgid < 0) {
1095 		pg = sigio->sio_pgrp;
1096 		sx_assert(&proctree_lock, SX_XLOCKED);
1097 		PGRP_LOCK(pg);
1098 	} else /* if (sigio->sio_pgid > 0) */ {
1099 		p = sigio->sio_proc;
1100 		PROC_LOCK(p);
1101 		KASSERT((p->p_flag & P_WEXIT) != 0,
1102 		    ("%s: process %p is not exiting", __func__, p));
1103 	}
1104 
1105 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1106 		*sigio->sio_myref = NULL;
1107 		if (pg != NULL) {
1108 			KASSERT(sigio->sio_pgid < 0,
1109 			    ("Proc sigio in pgrp sigio list"));
1110 			KASSERT(sigio->sio_pgrp == pg,
1111 			    ("Bogus pgrp in sigio list"));
1112 		} else /* if (p != NULL) */ {
1113 			KASSERT(sigio->sio_pgid > 0,
1114 			    ("Pgrp sigio in proc sigio list"));
1115 			KASSERT(sigio->sio_proc == p,
1116 			    ("Bogus proc in sigio list"));
1117 		}
1118 	}
1119 
1120 	if (pg != NULL)
1121 		PGRP_UNLOCK(pg);
1122 	else
1123 		PROC_UNLOCK(p);
1124 	SIGIO_UNLOCK();
1125 
1126 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1127 		sigiofree(sigio);
1128 }
1129 
1130 /*
1131  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1132  *
1133  * After permission checking, add a sigio structure to the sigio list for
1134  * the process or process group.
1135  */
1136 int
1137 fsetown(pid_t pgid, struct sigio **sigiop)
1138 {
1139 	struct proc *proc;
1140 	struct pgrp *pgrp;
1141 	struct sigio *osigio, *sigio;
1142 	int ret;
1143 
1144 	if (pgid == 0) {
1145 		funsetown(sigiop);
1146 		return (0);
1147 	}
1148 
1149 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1150 	sigio->sio_pgid = pgid;
1151 	sigio->sio_ucred = crhold(curthread->td_ucred);
1152 	sigio->sio_myref = sigiop;
1153 
1154 	ret = 0;
1155 	if (pgid > 0) {
1156 		ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1157 		SIGIO_LOCK();
1158 		osigio = funsetown_locked(*sigiop);
1159 		if (ret == 0) {
1160 			PROC_LOCK(proc);
1161 			_PRELE(proc);
1162 			if ((proc->p_flag & P_WEXIT) != 0) {
1163 				ret = ESRCH;
1164 			} else if (proc->p_session !=
1165 			    curthread->td_proc->p_session) {
1166 				/*
1167 				 * Policy - Don't allow a process to FSETOWN a
1168 				 * process in another session.
1169 				 *
1170 				 * Remove this test to allow maximum flexibility
1171 				 * or restrict FSETOWN to the current process or
1172 				 * process group for maximum safety.
1173 				 */
1174 				ret = EPERM;
1175 			} else {
1176 				sigio->sio_proc = proc;
1177 				SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1178 				    sio_pgsigio);
1179 			}
1180 			PROC_UNLOCK(proc);
1181 		}
1182 	} else /* if (pgid < 0) */ {
1183 		sx_slock(&proctree_lock);
1184 		SIGIO_LOCK();
1185 		osigio = funsetown_locked(*sigiop);
1186 		pgrp = pgfind(-pgid);
1187 		if (pgrp == NULL) {
1188 			ret = ESRCH;
1189 		} else {
1190 			if (pgrp->pg_session != curthread->td_proc->p_session) {
1191 				/*
1192 				 * Policy - Don't allow a process to FSETOWN a
1193 				 * process in another session.
1194 				 *
1195 				 * Remove this test to allow maximum flexibility
1196 				 * or restrict FSETOWN to the current process or
1197 				 * process group for maximum safety.
1198 				 */
1199 				ret = EPERM;
1200 			} else {
1201 				sigio->sio_pgrp = pgrp;
1202 				SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1203 				    sio_pgsigio);
1204 			}
1205 			PGRP_UNLOCK(pgrp);
1206 		}
1207 		sx_sunlock(&proctree_lock);
1208 	}
1209 	if (ret == 0)
1210 		*sigiop = sigio;
1211 	SIGIO_UNLOCK();
1212 	if (osigio != NULL)
1213 		sigiofree(osigio);
1214 	return (ret);
1215 }
1216 
1217 /*
1218  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1219  */
1220 pid_t
1221 fgetown(struct sigio **sigiop)
1222 {
1223 	pid_t pgid;
1224 
1225 	SIGIO_LOCK();
1226 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1227 	SIGIO_UNLOCK();
1228 	return (pgid);
1229 }
1230 
1231 static int
1232 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1233     bool audit)
1234 {
1235 	int error;
1236 
1237 	FILEDESC_XLOCK_ASSERT(fdp);
1238 
1239 	/*
1240 	 * We now hold the fp reference that used to be owned by the
1241 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1242 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1243 	 * added, and deleteing a knote for the new fd.
1244 	 */
1245 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1246 		knote_fdclose(td, fd);
1247 
1248 	/*
1249 	 * We need to notify mqueue if the object is of type mqueue.
1250 	 */
1251 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1252 		mq_fdclose(td, fd, fp);
1253 	FILEDESC_XUNLOCK(fdp);
1254 
1255 #ifdef AUDIT
1256 	if (AUDITING_TD(td) && audit)
1257 		audit_sysclose(td, fd, fp);
1258 #endif
1259 	error = closef(fp, td);
1260 
1261 	/*
1262 	 * All paths leading up to closefp() will have already removed or
1263 	 * replaced the fd in the filedesc table, so a restart would not
1264 	 * operate on the same file.
1265 	 */
1266 	if (error == ERESTART)
1267 		error = EINTR;
1268 
1269 	return (error);
1270 }
1271 
1272 static int
1273 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1274     bool holdleaders, bool audit)
1275 {
1276 	int error;
1277 
1278 	FILEDESC_XLOCK_ASSERT(fdp);
1279 
1280 	if (holdleaders) {
1281 		if (td->td_proc->p_fdtol != NULL) {
1282 			/*
1283 			 * Ask fdfree() to sleep to ensure that all relevant
1284 			 * process leaders can be traversed in closef().
1285 			 */
1286 			fdp->fd_holdleaderscount++;
1287 		} else {
1288 			holdleaders = false;
1289 		}
1290 	}
1291 
1292 	error = closefp_impl(fdp, fd, fp, td, audit);
1293 	if (holdleaders) {
1294 		FILEDESC_XLOCK(fdp);
1295 		fdp->fd_holdleaderscount--;
1296 		if (fdp->fd_holdleaderscount == 0 &&
1297 		    fdp->fd_holdleaderswakeup != 0) {
1298 			fdp->fd_holdleaderswakeup = 0;
1299 			wakeup(&fdp->fd_holdleaderscount);
1300 		}
1301 		FILEDESC_XUNLOCK(fdp);
1302 	}
1303 	return (error);
1304 }
1305 
1306 static int
1307 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1308     bool holdleaders, bool audit)
1309 {
1310 
1311 	FILEDESC_XLOCK_ASSERT(fdp);
1312 
1313 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1314 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1315 	} else {
1316 		return (closefp_impl(fdp, fd, fp, td, audit));
1317 	}
1318 }
1319 
1320 /*
1321  * Close a file descriptor.
1322  */
1323 #ifndef _SYS_SYSPROTO_H_
1324 struct close_args {
1325 	int     fd;
1326 };
1327 #endif
1328 /* ARGSUSED */
1329 int
1330 sys_close(struct thread *td, struct close_args *uap)
1331 {
1332 
1333 	return (kern_close(td, uap->fd));
1334 }
1335 
1336 int
1337 kern_close(struct thread *td, int fd)
1338 {
1339 	struct filedesc *fdp;
1340 	struct file *fp;
1341 
1342 	fdp = td->td_proc->p_fd;
1343 
1344 	FILEDESC_XLOCK(fdp);
1345 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1346 		FILEDESC_XUNLOCK(fdp);
1347 		return (EBADF);
1348 	}
1349 	fdfree(fdp, fd);
1350 
1351 	/* closefp() drops the FILEDESC lock for us. */
1352 	return (closefp(fdp, fd, fp, td, true, true));
1353 }
1354 
1355 int
1356 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1357 {
1358 	struct filedesc *fdp;
1359 	const struct fdescenttbl *fdt;
1360 	struct file *fp;
1361 	int fd;
1362 
1363 	/*
1364 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1365 	 * open should not be a usage error.  From a close_range() perspective,
1366 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1367 	 * be a usage error as all fd above 3 are in-fact already closed.
1368 	 */
1369 	if (highfd < lowfd) {
1370 		return (EINVAL);
1371 	}
1372 
1373 	fdp = td->td_proc->p_fd;
1374 	FILEDESC_XLOCK(fdp);
1375 	fdt = atomic_load_ptr(&fdp->fd_files);
1376 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1377 	fd = lowfd;
1378 	if (__predict_false(fd > highfd)) {
1379 		goto out_locked;
1380 	}
1381 	for (;;) {
1382 		fp = fdt->fdt_ofiles[fd].fde_file;
1383 		if (fp == NULL) {
1384 			if (fd == highfd)
1385 				goto out_locked;
1386 		} else {
1387 			fdfree(fdp, fd);
1388 			(void) closefp(fdp, fd, fp, td, true, true);
1389 			if (fd == highfd)
1390 				goto out_unlocked;
1391 			FILEDESC_XLOCK(fdp);
1392 			fdt = atomic_load_ptr(&fdp->fd_files);
1393 		}
1394 		fd++;
1395 	}
1396 out_locked:
1397 	FILEDESC_XUNLOCK(fdp);
1398 out_unlocked:
1399 	return (0);
1400 }
1401 
1402 #ifndef _SYS_SYSPROTO_H_
1403 struct close_range_args {
1404 	u_int	lowfd;
1405 	u_int	highfd;
1406 	int	flags;
1407 };
1408 #endif
1409 int
1410 sys_close_range(struct thread *td, struct close_range_args *uap)
1411 {
1412 
1413 	AUDIT_ARG_FD(uap->lowfd);
1414 	AUDIT_ARG_CMD(uap->highfd);
1415 	AUDIT_ARG_FFLAGS(uap->flags);
1416 
1417 	/* No flags currently defined */
1418 	if (uap->flags != 0)
1419 		return (EINVAL);
1420 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1421 }
1422 
1423 #ifdef COMPAT_FREEBSD12
1424 /*
1425  * Close open file descriptors.
1426  */
1427 #ifndef _SYS_SYSPROTO_H_
1428 struct freebsd12_closefrom_args {
1429 	int	lowfd;
1430 };
1431 #endif
1432 /* ARGSUSED */
1433 int
1434 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1435 {
1436 	u_int lowfd;
1437 
1438 	AUDIT_ARG_FD(uap->lowfd);
1439 
1440 	/*
1441 	 * Treat negative starting file descriptor values identical to
1442 	 * closefrom(0) which closes all files.
1443 	 */
1444 	lowfd = MAX(0, uap->lowfd);
1445 	return (kern_close_range(td, lowfd, ~0U));
1446 }
1447 #endif	/* COMPAT_FREEBSD12 */
1448 
1449 #if defined(COMPAT_43)
1450 /*
1451  * Return status information about a file descriptor.
1452  */
1453 #ifndef _SYS_SYSPROTO_H_
1454 struct ofstat_args {
1455 	int	fd;
1456 	struct	ostat *sb;
1457 };
1458 #endif
1459 /* ARGSUSED */
1460 int
1461 ofstat(struct thread *td, struct ofstat_args *uap)
1462 {
1463 	struct ostat oub;
1464 	struct stat ub;
1465 	int error;
1466 
1467 	error = kern_fstat(td, uap->fd, &ub);
1468 	if (error == 0) {
1469 		cvtstat(&ub, &oub);
1470 		error = copyout(&oub, uap->sb, sizeof(oub));
1471 	}
1472 	return (error);
1473 }
1474 #endif /* COMPAT_43 */
1475 
1476 #if defined(COMPAT_FREEBSD11)
1477 int
1478 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1479 {
1480 	struct stat sb;
1481 	struct freebsd11_stat osb;
1482 	int error;
1483 
1484 	error = kern_fstat(td, uap->fd, &sb);
1485 	if (error != 0)
1486 		return (error);
1487 	error = freebsd11_cvtstat(&sb, &osb);
1488 	if (error == 0)
1489 		error = copyout(&osb, uap->sb, sizeof(osb));
1490 	return (error);
1491 }
1492 #endif	/* COMPAT_FREEBSD11 */
1493 
1494 /*
1495  * Return status information about a file descriptor.
1496  */
1497 #ifndef _SYS_SYSPROTO_H_
1498 struct fstat_args {
1499 	int	fd;
1500 	struct	stat *sb;
1501 };
1502 #endif
1503 /* ARGSUSED */
1504 int
1505 sys_fstat(struct thread *td, struct fstat_args *uap)
1506 {
1507 	struct stat ub;
1508 	int error;
1509 
1510 	error = kern_fstat(td, uap->fd, &ub);
1511 	if (error == 0)
1512 		error = copyout(&ub, uap->sb, sizeof(ub));
1513 	return (error);
1514 }
1515 
1516 int
1517 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1518 {
1519 	struct file *fp;
1520 	int error;
1521 
1522 	AUDIT_ARG_FD(fd);
1523 
1524 	error = fget(td, fd, &cap_fstat_rights, &fp);
1525 	if (__predict_false(error != 0))
1526 		return (error);
1527 
1528 	AUDIT_ARG_FILE(td->td_proc, fp);
1529 
1530 	error = fo_stat(fp, sbp, td->td_ucred);
1531 	fdrop(fp, td);
1532 #ifdef __STAT_TIME_T_EXT
1533 	sbp->st_atim_ext = 0;
1534 	sbp->st_mtim_ext = 0;
1535 	sbp->st_ctim_ext = 0;
1536 	sbp->st_btim_ext = 0;
1537 #endif
1538 #ifdef KTRACE
1539 	if (KTRPOINT(td, KTR_STRUCT))
1540 		ktrstat_error(sbp, error);
1541 #endif
1542 	return (error);
1543 }
1544 
1545 #if defined(COMPAT_FREEBSD11)
1546 /*
1547  * Return status information about a file descriptor.
1548  */
1549 #ifndef _SYS_SYSPROTO_H_
1550 struct freebsd11_nfstat_args {
1551 	int	fd;
1552 	struct	nstat *sb;
1553 };
1554 #endif
1555 /* ARGSUSED */
1556 int
1557 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1558 {
1559 	struct nstat nub;
1560 	struct stat ub;
1561 	int error;
1562 
1563 	error = kern_fstat(td, uap->fd, &ub);
1564 	if (error != 0)
1565 		return (error);
1566 	error = freebsd11_cvtnstat(&ub, &nub);
1567 	if (error != 0)
1568 		error = copyout(&nub, uap->sb, sizeof(nub));
1569 	return (error);
1570 }
1571 #endif /* COMPAT_FREEBSD11 */
1572 
1573 /*
1574  * Return pathconf information about a file descriptor.
1575  */
1576 #ifndef _SYS_SYSPROTO_H_
1577 struct fpathconf_args {
1578 	int	fd;
1579 	int	name;
1580 };
1581 #endif
1582 /* ARGSUSED */
1583 int
1584 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1585 {
1586 	long value;
1587 	int error;
1588 
1589 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1590 	if (error == 0)
1591 		td->td_retval[0] = value;
1592 	return (error);
1593 }
1594 
1595 int
1596 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1597 {
1598 	struct file *fp;
1599 	struct vnode *vp;
1600 	int error;
1601 
1602 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1603 	if (error != 0)
1604 		return (error);
1605 
1606 	if (name == _PC_ASYNC_IO) {
1607 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1608 		goto out;
1609 	}
1610 	vp = fp->f_vnode;
1611 	if (vp != NULL) {
1612 		vn_lock(vp, LK_SHARED | LK_RETRY);
1613 		error = VOP_PATHCONF(vp, name, valuep);
1614 		VOP_UNLOCK(vp);
1615 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1616 		if (name != _PC_PIPE_BUF) {
1617 			error = EINVAL;
1618 		} else {
1619 			*valuep = PIPE_BUF;
1620 			error = 0;
1621 		}
1622 	} else {
1623 		error = EOPNOTSUPP;
1624 	}
1625 out:
1626 	fdrop(fp, td);
1627 	return (error);
1628 }
1629 
1630 /*
1631  * Copy filecaps structure allocating memory for ioctls array if needed.
1632  *
1633  * The last parameter indicates whether the fdtable is locked. If it is not and
1634  * ioctls are encountered, copying fails and the caller must lock the table.
1635  *
1636  * Note that if the table was not locked, the caller has to check the relevant
1637  * sequence counter to determine whether the operation was successful.
1638  */
1639 bool
1640 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1641 {
1642 	size_t size;
1643 
1644 	if (src->fc_ioctls != NULL && !locked)
1645 		return (false);
1646 	memcpy(dst, src, sizeof(*src));
1647 	if (src->fc_ioctls == NULL)
1648 		return (true);
1649 
1650 	KASSERT(src->fc_nioctls > 0,
1651 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1652 
1653 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1654 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1655 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1656 	return (true);
1657 }
1658 
1659 static u_long *
1660 filecaps_copy_prep(const struct filecaps *src)
1661 {
1662 	u_long *ioctls;
1663 	size_t size;
1664 
1665 	if (__predict_true(src->fc_ioctls == NULL))
1666 		return (NULL);
1667 
1668 	KASSERT(src->fc_nioctls > 0,
1669 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1670 
1671 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1672 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1673 	return (ioctls);
1674 }
1675 
1676 static void
1677 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1678     u_long *ioctls)
1679 {
1680 	size_t size;
1681 
1682 	*dst = *src;
1683 	if (__predict_true(src->fc_ioctls == NULL)) {
1684 		MPASS(ioctls == NULL);
1685 		return;
1686 	}
1687 
1688 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1689 	dst->fc_ioctls = ioctls;
1690 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1691 }
1692 
1693 /*
1694  * Move filecaps structure to the new place and clear the old place.
1695  */
1696 void
1697 filecaps_move(struct filecaps *src, struct filecaps *dst)
1698 {
1699 
1700 	*dst = *src;
1701 	bzero(src, sizeof(*src));
1702 }
1703 
1704 /*
1705  * Fill the given filecaps structure with full rights.
1706  */
1707 static void
1708 filecaps_fill(struct filecaps *fcaps)
1709 {
1710 
1711 	CAP_ALL(&fcaps->fc_rights);
1712 	fcaps->fc_ioctls = NULL;
1713 	fcaps->fc_nioctls = -1;
1714 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1715 }
1716 
1717 /*
1718  * Free memory allocated within filecaps structure.
1719  */
1720 void
1721 filecaps_free(struct filecaps *fcaps)
1722 {
1723 
1724 	free(fcaps->fc_ioctls, M_FILECAPS);
1725 	bzero(fcaps, sizeof(*fcaps));
1726 }
1727 
1728 static u_long *
1729 filecaps_free_prep(struct filecaps *fcaps)
1730 {
1731 	u_long *ioctls;
1732 
1733 	ioctls = fcaps->fc_ioctls;
1734 	bzero(fcaps, sizeof(*fcaps));
1735 	return (ioctls);
1736 }
1737 
1738 static void
1739 filecaps_free_finish(u_long *ioctls)
1740 {
1741 
1742 	free(ioctls, M_FILECAPS);
1743 }
1744 
1745 /*
1746  * Validate the given filecaps structure.
1747  */
1748 static void
1749 filecaps_validate(const struct filecaps *fcaps, const char *func)
1750 {
1751 
1752 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1753 	    ("%s: invalid rights", func));
1754 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1755 	    ("%s: invalid fcntls", func));
1756 	KASSERT(fcaps->fc_fcntls == 0 ||
1757 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1758 	    ("%s: fcntls without CAP_FCNTL", func));
1759 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1760 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1761 	    ("%s: invalid ioctls", func));
1762 	KASSERT(fcaps->fc_nioctls == 0 ||
1763 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1764 	    ("%s: ioctls without CAP_IOCTL", func));
1765 }
1766 
1767 static void
1768 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1769 {
1770 	int nfd1;
1771 
1772 	FILEDESC_XLOCK_ASSERT(fdp);
1773 
1774 	nfd1 = fdp->fd_nfiles * 2;
1775 	if (nfd1 < nfd)
1776 		nfd1 = nfd;
1777 	fdgrowtable(fdp, nfd1);
1778 }
1779 
1780 /*
1781  * Grow the file table to accommodate (at least) nfd descriptors.
1782  */
1783 static void
1784 fdgrowtable(struct filedesc *fdp, int nfd)
1785 {
1786 	struct filedesc0 *fdp0;
1787 	struct freetable *ft;
1788 	struct fdescenttbl *ntable;
1789 	struct fdescenttbl *otable;
1790 	int nnfiles, onfiles;
1791 	NDSLOTTYPE *nmap, *omap;
1792 
1793 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1794 
1795 	/* save old values */
1796 	onfiles = fdp->fd_nfiles;
1797 	otable = fdp->fd_files;
1798 	omap = fdp->fd_map;
1799 
1800 	/* compute the size of the new table */
1801 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1802 	if (nnfiles <= onfiles)
1803 		/* the table is already large enough */
1804 		return;
1805 
1806 	/*
1807 	 * Allocate a new table.  We need enough space for the number of
1808 	 * entries, file entries themselves and the struct freetable we will use
1809 	 * when we decommission the table and place it on the freelist.
1810 	 * We place the struct freetable in the middle so we don't have
1811 	 * to worry about padding.
1812 	 */
1813 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1814 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1815 	    sizeof(struct freetable),
1816 	    M_FILEDESC, M_ZERO | M_WAITOK);
1817 	/* copy the old data */
1818 	ntable->fdt_nfiles = nnfiles;
1819 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1820 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1821 
1822 	/*
1823 	 * Allocate a new map only if the old is not large enough.  It will
1824 	 * grow at a slower rate than the table as it can map more
1825 	 * entries than the table can hold.
1826 	 */
1827 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1828 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1829 		    M_ZERO | M_WAITOK);
1830 		/* copy over the old data and update the pointer */
1831 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1832 		fdp->fd_map = nmap;
1833 	}
1834 
1835 	/*
1836 	 * Make sure that ntable is correctly initialized before we replace
1837 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1838 	 * data.
1839 	 */
1840 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1841 
1842 	/*
1843 	 * Free the old file table when not shared by other threads or processes.
1844 	 * The old file table is considered to be shared when either are true:
1845 	 * - The process has more than one thread.
1846 	 * - The file descriptor table has been shared via fdshare().
1847 	 *
1848 	 * When shared, the old file table will be placed on a freelist
1849 	 * which will be processed when the struct filedesc is released.
1850 	 *
1851 	 * Note that if onfiles == NDFILE, we're dealing with the original
1852 	 * static allocation contained within (struct filedesc0 *)fdp,
1853 	 * which must not be freed.
1854 	 */
1855 	if (onfiles > NDFILE) {
1856 		/*
1857 		 * Note we may be called here from fdinit while allocating a
1858 		 * table for a new process in which case ->p_fd points
1859 		 * elsewhere.
1860 		 */
1861 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1862 			free(otable, M_FILEDESC);
1863 		} else {
1864 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1865 			fdp0 = (struct filedesc0 *)fdp;
1866 			ft->ft_table = otable;
1867 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1868 		}
1869 	}
1870 	/*
1871 	 * The map does not have the same possibility of threads still
1872 	 * holding references to it.  So always free it as long as it
1873 	 * does not reference the original static allocation.
1874 	 */
1875 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1876 		free(omap, M_FILEDESC);
1877 }
1878 
1879 /*
1880  * Allocate a file descriptor for the process.
1881  */
1882 int
1883 fdalloc(struct thread *td, int minfd, int *result)
1884 {
1885 	struct proc *p = td->td_proc;
1886 	struct filedesc *fdp = p->p_fd;
1887 	int fd, maxfd, allocfd;
1888 #ifdef RACCT
1889 	int error;
1890 #endif
1891 
1892 	FILEDESC_XLOCK_ASSERT(fdp);
1893 
1894 	if (fdp->fd_freefile > minfd)
1895 		minfd = fdp->fd_freefile;
1896 
1897 	maxfd = getmaxfd(td);
1898 
1899 	/*
1900 	 * Search the bitmap for a free descriptor starting at minfd.
1901 	 * If none is found, grow the file table.
1902 	 */
1903 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1904 	if (__predict_false(fd >= maxfd))
1905 		return (EMFILE);
1906 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1907 		allocfd = min(fd * 2, maxfd);
1908 #ifdef RACCT
1909 		if (RACCT_ENABLED()) {
1910 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1911 			if (error != 0)
1912 				return (EMFILE);
1913 		}
1914 #endif
1915 		/*
1916 		 * fd is already equal to first free descriptor >= minfd, so
1917 		 * we only need to grow the table and we are done.
1918 		 */
1919 		fdgrowtable_exp(fdp, allocfd);
1920 	}
1921 
1922 	/*
1923 	 * Perform some sanity checks, then mark the file descriptor as
1924 	 * used and return it to the caller.
1925 	 */
1926 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1927 	    ("invalid descriptor %d", fd));
1928 	KASSERT(!fdisused(fdp, fd),
1929 	    ("fd_first_free() returned non-free descriptor"));
1930 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1931 	    ("file descriptor isn't free"));
1932 	fdused(fdp, fd);
1933 	*result = fd;
1934 	return (0);
1935 }
1936 
1937 /*
1938  * Allocate n file descriptors for the process.
1939  */
1940 int
1941 fdallocn(struct thread *td, int minfd, int *fds, int n)
1942 {
1943 	struct proc *p = td->td_proc;
1944 	struct filedesc *fdp = p->p_fd;
1945 	int i;
1946 
1947 	FILEDESC_XLOCK_ASSERT(fdp);
1948 
1949 	for (i = 0; i < n; i++)
1950 		if (fdalloc(td, 0, &fds[i]) != 0)
1951 			break;
1952 
1953 	if (i < n) {
1954 		for (i--; i >= 0; i--)
1955 			fdunused(fdp, fds[i]);
1956 		return (EMFILE);
1957 	}
1958 
1959 	return (0);
1960 }
1961 
1962 /*
1963  * Create a new open file structure and allocate a file descriptor for the
1964  * process that refers to it.  We add one reference to the file for the
1965  * descriptor table and one reference for resultfp. This is to prevent us
1966  * being preempted and the entry in the descriptor table closed after we
1967  * release the FILEDESC lock.
1968  */
1969 int
1970 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1971     struct filecaps *fcaps)
1972 {
1973 	struct file *fp;
1974 	int error, fd;
1975 
1976 	MPASS(resultfp != NULL);
1977 	MPASS(resultfd != NULL);
1978 
1979 	error = _falloc_noinstall(td, &fp, 2);
1980 	if (__predict_false(error != 0)) {
1981 		return (error);
1982 	}
1983 
1984 	error = finstall_refed(td, fp, &fd, flags, fcaps);
1985 	if (__predict_false(error != 0)) {
1986 		falloc_abort(td, fp);
1987 		return (error);
1988 	}
1989 
1990 	*resultfp = fp;
1991 	*resultfd = fd;
1992 
1993 	return (0);
1994 }
1995 
1996 /*
1997  * Create a new open file structure without allocating a file descriptor.
1998  */
1999 int
2000 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2001 {
2002 	struct file *fp;
2003 	int maxuserfiles = maxfiles - (maxfiles / 20);
2004 	int openfiles_new;
2005 	static struct timeval lastfail;
2006 	static int curfail;
2007 
2008 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2009 	MPASS(n > 0);
2010 
2011 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2012 	if ((openfiles_new >= maxuserfiles &&
2013 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2014 	    openfiles_new >= maxfiles) {
2015 		atomic_subtract_int(&openfiles, 1);
2016 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2017 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2018 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2019 		}
2020 		return (ENFILE);
2021 	}
2022 	fp = uma_zalloc(file_zone, M_WAITOK);
2023 	bzero(fp, sizeof(*fp));
2024 	refcount_init(&fp->f_count, n);
2025 	fp->f_cred = crhold(td->td_ucred);
2026 	fp->f_ops = &badfileops;
2027 	*resultfp = fp;
2028 	return (0);
2029 }
2030 
2031 void
2032 falloc_abort(struct thread *td, struct file *fp)
2033 {
2034 
2035 	/*
2036 	 * For assertion purposes.
2037 	 */
2038 	refcount_init(&fp->f_count, 0);
2039 	_fdrop(fp, td);
2040 }
2041 
2042 /*
2043  * Install a file in a file descriptor table.
2044  */
2045 void
2046 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2047     struct filecaps *fcaps)
2048 {
2049 	struct filedescent *fde;
2050 
2051 	MPASS(fp != NULL);
2052 	if (fcaps != NULL)
2053 		filecaps_validate(fcaps, __func__);
2054 	FILEDESC_XLOCK_ASSERT(fdp);
2055 
2056 	fde = &fdp->fd_ofiles[fd];
2057 #ifdef CAPABILITIES
2058 	seqc_write_begin(&fde->fde_seqc);
2059 #endif
2060 	fde->fde_file = fp;
2061 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2062 	if (fcaps != NULL)
2063 		filecaps_move(fcaps, &fde->fde_caps);
2064 	else
2065 		filecaps_fill(&fde->fde_caps);
2066 #ifdef CAPABILITIES
2067 	seqc_write_end(&fde->fde_seqc);
2068 #endif
2069 }
2070 
2071 int
2072 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2073     struct filecaps *fcaps)
2074 {
2075 	struct filedesc *fdp = td->td_proc->p_fd;
2076 	int error;
2077 
2078 	MPASS(fd != NULL);
2079 
2080 	FILEDESC_XLOCK(fdp);
2081 	error = fdalloc(td, 0, fd);
2082 	if (__predict_true(error == 0)) {
2083 		_finstall(fdp, fp, *fd, flags, fcaps);
2084 	}
2085 	FILEDESC_XUNLOCK(fdp);
2086 	return (error);
2087 }
2088 
2089 int
2090 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2091     struct filecaps *fcaps)
2092 {
2093 	int error;
2094 
2095 	MPASS(fd != NULL);
2096 
2097 	if (!fhold(fp))
2098 		return (EBADF);
2099 	error = finstall_refed(td, fp, fd, flags, fcaps);
2100 	if (__predict_false(error != 0)) {
2101 		fdrop(fp, td);
2102 	}
2103 	return (error);
2104 }
2105 
2106 /*
2107  * Build a new filedesc structure from another.
2108  *
2109  * If fdp is not NULL, return with it shared locked.
2110  */
2111 struct filedesc *
2112 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2113 {
2114 	struct filedesc0 *newfdp0;
2115 	struct filedesc *newfdp;
2116 
2117 	if (prepfiles)
2118 		MPASS(lastfile != NULL);
2119 	else
2120 		MPASS(lastfile == NULL);
2121 
2122 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2123 	newfdp = &newfdp0->fd_fd;
2124 
2125 	/* Create the file descriptor table. */
2126 	FILEDESC_LOCK_INIT(newfdp);
2127 	refcount_init(&newfdp->fd_refcnt, 1);
2128 	refcount_init(&newfdp->fd_holdcnt, 1);
2129 	newfdp->fd_map = newfdp0->fd_dmap;
2130 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2131 	newfdp->fd_files->fdt_nfiles = NDFILE;
2132 
2133 	if (fdp == NULL)
2134 		return (newfdp);
2135 
2136 	FILEDESC_SLOCK(fdp);
2137 	if (!prepfiles) {
2138 		FILEDESC_SUNLOCK(fdp);
2139 		return (newfdp);
2140 	}
2141 
2142 	for (;;) {
2143 		*lastfile = fdlastfile(fdp);
2144 		if (*lastfile < newfdp->fd_nfiles)
2145 			break;
2146 		FILEDESC_SUNLOCK(fdp);
2147 		fdgrowtable(newfdp, *lastfile + 1);
2148 		FILEDESC_SLOCK(fdp);
2149 	}
2150 
2151 	return (newfdp);
2152 }
2153 
2154 /*
2155  * Build a pwddesc structure from another.
2156  * Copy the current, root, and jail root vnode references.
2157  *
2158  * If pdp is not NULL, return with it shared locked.
2159  */
2160 struct pwddesc *
2161 pdinit(struct pwddesc *pdp, bool keeplock)
2162 {
2163 	struct pwddesc *newpdp;
2164 	struct pwd *newpwd;
2165 
2166 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2167 
2168 	PWDDESC_LOCK_INIT(newpdp);
2169 	refcount_init(&newpdp->pd_refcount, 1);
2170 	newpdp->pd_cmask = CMASK;
2171 
2172 	if (pdp == NULL) {
2173 		newpwd = pwd_alloc();
2174 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2175 		return (newpdp);
2176 	}
2177 
2178 	PWDDESC_XLOCK(pdp);
2179 	newpwd = pwd_hold_pwddesc(pdp);
2180 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2181 	if (!keeplock)
2182 		PWDDESC_XUNLOCK(pdp);
2183 	return (newpdp);
2184 }
2185 
2186 /*
2187  * Hold either filedesc or pwddesc of the passed process.
2188  *
2189  * The process lock is used to synchronize against the target exiting and
2190  * freeing the data.
2191  *
2192  * Clearing can be ilustrated in 3 steps:
2193  * 1. set the pointer to NULL. Either routine can race against it, hence
2194  *   atomic_load_ptr.
2195  * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2196  *   race to either still see the pointer or find NULL. It is still safe to
2197  *   grab a reference as clearing is stalled.
2198  * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2199  *   guaranteed to see NULL, making it safe to finish clearing
2200  */
2201 static struct filedesc *
2202 fdhold(struct proc *p)
2203 {
2204 	struct filedesc *fdp;
2205 
2206 	PROC_LOCK_ASSERT(p, MA_OWNED);
2207 	fdp = atomic_load_ptr(&p->p_fd);
2208 	if (fdp != NULL)
2209 		refcount_acquire(&fdp->fd_holdcnt);
2210 	return (fdp);
2211 }
2212 
2213 static struct pwddesc *
2214 pdhold(struct proc *p)
2215 {
2216 	struct pwddesc *pdp;
2217 
2218 	PROC_LOCK_ASSERT(p, MA_OWNED);
2219 	pdp = atomic_load_ptr(&p->p_pd);
2220 	if (pdp != NULL)
2221 		refcount_acquire(&pdp->pd_refcount);
2222 	return (pdp);
2223 }
2224 
2225 static void
2226 fddrop(struct filedesc *fdp)
2227 {
2228 
2229 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2230 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2231 			return;
2232 	}
2233 
2234 	FILEDESC_LOCK_DESTROY(fdp);
2235 	uma_zfree(filedesc0_zone, fdp);
2236 }
2237 
2238 static void
2239 pddrop(struct pwddesc *pdp)
2240 {
2241 	struct pwd *pwd;
2242 
2243 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2244 		return;
2245 
2246 	PWDDESC_XLOCK(pdp);
2247 	if (refcount_release(&pdp->pd_refcount) == 0) {
2248 		PWDDESC_XUNLOCK(pdp);
2249 		return;
2250 	}
2251 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2252 	pwd_set(pdp, NULL);
2253 	PWDDESC_XUNLOCK(pdp);
2254 	pwd_drop(pwd);
2255 
2256 	PWDDESC_LOCK_DESTROY(pdp);
2257 	free(pdp, M_PWDDESC);
2258 }
2259 
2260 /*
2261  * Share a filedesc structure.
2262  */
2263 struct filedesc *
2264 fdshare(struct filedesc *fdp)
2265 {
2266 
2267 	refcount_acquire(&fdp->fd_refcnt);
2268 	return (fdp);
2269 }
2270 
2271 /*
2272  * Share a pwddesc structure.
2273  */
2274 struct pwddesc *
2275 pdshare(struct pwddesc *pdp)
2276 {
2277 	refcount_acquire(&pdp->pd_refcount);
2278 	return (pdp);
2279 }
2280 
2281 /*
2282  * Unshare a filedesc structure, if necessary by making a copy
2283  */
2284 void
2285 fdunshare(struct thread *td)
2286 {
2287 	struct filedesc *tmp;
2288 	struct proc *p = td->td_proc;
2289 
2290 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2291 		return;
2292 
2293 	tmp = fdcopy(p->p_fd);
2294 	fdescfree(td);
2295 	p->p_fd = tmp;
2296 }
2297 
2298 /*
2299  * Unshare a pwddesc structure.
2300  */
2301 void
2302 pdunshare(struct thread *td)
2303 {
2304 	struct pwddesc *pdp;
2305 	struct proc *p;
2306 
2307 	p = td->td_proc;
2308 	/* Not shared. */
2309 	if (p->p_pd->pd_refcount == 1)
2310 		return;
2311 
2312 	pdp = pdcopy(p->p_pd);
2313 	pdescfree(td);
2314 	p->p_pd = pdp;
2315 }
2316 
2317 /*
2318  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2319  * this is to ease callers, not catch errors.
2320  */
2321 struct filedesc *
2322 fdcopy(struct filedesc *fdp)
2323 {
2324 	struct filedesc *newfdp;
2325 	struct filedescent *nfde, *ofde;
2326 	int i, lastfile;
2327 
2328 	MPASS(fdp != NULL);
2329 
2330 	newfdp = fdinit(fdp, true, &lastfile);
2331 	/* copy all passable descriptors (i.e. not kqueue) */
2332 	newfdp->fd_freefile = -1;
2333 	for (i = 0; i <= lastfile; ++i) {
2334 		ofde = &fdp->fd_ofiles[i];
2335 		if (ofde->fde_file == NULL ||
2336 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2337 		    !fhold(ofde->fde_file)) {
2338 			if (newfdp->fd_freefile == -1)
2339 				newfdp->fd_freefile = i;
2340 			continue;
2341 		}
2342 		nfde = &newfdp->fd_ofiles[i];
2343 		*nfde = *ofde;
2344 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2345 		fdused_init(newfdp, i);
2346 	}
2347 	if (newfdp->fd_freefile == -1)
2348 		newfdp->fd_freefile = i;
2349 	FILEDESC_SUNLOCK(fdp);
2350 	return (newfdp);
2351 }
2352 
2353 /*
2354  * Copy a pwddesc structure.
2355  */
2356 struct pwddesc *
2357 pdcopy(struct pwddesc *pdp)
2358 {
2359 	struct pwddesc *newpdp;
2360 
2361 	MPASS(pdp != NULL);
2362 
2363 	newpdp = pdinit(pdp, true);
2364 	newpdp->pd_cmask = pdp->pd_cmask;
2365 	PWDDESC_XUNLOCK(pdp);
2366 	return (newpdp);
2367 }
2368 
2369 /*
2370  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2371  * one of processes using it exits) and the table used to be shared.
2372  */
2373 static void
2374 fdclearlocks(struct thread *td)
2375 {
2376 	struct filedesc *fdp;
2377 	struct filedesc_to_leader *fdtol;
2378 	struct flock lf;
2379 	struct file *fp;
2380 	struct proc *p;
2381 	struct vnode *vp;
2382 	int i, lastfile;
2383 
2384 	p = td->td_proc;
2385 	fdp = p->p_fd;
2386 	fdtol = p->p_fdtol;
2387 	MPASS(fdtol != NULL);
2388 
2389 	FILEDESC_XLOCK(fdp);
2390 	KASSERT(fdtol->fdl_refcount > 0,
2391 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2392 	    fdtol->fdl_refcount));
2393 	if (fdtol->fdl_refcount == 1 &&
2394 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2395 		lastfile = fdlastfile(fdp);
2396 		for (i = 0; i <= lastfile; i++) {
2397 			fp = fdp->fd_ofiles[i].fde_file;
2398 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2399 			    !fhold(fp))
2400 				continue;
2401 			FILEDESC_XUNLOCK(fdp);
2402 			lf.l_whence = SEEK_SET;
2403 			lf.l_start = 0;
2404 			lf.l_len = 0;
2405 			lf.l_type = F_UNLCK;
2406 			vp = fp->f_vnode;
2407 			(void) VOP_ADVLOCK(vp,
2408 			    (caddr_t)p->p_leader, F_UNLCK,
2409 			    &lf, F_POSIX);
2410 			FILEDESC_XLOCK(fdp);
2411 			fdrop(fp, td);
2412 		}
2413 	}
2414 retry:
2415 	if (fdtol->fdl_refcount == 1) {
2416 		if (fdp->fd_holdleaderscount > 0 &&
2417 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2418 			/*
2419 			 * close() or kern_dup() has cleared a reference
2420 			 * in a shared file descriptor table.
2421 			 */
2422 			fdp->fd_holdleaderswakeup = 1;
2423 			sx_sleep(&fdp->fd_holdleaderscount,
2424 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2425 			goto retry;
2426 		}
2427 		if (fdtol->fdl_holdcount > 0) {
2428 			/*
2429 			 * Ensure that fdtol->fdl_leader remains
2430 			 * valid in closef().
2431 			 */
2432 			fdtol->fdl_wakeup = 1;
2433 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2434 			    "fdlhold", 0);
2435 			goto retry;
2436 		}
2437 	}
2438 	fdtol->fdl_refcount--;
2439 	if (fdtol->fdl_refcount == 0 &&
2440 	    fdtol->fdl_holdcount == 0) {
2441 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2442 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2443 	} else
2444 		fdtol = NULL;
2445 	p->p_fdtol = NULL;
2446 	FILEDESC_XUNLOCK(fdp);
2447 	if (fdtol != NULL)
2448 		free(fdtol, M_FILEDESC_TO_LEADER);
2449 }
2450 
2451 /*
2452  * Release a filedesc structure.
2453  */
2454 static void
2455 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2456 {
2457 	struct filedesc0 *fdp0;
2458 	struct freetable *ft, *tft;
2459 	struct filedescent *fde;
2460 	struct file *fp;
2461 	int i, lastfile;
2462 
2463 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2464 	    ("%s: fd table %p carries references", __func__, fdp));
2465 
2466 	/*
2467 	 * Serialize with threads iterating over the table, if any.
2468 	 */
2469 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2470 		FILEDESC_XLOCK(fdp);
2471 		FILEDESC_XUNLOCK(fdp);
2472 	}
2473 
2474 	lastfile = fdlastfile_single(fdp);
2475 	for (i = 0; i <= lastfile; i++) {
2476 		fde = &fdp->fd_ofiles[i];
2477 		fp = fde->fde_file;
2478 		if (fp != NULL) {
2479 			fdefree_last(fde);
2480 			(void) closef(fp, td);
2481 		}
2482 	}
2483 
2484 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2485 		free(fdp->fd_map, M_FILEDESC);
2486 	if (fdp->fd_nfiles > NDFILE)
2487 		free(fdp->fd_files, M_FILEDESC);
2488 
2489 	fdp0 = (struct filedesc0 *)fdp;
2490 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2491 		free(ft->ft_table, M_FILEDESC);
2492 
2493 	fddrop(fdp);
2494 }
2495 
2496 void
2497 fdescfree(struct thread *td)
2498 {
2499 	struct proc *p;
2500 	struct filedesc *fdp;
2501 
2502 	p = td->td_proc;
2503 	fdp = p->p_fd;
2504 	MPASS(fdp != NULL);
2505 
2506 #ifdef RACCT
2507 	if (RACCT_ENABLED())
2508 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2509 #endif
2510 
2511 	if (p->p_fdtol != NULL)
2512 		fdclearlocks(td);
2513 
2514 	/*
2515 	 * Check fdhold for an explanation.
2516 	 */
2517 	atomic_store_ptr(&p->p_fd, NULL);
2518 	atomic_thread_fence_seq_cst();
2519 	PROC_WAIT_UNLOCKED(p);
2520 
2521 	if (refcount_release(&fdp->fd_refcnt) == 0)
2522 		return;
2523 
2524 	fdescfree_fds(td, fdp);
2525 }
2526 
2527 void
2528 pdescfree(struct thread *td)
2529 {
2530 	struct proc *p;
2531 	struct pwddesc *pdp;
2532 
2533 	p = td->td_proc;
2534 	pdp = p->p_pd;
2535 	MPASS(pdp != NULL);
2536 
2537 	/*
2538 	 * Check pdhold for an explanation.
2539 	 */
2540 	atomic_store_ptr(&p->p_pd, NULL);
2541 	atomic_thread_fence_seq_cst();
2542 	PROC_WAIT_UNLOCKED(p);
2543 
2544 	pddrop(pdp);
2545 }
2546 
2547 /*
2548  * For setugid programs, we don't want to people to use that setugidness
2549  * to generate error messages which write to a file which otherwise would
2550  * otherwise be off-limits to the process.  We check for filesystems where
2551  * the vnode can change out from under us after execve (like [lin]procfs).
2552  *
2553  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2554  * sufficient.  We also don't check for setugidness since we know we are.
2555  */
2556 static bool
2557 is_unsafe(struct file *fp)
2558 {
2559 	struct vnode *vp;
2560 
2561 	if (fp->f_type != DTYPE_VNODE)
2562 		return (false);
2563 
2564 	vp = fp->f_vnode;
2565 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2566 }
2567 
2568 /*
2569  * Make this setguid thing safe, if at all possible.
2570  */
2571 void
2572 fdsetugidsafety(struct thread *td)
2573 {
2574 	struct filedesc *fdp;
2575 	struct file *fp;
2576 	int i;
2577 
2578 	fdp = td->td_proc->p_fd;
2579 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2580 	    ("the fdtable should not be shared"));
2581 	MPASS(fdp->fd_nfiles >= 3);
2582 	for (i = 0; i <= 2; i++) {
2583 		fp = fdp->fd_ofiles[i].fde_file;
2584 		if (fp != NULL && is_unsafe(fp)) {
2585 			FILEDESC_XLOCK(fdp);
2586 			knote_fdclose(td, i);
2587 			/*
2588 			 * NULL-out descriptor prior to close to avoid
2589 			 * a race while close blocks.
2590 			 */
2591 			fdfree(fdp, i);
2592 			FILEDESC_XUNLOCK(fdp);
2593 			(void) closef(fp, td);
2594 		}
2595 	}
2596 }
2597 
2598 /*
2599  * If a specific file object occupies a specific file descriptor, close the
2600  * file descriptor entry and drop a reference on the file object.  This is a
2601  * convenience function to handle a subsequent error in a function that calls
2602  * falloc() that handles the race that another thread might have closed the
2603  * file descriptor out from under the thread creating the file object.
2604  */
2605 void
2606 fdclose(struct thread *td, struct file *fp, int idx)
2607 {
2608 	struct filedesc *fdp = td->td_proc->p_fd;
2609 
2610 	FILEDESC_XLOCK(fdp);
2611 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2612 		fdfree(fdp, idx);
2613 		FILEDESC_XUNLOCK(fdp);
2614 		fdrop(fp, td);
2615 	} else
2616 		FILEDESC_XUNLOCK(fdp);
2617 }
2618 
2619 /*
2620  * Close any files on exec?
2621  */
2622 void
2623 fdcloseexec(struct thread *td)
2624 {
2625 	struct filedesc *fdp;
2626 	struct filedescent *fde;
2627 	struct file *fp;
2628 	int i, lastfile;
2629 
2630 	fdp = td->td_proc->p_fd;
2631 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2632 	    ("the fdtable should not be shared"));
2633 	lastfile = fdlastfile_single(fdp);
2634 	for (i = 0; i <= lastfile; i++) {
2635 		fde = &fdp->fd_ofiles[i];
2636 		fp = fde->fde_file;
2637 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2638 		    (fde->fde_flags & UF_EXCLOSE))) {
2639 			FILEDESC_XLOCK(fdp);
2640 			fdfree(fdp, i);
2641 			(void) closefp(fdp, i, fp, td, false, false);
2642 			FILEDESC_UNLOCK_ASSERT(fdp);
2643 		}
2644 	}
2645 }
2646 
2647 /*
2648  * It is unsafe for set[ug]id processes to be started with file
2649  * descriptors 0..2 closed, as these descriptors are given implicit
2650  * significance in the Standard C library.  fdcheckstd() will create a
2651  * descriptor referencing /dev/null for each of stdin, stdout, and
2652  * stderr that is not already open.
2653  */
2654 int
2655 fdcheckstd(struct thread *td)
2656 {
2657 	struct filedesc *fdp;
2658 	register_t save;
2659 	int i, error, devnull;
2660 
2661 	fdp = td->td_proc->p_fd;
2662 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2663 	    ("the fdtable should not be shared"));
2664 	MPASS(fdp->fd_nfiles >= 3);
2665 	devnull = -1;
2666 	for (i = 0; i <= 2; i++) {
2667 		if (fdp->fd_ofiles[i].fde_file != NULL)
2668 			continue;
2669 
2670 		save = td->td_retval[0];
2671 		if (devnull != -1) {
2672 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2673 		} else {
2674 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2675 			    UIO_SYSSPACE, O_RDWR, 0);
2676 			if (error == 0) {
2677 				devnull = td->td_retval[0];
2678 				KASSERT(devnull == i, ("we didn't get our fd"));
2679 			}
2680 		}
2681 		td->td_retval[0] = save;
2682 		if (error != 0)
2683 			return (error);
2684 	}
2685 	return (0);
2686 }
2687 
2688 /*
2689  * Internal form of close.  Decrement reference count on file structure.
2690  * Note: td may be NULL when closing a file that was being passed in a
2691  * message.
2692  */
2693 int
2694 closef(struct file *fp, struct thread *td)
2695 {
2696 	struct vnode *vp;
2697 	struct flock lf;
2698 	struct filedesc_to_leader *fdtol;
2699 	struct filedesc *fdp;
2700 
2701 	MPASS(td != NULL);
2702 
2703 	/*
2704 	 * POSIX record locking dictates that any close releases ALL
2705 	 * locks owned by this process.  This is handled by setting
2706 	 * a flag in the unlock to free ONLY locks obeying POSIX
2707 	 * semantics, and not to free BSD-style file locks.
2708 	 * If the descriptor was in a message, POSIX-style locks
2709 	 * aren't passed with the descriptor, and the thread pointer
2710 	 * will be NULL.  Callers should be careful only to pass a
2711 	 * NULL thread pointer when there really is no owning
2712 	 * context that might have locks, or the locks will be
2713 	 * leaked.
2714 	 */
2715 	if (fp->f_type == DTYPE_VNODE) {
2716 		vp = fp->f_vnode;
2717 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2718 			lf.l_whence = SEEK_SET;
2719 			lf.l_start = 0;
2720 			lf.l_len = 0;
2721 			lf.l_type = F_UNLCK;
2722 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2723 			    F_UNLCK, &lf, F_POSIX);
2724 		}
2725 		fdtol = td->td_proc->p_fdtol;
2726 		if (fdtol != NULL) {
2727 			/*
2728 			 * Handle special case where file descriptor table is
2729 			 * shared between multiple process leaders.
2730 			 */
2731 			fdp = td->td_proc->p_fd;
2732 			FILEDESC_XLOCK(fdp);
2733 			for (fdtol = fdtol->fdl_next;
2734 			    fdtol != td->td_proc->p_fdtol;
2735 			    fdtol = fdtol->fdl_next) {
2736 				if ((fdtol->fdl_leader->p_flag &
2737 				    P_ADVLOCK) == 0)
2738 					continue;
2739 				fdtol->fdl_holdcount++;
2740 				FILEDESC_XUNLOCK(fdp);
2741 				lf.l_whence = SEEK_SET;
2742 				lf.l_start = 0;
2743 				lf.l_len = 0;
2744 				lf.l_type = F_UNLCK;
2745 				vp = fp->f_vnode;
2746 				(void) VOP_ADVLOCK(vp,
2747 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2748 				    F_POSIX);
2749 				FILEDESC_XLOCK(fdp);
2750 				fdtol->fdl_holdcount--;
2751 				if (fdtol->fdl_holdcount == 0 &&
2752 				    fdtol->fdl_wakeup != 0) {
2753 					fdtol->fdl_wakeup = 0;
2754 					wakeup(fdtol);
2755 				}
2756 			}
2757 			FILEDESC_XUNLOCK(fdp);
2758 		}
2759 	}
2760 	return (fdrop_close(fp, td));
2761 }
2762 
2763 /*
2764  * Hack for file descriptor passing code.
2765  */
2766 void
2767 closef_nothread(struct file *fp)
2768 {
2769 
2770 	fdrop(fp, NULL);
2771 }
2772 
2773 /*
2774  * Initialize the file pointer with the specified properties.
2775  *
2776  * The ops are set with release semantics to be certain that the flags, type,
2777  * and data are visible when ops is.  This is to prevent ops methods from being
2778  * called with bad data.
2779  */
2780 void
2781 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2782 {
2783 	fp->f_data = data;
2784 	fp->f_flag = flag;
2785 	fp->f_type = type;
2786 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2787 }
2788 
2789 void
2790 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2791 {
2792 	fp->f_seqcount[UIO_READ] = 1;
2793 	fp->f_seqcount[UIO_WRITE] = 1;
2794 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2795 	    data, ops);
2796 }
2797 
2798 int
2799 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2800     struct file **fpp, struct filecaps *havecapsp)
2801 {
2802 	struct filedescent *fde;
2803 	int error;
2804 
2805 	FILEDESC_LOCK_ASSERT(fdp);
2806 
2807 	fde = fdeget_locked(fdp, fd);
2808 	if (fde == NULL) {
2809 		error = EBADF;
2810 		goto out;
2811 	}
2812 
2813 #ifdef CAPABILITIES
2814 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2815 	if (error != 0)
2816 		goto out;
2817 #endif
2818 
2819 	if (havecapsp != NULL)
2820 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2821 
2822 	*fpp = fde->fde_file;
2823 
2824 	error = 0;
2825 out:
2826 	return (error);
2827 }
2828 
2829 int
2830 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2831     struct file **fpp, struct filecaps *havecapsp)
2832 {
2833 	struct filedesc *fdp = td->td_proc->p_fd;
2834 	int error;
2835 #ifndef CAPABILITIES
2836 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2837 	if (havecapsp != NULL && error == 0)
2838 		filecaps_fill(havecapsp);
2839 #else
2840 	struct file *fp;
2841 	seqc_t seq;
2842 
2843 	*fpp = NULL;
2844 	for (;;) {
2845 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2846 		if (error != 0)
2847 			return (error);
2848 
2849 		if (havecapsp != NULL) {
2850 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2851 			    havecapsp, false)) {
2852 				fdrop(fp, td);
2853 				goto get_locked;
2854 			}
2855 		}
2856 
2857 		if (!fd_modified(fdp, fd, seq))
2858 			break;
2859 		fdrop(fp, td);
2860 	}
2861 
2862 	*fpp = fp;
2863 	return (0);
2864 
2865 get_locked:
2866 	FILEDESC_SLOCK(fdp);
2867 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2868 	if (error == 0 && !fhold(*fpp))
2869 		error = EBADF;
2870 	FILEDESC_SUNLOCK(fdp);
2871 #endif
2872 	return (error);
2873 }
2874 
2875 #ifdef CAPABILITIES
2876 int
2877 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2878 {
2879 	const struct filedescent *fde;
2880 	const struct fdescenttbl *fdt;
2881 	struct filedesc *fdp;
2882 	struct file *fp;
2883 	struct vnode *vp;
2884 	const cap_rights_t *haverights;
2885 	cap_rights_t rights;
2886 	seqc_t seq;
2887 
2888 	VFS_SMR_ASSERT_ENTERED();
2889 
2890 	rights = *ndp->ni_rightsneeded;
2891 	cap_rights_set_one(&rights, CAP_LOOKUP);
2892 
2893 	fdp = curproc->p_fd;
2894 	fdt = fdp->fd_files;
2895 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2896 		return (EBADF);
2897 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2898 	fde = &fdt->fdt_ofiles[fd];
2899 	haverights = cap_rights_fde_inline(fde);
2900 	fp = fde->fde_file;
2901 	if (__predict_false(fp == NULL))
2902 		return (EAGAIN);
2903 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2904 		return (EAGAIN);
2905 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2906 	vp = fp->f_vnode;
2907 	if (__predict_false(vp == NULL)) {
2908 		return (EAGAIN);
2909 	}
2910 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2911 		return (EAGAIN);
2912 	}
2913 	/*
2914 	 * Use an acquire barrier to force re-reading of fdt so it is
2915 	 * refreshed for verification.
2916 	 */
2917 	atomic_thread_fence_acq();
2918 	fdt = fdp->fd_files;
2919 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2920 		return (EAGAIN);
2921 	/*
2922 	 * If file descriptor doesn't have all rights,
2923 	 * all lookups relative to it must also be
2924 	 * strictly relative.
2925 	 *
2926 	 * Not yet supported by fast path.
2927 	 */
2928 	CAP_ALL(&rights);
2929 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2930 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2931 	    ndp->ni_filecaps.fc_nioctls != -1) {
2932 #ifdef notyet
2933 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2934 #else
2935 		return (EAGAIN);
2936 #endif
2937 	}
2938 	*vpp = vp;
2939 	return (0);
2940 }
2941 #else
2942 int
2943 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2944 {
2945 	const struct fdescenttbl *fdt;
2946 	struct filedesc *fdp;
2947 	struct file *fp;
2948 	struct vnode *vp;
2949 
2950 	VFS_SMR_ASSERT_ENTERED();
2951 
2952 	fdp = curproc->p_fd;
2953 	fdt = fdp->fd_files;
2954 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2955 		return (EBADF);
2956 	fp = fdt->fdt_ofiles[fd].fde_file;
2957 	if (__predict_false(fp == NULL))
2958 		return (EAGAIN);
2959 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2960 	vp = fp->f_vnode;
2961 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2962 		return (EAGAIN);
2963 	}
2964 	/*
2965 	 * Use an acquire barrier to force re-reading of fdt so it is
2966 	 * refreshed for verification.
2967 	 */
2968 	atomic_thread_fence_acq();
2969 	fdt = fdp->fd_files;
2970 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2971 		return (EAGAIN);
2972 	filecaps_fill(&ndp->ni_filecaps);
2973 	*vpp = vp;
2974 	return (0);
2975 }
2976 #endif
2977 
2978 int
2979 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2980     struct file **fpp, seqc_t *seqp)
2981 {
2982 #ifdef CAPABILITIES
2983 	const struct filedescent *fde;
2984 #endif
2985 	const struct fdescenttbl *fdt;
2986 	struct file *fp;
2987 #ifdef CAPABILITIES
2988 	seqc_t seq;
2989 	cap_rights_t haverights;
2990 	int error;
2991 #endif
2992 
2993 	fdt = fdp->fd_files;
2994 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2995 		return (EBADF);
2996 	/*
2997 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2998 	 * never raising a refcount above 0.  To accomplish this we have
2999 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3000 	 * must be re-verified once we acquire a reference to be certain
3001 	 * that the identity is still correct and we did not lose a race
3002 	 * due to preemption.
3003 	 */
3004 	for (;;) {
3005 #ifdef CAPABILITIES
3006 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3007 		fde = &fdt->fdt_ofiles[fd];
3008 		haverights = *cap_rights_fde_inline(fde);
3009 		fp = fde->fde_file;
3010 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3011 			continue;
3012 #else
3013 		fp = fdt->fdt_ofiles[fd].fde_file;
3014 #endif
3015 		if (fp == NULL)
3016 			return (EBADF);
3017 #ifdef CAPABILITIES
3018 		error = cap_check_inline(&haverights, needrightsp);
3019 		if (error != 0)
3020 			return (error);
3021 #endif
3022 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3023 			/*
3024 			 * Force a reload. Other thread could reallocate the
3025 			 * table before this fd was closed, so it is possible
3026 			 * that there is a stale fp pointer in cached version.
3027 			 */
3028 			fdt = atomic_load_ptr(&fdp->fd_files);
3029 			continue;
3030 		}
3031 		/*
3032 		 * Use an acquire barrier to force re-reading of fdt so it is
3033 		 * refreshed for verification.
3034 		 */
3035 		atomic_thread_fence_acq();
3036 		fdt = fdp->fd_files;
3037 #ifdef	CAPABILITIES
3038 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3039 #else
3040 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3041 #endif
3042 			break;
3043 		fdrop(fp, curthread);
3044 	}
3045 	*fpp = fp;
3046 	if (seqp != NULL) {
3047 #ifdef CAPABILITIES
3048 		*seqp = seq;
3049 #endif
3050 	}
3051 	return (0);
3052 }
3053 
3054 /*
3055  * See the comments in fget_unlocked_seq for an explanation of how this works.
3056  *
3057  * This is a simplified variant which bails out to the aforementioned routine
3058  * if anything goes wrong. In practice this only happens when userspace is
3059  * racing with itself.
3060  */
3061 int
3062 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3063     struct file **fpp)
3064 {
3065 #ifdef CAPABILITIES
3066 	const struct filedescent *fde;
3067 #endif
3068 	const struct fdescenttbl *fdt;
3069 	struct file *fp;
3070 #ifdef CAPABILITIES
3071 	seqc_t seq;
3072 	const cap_rights_t *haverights;
3073 #endif
3074 
3075 	fdt = fdp->fd_files;
3076 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3077 		return (EBADF);
3078 #ifdef CAPABILITIES
3079 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3080 	fde = &fdt->fdt_ofiles[fd];
3081 	haverights = cap_rights_fde_inline(fde);
3082 	fp = fde->fde_file;
3083 #else
3084 	fp = fdt->fdt_ofiles[fd].fde_file;
3085 #endif
3086 	if (__predict_false(fp == NULL))
3087 		goto out_fallback;
3088 #ifdef CAPABILITIES
3089 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3090 		goto out_fallback;
3091 #endif
3092 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3093 		goto out_fallback;
3094 
3095 	/*
3096 	 * Use an acquire barrier to force re-reading of fdt so it is
3097 	 * refreshed for verification.
3098 	 */
3099 	atomic_thread_fence_acq();
3100 	fdt = fdp->fd_files;
3101 #ifdef	CAPABILITIES
3102 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3103 #else
3104 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3105 #endif
3106 		goto out_fdrop;
3107 	*fpp = fp;
3108 	return (0);
3109 out_fdrop:
3110 	fdrop(fp, curthread);
3111 out_fallback:
3112 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3113 }
3114 
3115 /*
3116  * Translate fd -> file when the caller guarantees the file descriptor table
3117  * can't be changed by others.
3118  *
3119  * Note this does not mean the file object itself is only visible to the caller,
3120  * merely that it wont disappear without having to be referenced.
3121  *
3122  * Must be paired with fput_only_user.
3123  */
3124 #ifdef	CAPABILITIES
3125 int
3126 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3127     struct file **fpp)
3128 {
3129 	const struct filedescent *fde;
3130 	const struct fdescenttbl *fdt;
3131 	const cap_rights_t *haverights;
3132 	struct file *fp;
3133 	int error;
3134 
3135 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3136 
3137 	if (__predict_false(fd >= fdp->fd_nfiles))
3138 		return (EBADF);
3139 
3140 	fdt = fdp->fd_files;
3141 	fde = &fdt->fdt_ofiles[fd];
3142 	fp = fde->fde_file;
3143 	if (__predict_false(fp == NULL))
3144 		return (EBADF);
3145 	MPASS(refcount_load(&fp->f_count) > 0);
3146 	haverights = cap_rights_fde_inline(fde);
3147 	error = cap_check_inline(haverights, needrightsp);
3148 	if (__predict_false(error != 0))
3149 		return (error);
3150 	*fpp = fp;
3151 	return (0);
3152 }
3153 #else
3154 int
3155 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3156     struct file **fpp)
3157 {
3158 	struct file *fp;
3159 
3160 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3161 
3162 	if (__predict_false(fd >= fdp->fd_nfiles))
3163 		return (EBADF);
3164 
3165 	fp = fdp->fd_ofiles[fd].fde_file;
3166 	if (__predict_false(fp == NULL))
3167 		return (EBADF);
3168 
3169 	MPASS(refcount_load(&fp->f_count) > 0);
3170 	*fpp = fp;
3171 	return (0);
3172 }
3173 #endif
3174 
3175 /*
3176  * Extract the file pointer associated with the specified descriptor for the
3177  * current user process.
3178  *
3179  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3180  * returned.
3181  *
3182  * File's rights will be checked against the capability rights mask.
3183  *
3184  * If an error occurred the non-zero error is returned and *fpp is set to
3185  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3186  * responsible for fdrop().
3187  */
3188 static __inline int
3189 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3190     cap_rights_t *needrightsp)
3191 {
3192 	struct filedesc *fdp;
3193 	struct file *fp;
3194 	int error;
3195 
3196 	*fpp = NULL;
3197 	fdp = td->td_proc->p_fd;
3198 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3199 	if (__predict_false(error != 0))
3200 		return (error);
3201 	if (__predict_false(fp->f_ops == &badfileops)) {
3202 		fdrop(fp, td);
3203 		return (EBADF);
3204 	}
3205 
3206 	/*
3207 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3208 	 */
3209 	error = 0;
3210 	switch (flags) {
3211 	case FREAD:
3212 	case FWRITE:
3213 		if ((fp->f_flag & flags) == 0)
3214 			error = EBADF;
3215 		break;
3216 	case FEXEC:
3217 		if (fp->f_ops != &path_fileops &&
3218 		    ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3219 		    (fp->f_flag & FWRITE) != 0))
3220 			error = EBADF;
3221 		break;
3222 	case 0:
3223 		break;
3224 	default:
3225 		KASSERT(0, ("wrong flags"));
3226 	}
3227 
3228 	if (error != 0) {
3229 		fdrop(fp, td);
3230 		return (error);
3231 	}
3232 
3233 	*fpp = fp;
3234 	return (0);
3235 }
3236 
3237 int
3238 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3239 {
3240 
3241 	return (_fget(td, fd, fpp, 0, rightsp));
3242 }
3243 
3244 int
3245 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3246     struct file **fpp)
3247 {
3248 	int error;
3249 #ifndef CAPABILITIES
3250 	error = _fget(td, fd, fpp, 0, rightsp);
3251 	if (maxprotp != NULL)
3252 		*maxprotp = VM_PROT_ALL;
3253 	return (error);
3254 #else
3255 	cap_rights_t fdrights;
3256 	struct filedesc *fdp;
3257 	struct file *fp;
3258 	seqc_t seq;
3259 
3260 	*fpp = NULL;
3261 	fdp = td->td_proc->p_fd;
3262 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3263 	for (;;) {
3264 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3265 		if (__predict_false(error != 0))
3266 			return (error);
3267 		if (__predict_false(fp->f_ops == &badfileops)) {
3268 			fdrop(fp, td);
3269 			return (EBADF);
3270 		}
3271 		if (maxprotp != NULL)
3272 			fdrights = *cap_rights(fdp, fd);
3273 		if (!fd_modified(fdp, fd, seq))
3274 			break;
3275 		fdrop(fp, td);
3276 	}
3277 
3278 	/*
3279 	 * If requested, convert capability rights to access flags.
3280 	 */
3281 	if (maxprotp != NULL)
3282 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3283 	*fpp = fp;
3284 	return (0);
3285 #endif
3286 }
3287 
3288 int
3289 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3290 {
3291 
3292 	return (_fget(td, fd, fpp, FREAD, rightsp));
3293 }
3294 
3295 int
3296 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3297 {
3298 
3299 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3300 }
3301 
3302 int
3303 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3304     struct file **fpp)
3305 {
3306 	struct filedesc *fdp = td->td_proc->p_fd;
3307 #ifndef CAPABILITIES
3308 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3309 #else
3310 	struct file *fp;
3311 	int error;
3312 	seqc_t seq;
3313 
3314 	*fpp = NULL;
3315 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3316 	for (;;) {
3317 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3318 		if (error != 0)
3319 			return (error);
3320 		error = cap_fcntl_check(fdp, fd, needfcntl);
3321 		if (!fd_modified(fdp, fd, seq))
3322 			break;
3323 		fdrop(fp, td);
3324 	}
3325 	if (error != 0) {
3326 		fdrop(fp, td);
3327 		return (error);
3328 	}
3329 	*fpp = fp;
3330 	return (0);
3331 #endif
3332 }
3333 
3334 /*
3335  * Like fget() but loads the underlying vnode, or returns an error if the
3336  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3337  * never have VM objects.  The returned vnode will be vref()'d.
3338  *
3339  * XXX: what about the unused flags ?
3340  */
3341 static __inline int
3342 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3343     struct vnode **vpp)
3344 {
3345 	struct file *fp;
3346 	int error;
3347 
3348 	*vpp = NULL;
3349 	error = _fget(td, fd, &fp, flags, needrightsp);
3350 	if (error != 0)
3351 		return (error);
3352 	if (fp->f_vnode == NULL) {
3353 		error = EINVAL;
3354 	} else {
3355 		*vpp = fp->f_vnode;
3356 		vref(*vpp);
3357 	}
3358 	fdrop(fp, td);
3359 
3360 	return (error);
3361 }
3362 
3363 int
3364 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3365 {
3366 
3367 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3368 }
3369 
3370 int
3371 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3372     struct filecaps *havecaps, struct vnode **vpp)
3373 {
3374 	struct filecaps caps;
3375 	struct file *fp;
3376 	int error;
3377 
3378 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3379 	if (error != 0)
3380 		return (error);
3381 	if (fp->f_ops == &badfileops) {
3382 		error = EBADF;
3383 		goto out;
3384 	}
3385 	if (fp->f_vnode == NULL) {
3386 		error = EINVAL;
3387 		goto out;
3388 	}
3389 
3390 	*havecaps = caps;
3391 	*vpp = fp->f_vnode;
3392 	vref(*vpp);
3393 	fdrop(fp, td);
3394 
3395 	return (0);
3396 out:
3397 	filecaps_free(&caps);
3398 	fdrop(fp, td);
3399 	return (error);
3400 }
3401 
3402 int
3403 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3404 {
3405 
3406 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3407 }
3408 
3409 int
3410 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3411 {
3412 
3413 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3414 }
3415 
3416 #ifdef notyet
3417 int
3418 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3419     struct vnode **vpp)
3420 {
3421 
3422 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3423 }
3424 #endif
3425 
3426 /*
3427  * Handle the last reference to a file being closed.
3428  *
3429  * Without the noinline attribute clang keeps inlining the func thorough this
3430  * file when fdrop is used.
3431  */
3432 int __noinline
3433 _fdrop(struct file *fp, struct thread *td)
3434 {
3435 	int error;
3436 #ifdef INVARIANTS
3437 	int count;
3438 
3439 	count = refcount_load(&fp->f_count);
3440 	if (count != 0)
3441 		panic("fdrop: fp %p count %d", fp, count);
3442 #endif
3443 	error = fo_close(fp, td);
3444 	atomic_subtract_int(&openfiles, 1);
3445 	crfree(fp->f_cred);
3446 	free(fp->f_advice, M_FADVISE);
3447 	uma_zfree(file_zone, fp);
3448 
3449 	return (error);
3450 }
3451 
3452 /*
3453  * Apply an advisory lock on a file descriptor.
3454  *
3455  * Just attempt to get a record lock of the requested type on the entire file
3456  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3457  */
3458 #ifndef _SYS_SYSPROTO_H_
3459 struct flock_args {
3460 	int	fd;
3461 	int	how;
3462 };
3463 #endif
3464 /* ARGSUSED */
3465 int
3466 sys_flock(struct thread *td, struct flock_args *uap)
3467 {
3468 	struct file *fp;
3469 	struct vnode *vp;
3470 	struct flock lf;
3471 	int error;
3472 
3473 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3474 	if (error != 0)
3475 		return (error);
3476 	error = EOPNOTSUPP;
3477 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3478 		goto done;
3479 	}
3480 	if (fp->f_ops == &path_fileops) {
3481 		goto done;
3482 	}
3483 
3484 	error = 0;
3485 	vp = fp->f_vnode;
3486 	lf.l_whence = SEEK_SET;
3487 	lf.l_start = 0;
3488 	lf.l_len = 0;
3489 	if (uap->how & LOCK_UN) {
3490 		lf.l_type = F_UNLCK;
3491 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3492 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3493 		goto done;
3494 	}
3495 	if (uap->how & LOCK_EX)
3496 		lf.l_type = F_WRLCK;
3497 	else if (uap->how & LOCK_SH)
3498 		lf.l_type = F_RDLCK;
3499 	else {
3500 		error = EBADF;
3501 		goto done;
3502 	}
3503 	atomic_set_int(&fp->f_flag, FHASLOCK);
3504 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3505 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3506 done:
3507 	fdrop(fp, td);
3508 	return (error);
3509 }
3510 /*
3511  * Duplicate the specified descriptor to a free descriptor.
3512  */
3513 int
3514 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3515     int openerror, int *indxp)
3516 {
3517 	struct filedescent *newfde, *oldfde;
3518 	struct file *fp;
3519 	u_long *ioctls;
3520 	int error, indx;
3521 
3522 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3523 	    ("unexpected error %d in %s", openerror, __func__));
3524 
3525 	/*
3526 	 * If the to-be-dup'd fd number is greater than the allowed number
3527 	 * of file descriptors, or the fd to be dup'd has already been
3528 	 * closed, then reject.
3529 	 */
3530 	FILEDESC_XLOCK(fdp);
3531 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3532 		FILEDESC_XUNLOCK(fdp);
3533 		return (EBADF);
3534 	}
3535 
3536 	error = fdalloc(td, 0, &indx);
3537 	if (error != 0) {
3538 		FILEDESC_XUNLOCK(fdp);
3539 		return (error);
3540 	}
3541 
3542 	/*
3543 	 * There are two cases of interest here.
3544 	 *
3545 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3546 	 *
3547 	 * For ENXIO steal away the file structure from (dfd) and store it in
3548 	 * (indx).  (dfd) is effectively closed by this operation.
3549 	 */
3550 	switch (openerror) {
3551 	case ENODEV:
3552 		/*
3553 		 * Check that the mode the file is being opened for is a
3554 		 * subset of the mode of the existing descriptor.
3555 		 */
3556 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3557 			fdunused(fdp, indx);
3558 			FILEDESC_XUNLOCK(fdp);
3559 			return (EACCES);
3560 		}
3561 		if (!fhold(fp)) {
3562 			fdunused(fdp, indx);
3563 			FILEDESC_XUNLOCK(fdp);
3564 			return (EBADF);
3565 		}
3566 		newfde = &fdp->fd_ofiles[indx];
3567 		oldfde = &fdp->fd_ofiles[dfd];
3568 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3569 #ifdef CAPABILITIES
3570 		seqc_write_begin(&newfde->fde_seqc);
3571 #endif
3572 		memcpy(newfde, oldfde, fde_change_size);
3573 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3574 		    ioctls);
3575 #ifdef CAPABILITIES
3576 		seqc_write_end(&newfde->fde_seqc);
3577 #endif
3578 		break;
3579 	case ENXIO:
3580 		/*
3581 		 * Steal away the file pointer from dfd and stuff it into indx.
3582 		 */
3583 		newfde = &fdp->fd_ofiles[indx];
3584 		oldfde = &fdp->fd_ofiles[dfd];
3585 #ifdef CAPABILITIES
3586 		seqc_write_begin(&newfde->fde_seqc);
3587 #endif
3588 		memcpy(newfde, oldfde, fde_change_size);
3589 		oldfde->fde_file = NULL;
3590 		fdunused(fdp, dfd);
3591 #ifdef CAPABILITIES
3592 		seqc_write_end(&newfde->fde_seqc);
3593 #endif
3594 		break;
3595 	}
3596 	FILEDESC_XUNLOCK(fdp);
3597 	*indxp = indx;
3598 	return (0);
3599 }
3600 
3601 /*
3602  * This sysctl determines if we will allow a process to chroot(2) if it
3603  * has a directory open:
3604  *	0: disallowed for all processes.
3605  *	1: allowed for processes that were not already chroot(2)'ed.
3606  *	2: allowed for all processes.
3607  */
3608 
3609 static int chroot_allow_open_directories = 1;
3610 
3611 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3612     &chroot_allow_open_directories, 0,
3613     "Allow a process to chroot(2) if it has a directory open");
3614 
3615 /*
3616  * Helper function for raised chroot(2) security function:  Refuse if
3617  * any filedescriptors are open directories.
3618  */
3619 static int
3620 chroot_refuse_vdir_fds(struct filedesc *fdp)
3621 {
3622 	struct vnode *vp;
3623 	struct file *fp;
3624 	int fd, lastfile;
3625 
3626 	FILEDESC_LOCK_ASSERT(fdp);
3627 
3628 	lastfile = fdlastfile(fdp);
3629 	for (fd = 0; fd <= lastfile; fd++) {
3630 		fp = fget_locked(fdp, fd);
3631 		if (fp == NULL)
3632 			continue;
3633 		if (fp->f_type == DTYPE_VNODE) {
3634 			vp = fp->f_vnode;
3635 			if (vp->v_type == VDIR)
3636 				return (EPERM);
3637 		}
3638 	}
3639 	return (0);
3640 }
3641 
3642 static void
3643 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3644 {
3645 
3646 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3647 		vrefact(oldpwd->pwd_cdir);
3648 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3649 	}
3650 
3651 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3652 		vrefact(oldpwd->pwd_rdir);
3653 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3654 	}
3655 
3656 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3657 		vrefact(oldpwd->pwd_jdir);
3658 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3659 	}
3660 }
3661 
3662 struct pwd *
3663 pwd_hold_pwddesc(struct pwddesc *pdp)
3664 {
3665 	struct pwd *pwd;
3666 
3667 	PWDDESC_ASSERT_XLOCKED(pdp);
3668 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3669 	if (pwd != NULL)
3670 		refcount_acquire(&pwd->pwd_refcount);
3671 	return (pwd);
3672 }
3673 
3674 bool
3675 pwd_hold_smr(struct pwd *pwd)
3676 {
3677 
3678 	MPASS(pwd != NULL);
3679 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3680 		return (true);
3681 	}
3682 	return (false);
3683 }
3684 
3685 struct pwd *
3686 pwd_hold(struct thread *td)
3687 {
3688 	struct pwddesc *pdp;
3689 	struct pwd *pwd;
3690 
3691 	pdp = td->td_proc->p_pd;
3692 
3693 	vfs_smr_enter();
3694 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3695 	if (pwd_hold_smr(pwd)) {
3696 		vfs_smr_exit();
3697 		return (pwd);
3698 	}
3699 	vfs_smr_exit();
3700 	PWDDESC_XLOCK(pdp);
3701 	pwd = pwd_hold_pwddesc(pdp);
3702 	MPASS(pwd != NULL);
3703 	PWDDESC_XUNLOCK(pdp);
3704 	return (pwd);
3705 }
3706 
3707 struct pwd *
3708 pwd_hold_proc(struct proc *p)
3709 {
3710 	struct pwddesc *pdp;
3711 	struct pwd *pwd;
3712 
3713 	PROC_ASSERT_HELD(p);
3714 	PROC_LOCK(p);
3715 	pdp = pdhold(p);
3716 	MPASS(pdp != NULL);
3717 	PROC_UNLOCK(p);
3718 
3719 	PWDDESC_XLOCK(pdp);
3720 	pwd = pwd_hold_pwddesc(pdp);
3721 	MPASS(pwd != NULL);
3722 	PWDDESC_XUNLOCK(pdp);
3723 	pddrop(pdp);
3724 	return (pwd);
3725 }
3726 
3727 static struct pwd *
3728 pwd_alloc(void)
3729 {
3730 	struct pwd *pwd;
3731 
3732 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3733 	bzero(pwd, sizeof(*pwd));
3734 	refcount_init(&pwd->pwd_refcount, 1);
3735 	return (pwd);
3736 }
3737 
3738 void
3739 pwd_drop(struct pwd *pwd)
3740 {
3741 
3742 	if (!refcount_release(&pwd->pwd_refcount))
3743 		return;
3744 
3745 	if (pwd->pwd_cdir != NULL)
3746 		vrele(pwd->pwd_cdir);
3747 	if (pwd->pwd_rdir != NULL)
3748 		vrele(pwd->pwd_rdir);
3749 	if (pwd->pwd_jdir != NULL)
3750 		vrele(pwd->pwd_jdir);
3751 	uma_zfree_smr(pwd_zone, pwd);
3752 }
3753 
3754 /*
3755 * The caller is responsible for invoking priv_check() and
3756 * mac_vnode_check_chroot() to authorize this operation.
3757 */
3758 int
3759 pwd_chroot(struct thread *td, struct vnode *vp)
3760 {
3761 	struct pwddesc *pdp;
3762 	struct filedesc *fdp;
3763 	struct pwd *newpwd, *oldpwd;
3764 	int error;
3765 
3766 	fdp = td->td_proc->p_fd;
3767 	pdp = td->td_proc->p_pd;
3768 	newpwd = pwd_alloc();
3769 	FILEDESC_SLOCK(fdp);
3770 	PWDDESC_XLOCK(pdp);
3771 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3772 	if (chroot_allow_open_directories == 0 ||
3773 	    (chroot_allow_open_directories == 1 &&
3774 	    oldpwd->pwd_rdir != rootvnode)) {
3775 		error = chroot_refuse_vdir_fds(fdp);
3776 		FILEDESC_SUNLOCK(fdp);
3777 		if (error != 0) {
3778 			PWDDESC_XUNLOCK(pdp);
3779 			pwd_drop(newpwd);
3780 			return (error);
3781 		}
3782 	} else {
3783 		FILEDESC_SUNLOCK(fdp);
3784 	}
3785 
3786 	vrefact(vp);
3787 	newpwd->pwd_rdir = vp;
3788 	if (oldpwd->pwd_jdir == NULL) {
3789 		vrefact(vp);
3790 		newpwd->pwd_jdir = vp;
3791 	}
3792 	pwd_fill(oldpwd, newpwd);
3793 	pwd_set(pdp, newpwd);
3794 	PWDDESC_XUNLOCK(pdp);
3795 	pwd_drop(oldpwd);
3796 	return (0);
3797 }
3798 
3799 void
3800 pwd_chdir(struct thread *td, struct vnode *vp)
3801 {
3802 	struct pwddesc *pdp;
3803 	struct pwd *newpwd, *oldpwd;
3804 
3805 	VNPASS(vp->v_usecount > 0, vp);
3806 
3807 	newpwd = pwd_alloc();
3808 	pdp = td->td_proc->p_pd;
3809 	PWDDESC_XLOCK(pdp);
3810 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3811 	newpwd->pwd_cdir = vp;
3812 	pwd_fill(oldpwd, newpwd);
3813 	pwd_set(pdp, newpwd);
3814 	PWDDESC_XUNLOCK(pdp);
3815 	pwd_drop(oldpwd);
3816 }
3817 
3818 /*
3819  * jail_attach(2) changes both root and working directories.
3820  */
3821 int
3822 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3823 {
3824 	struct pwddesc *pdp;
3825 	struct filedesc *fdp;
3826 	struct pwd *newpwd, *oldpwd;
3827 	int error;
3828 
3829 	fdp = td->td_proc->p_fd;
3830 	pdp = td->td_proc->p_pd;
3831 	newpwd = pwd_alloc();
3832 	FILEDESC_SLOCK(fdp);
3833 	PWDDESC_XLOCK(pdp);
3834 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3835 	error = chroot_refuse_vdir_fds(fdp);
3836 	FILEDESC_SUNLOCK(fdp);
3837 	if (error != 0) {
3838 		PWDDESC_XUNLOCK(pdp);
3839 		pwd_drop(newpwd);
3840 		return (error);
3841 	}
3842 
3843 	vrefact(vp);
3844 	newpwd->pwd_rdir = vp;
3845 	vrefact(vp);
3846 	newpwd->pwd_cdir = vp;
3847 	if (oldpwd->pwd_jdir == NULL) {
3848 		vrefact(vp);
3849 		newpwd->pwd_jdir = vp;
3850 	}
3851 	pwd_fill(oldpwd, newpwd);
3852 	pwd_set(pdp, newpwd);
3853 	PWDDESC_XUNLOCK(pdp);
3854 	pwd_drop(oldpwd);
3855 	return (0);
3856 }
3857 
3858 void
3859 pwd_ensure_dirs(void)
3860 {
3861 	struct pwddesc *pdp;
3862 	struct pwd *oldpwd, *newpwd;
3863 
3864 	pdp = curproc->p_pd;
3865 	PWDDESC_XLOCK(pdp);
3866 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3867 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3868 		PWDDESC_XUNLOCK(pdp);
3869 		return;
3870 	}
3871 	PWDDESC_XUNLOCK(pdp);
3872 
3873 	newpwd = pwd_alloc();
3874 	PWDDESC_XLOCK(pdp);
3875 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3876 	pwd_fill(oldpwd, newpwd);
3877 	if (newpwd->pwd_cdir == NULL) {
3878 		vrefact(rootvnode);
3879 		newpwd->pwd_cdir = rootvnode;
3880 	}
3881 	if (newpwd->pwd_rdir == NULL) {
3882 		vrefact(rootvnode);
3883 		newpwd->pwd_rdir = rootvnode;
3884 	}
3885 	pwd_set(pdp, newpwd);
3886 	PWDDESC_XUNLOCK(pdp);
3887 	pwd_drop(oldpwd);
3888 }
3889 
3890 void
3891 pwd_set_rootvnode(void)
3892 {
3893 	struct pwddesc *pdp;
3894 	struct pwd *oldpwd, *newpwd;
3895 
3896 	pdp = curproc->p_pd;
3897 
3898 	newpwd = pwd_alloc();
3899 	PWDDESC_XLOCK(pdp);
3900 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3901 	vrefact(rootvnode);
3902 	newpwd->pwd_cdir = rootvnode;
3903 	vrefact(rootvnode);
3904 	newpwd->pwd_rdir = rootvnode;
3905 	pwd_fill(oldpwd, newpwd);
3906 	pwd_set(pdp, newpwd);
3907 	PWDDESC_XUNLOCK(pdp);
3908 	pwd_drop(oldpwd);
3909 }
3910 
3911 /*
3912  * Scan all active processes and prisons to see if any of them have a current
3913  * or root directory of `olddp'. If so, replace them with the new mount point.
3914  */
3915 void
3916 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3917 {
3918 	struct pwddesc *pdp;
3919 	struct pwd *newpwd, *oldpwd;
3920 	struct prison *pr;
3921 	struct proc *p;
3922 	int nrele;
3923 
3924 	if (vrefcnt(olddp) == 1)
3925 		return;
3926 	nrele = 0;
3927 	newpwd = pwd_alloc();
3928 	sx_slock(&allproc_lock);
3929 	FOREACH_PROC_IN_SYSTEM(p) {
3930 		PROC_LOCK(p);
3931 		pdp = pdhold(p);
3932 		PROC_UNLOCK(p);
3933 		if (pdp == NULL)
3934 			continue;
3935 		PWDDESC_XLOCK(pdp);
3936 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3937 		if (oldpwd == NULL ||
3938 		    (oldpwd->pwd_cdir != olddp &&
3939 		    oldpwd->pwd_rdir != olddp &&
3940 		    oldpwd->pwd_jdir != olddp)) {
3941 			PWDDESC_XUNLOCK(pdp);
3942 			pddrop(pdp);
3943 			continue;
3944 		}
3945 		if (oldpwd->pwd_cdir == olddp) {
3946 			vrefact(newdp);
3947 			newpwd->pwd_cdir = newdp;
3948 		}
3949 		if (oldpwd->pwd_rdir == olddp) {
3950 			vrefact(newdp);
3951 			newpwd->pwd_rdir = newdp;
3952 		}
3953 		if (oldpwd->pwd_jdir == olddp) {
3954 			vrefact(newdp);
3955 			newpwd->pwd_jdir = newdp;
3956 		}
3957 		pwd_fill(oldpwd, newpwd);
3958 		pwd_set(pdp, newpwd);
3959 		PWDDESC_XUNLOCK(pdp);
3960 		pwd_drop(oldpwd);
3961 		pddrop(pdp);
3962 		newpwd = pwd_alloc();
3963 	}
3964 	sx_sunlock(&allproc_lock);
3965 	pwd_drop(newpwd);
3966 	if (rootvnode == olddp) {
3967 		vrefact(newdp);
3968 		rootvnode = newdp;
3969 		nrele++;
3970 	}
3971 	mtx_lock(&prison0.pr_mtx);
3972 	if (prison0.pr_root == olddp) {
3973 		vrefact(newdp);
3974 		prison0.pr_root = newdp;
3975 		nrele++;
3976 	}
3977 	mtx_unlock(&prison0.pr_mtx);
3978 	sx_slock(&allprison_lock);
3979 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3980 		mtx_lock(&pr->pr_mtx);
3981 		if (pr->pr_root == olddp) {
3982 			vrefact(newdp);
3983 			pr->pr_root = newdp;
3984 			nrele++;
3985 		}
3986 		mtx_unlock(&pr->pr_mtx);
3987 	}
3988 	sx_sunlock(&allprison_lock);
3989 	while (nrele--)
3990 		vrele(olddp);
3991 }
3992 
3993 struct filedesc_to_leader *
3994 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3995 {
3996 	struct filedesc_to_leader *fdtol;
3997 
3998 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3999 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4000 	fdtol->fdl_refcount = 1;
4001 	fdtol->fdl_holdcount = 0;
4002 	fdtol->fdl_wakeup = 0;
4003 	fdtol->fdl_leader = leader;
4004 	if (old != NULL) {
4005 		FILEDESC_XLOCK(fdp);
4006 		fdtol->fdl_next = old->fdl_next;
4007 		fdtol->fdl_prev = old;
4008 		old->fdl_next = fdtol;
4009 		fdtol->fdl_next->fdl_prev = fdtol;
4010 		FILEDESC_XUNLOCK(fdp);
4011 	} else {
4012 		fdtol->fdl_next = fdtol;
4013 		fdtol->fdl_prev = fdtol;
4014 	}
4015 	return (fdtol);
4016 }
4017 
4018 static int
4019 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4020 {
4021 	NDSLOTTYPE *map;
4022 	struct filedesc *fdp;
4023 	u_int namelen;
4024 	int count, off, minoff;
4025 
4026 	namelen = arg2;
4027 	if (namelen != 1)
4028 		return (EINVAL);
4029 
4030 	if (*(int *)arg1 != 0)
4031 		return (EINVAL);
4032 
4033 	fdp = curproc->p_fd;
4034 	count = 0;
4035 	FILEDESC_SLOCK(fdp);
4036 	map = fdp->fd_map;
4037 	off = NDSLOT(fdp->fd_nfiles - 1);
4038 	for (minoff = NDSLOT(0); off >= minoff; --off)
4039 		count += bitcountl(map[off]);
4040 	FILEDESC_SUNLOCK(fdp);
4041 
4042 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4043 }
4044 
4045 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4046     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4047     "Number of open file descriptors");
4048 
4049 /*
4050  * Get file structures globally.
4051  */
4052 static int
4053 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4054 {
4055 	struct xfile xf;
4056 	struct filedesc *fdp;
4057 	struct file *fp;
4058 	struct proc *p;
4059 	int error, n, lastfile;
4060 
4061 	error = sysctl_wire_old_buffer(req, 0);
4062 	if (error != 0)
4063 		return (error);
4064 	if (req->oldptr == NULL) {
4065 		n = 0;
4066 		sx_slock(&allproc_lock);
4067 		FOREACH_PROC_IN_SYSTEM(p) {
4068 			PROC_LOCK(p);
4069 			if (p->p_state == PRS_NEW) {
4070 				PROC_UNLOCK(p);
4071 				continue;
4072 			}
4073 			fdp = fdhold(p);
4074 			PROC_UNLOCK(p);
4075 			if (fdp == NULL)
4076 				continue;
4077 			/* overestimates sparse tables. */
4078 			n += fdp->fd_nfiles;
4079 			fddrop(fdp);
4080 		}
4081 		sx_sunlock(&allproc_lock);
4082 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4083 	}
4084 	error = 0;
4085 	bzero(&xf, sizeof(xf));
4086 	xf.xf_size = sizeof(xf);
4087 	sx_slock(&allproc_lock);
4088 	FOREACH_PROC_IN_SYSTEM(p) {
4089 		PROC_LOCK(p);
4090 		if (p->p_state == PRS_NEW) {
4091 			PROC_UNLOCK(p);
4092 			continue;
4093 		}
4094 		if (p_cansee(req->td, p) != 0) {
4095 			PROC_UNLOCK(p);
4096 			continue;
4097 		}
4098 		xf.xf_pid = p->p_pid;
4099 		xf.xf_uid = p->p_ucred->cr_uid;
4100 		fdp = fdhold(p);
4101 		PROC_UNLOCK(p);
4102 		if (fdp == NULL)
4103 			continue;
4104 		FILEDESC_SLOCK(fdp);
4105 		lastfile = fdlastfile(fdp);
4106 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4107 		    n++) {
4108 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4109 				continue;
4110 			xf.xf_fd = n;
4111 			xf.xf_file = (uintptr_t)fp;
4112 			xf.xf_data = (uintptr_t)fp->f_data;
4113 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4114 			xf.xf_type = (uintptr_t)fp->f_type;
4115 			xf.xf_count = refcount_load(&fp->f_count);
4116 			xf.xf_msgcount = 0;
4117 			xf.xf_offset = foffset_get(fp);
4118 			xf.xf_flag = fp->f_flag;
4119 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4120 			if (error)
4121 				break;
4122 		}
4123 		FILEDESC_SUNLOCK(fdp);
4124 		fddrop(fdp);
4125 		if (error)
4126 			break;
4127 	}
4128 	sx_sunlock(&allproc_lock);
4129 	return (error);
4130 }
4131 
4132 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4133     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4134 
4135 #ifdef KINFO_FILE_SIZE
4136 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4137 #endif
4138 
4139 static int
4140 xlate_fflags(int fflags)
4141 {
4142 	static const struct {
4143 		int	fflag;
4144 		int	kf_fflag;
4145 	} fflags_table[] = {
4146 		{ FAPPEND, KF_FLAG_APPEND },
4147 		{ FASYNC, KF_FLAG_ASYNC },
4148 		{ FFSYNC, KF_FLAG_FSYNC },
4149 		{ FHASLOCK, KF_FLAG_HASLOCK },
4150 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4151 		{ FREAD, KF_FLAG_READ },
4152 		{ FWRITE, KF_FLAG_WRITE },
4153 		{ O_CREAT, KF_FLAG_CREAT },
4154 		{ O_DIRECT, KF_FLAG_DIRECT },
4155 		{ O_EXCL, KF_FLAG_EXCL },
4156 		{ O_EXEC, KF_FLAG_EXEC },
4157 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4158 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4159 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4160 		{ O_TRUNC, KF_FLAG_TRUNC }
4161 	};
4162 	unsigned int i;
4163 	int kflags;
4164 
4165 	kflags = 0;
4166 	for (i = 0; i < nitems(fflags_table); i++)
4167 		if (fflags & fflags_table[i].fflag)
4168 			kflags |=  fflags_table[i].kf_fflag;
4169 	return (kflags);
4170 }
4171 
4172 /* Trim unused data from kf_path by truncating the structure size. */
4173 void
4174 pack_kinfo(struct kinfo_file *kif)
4175 {
4176 
4177 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4178 	    strlen(kif->kf_path) + 1;
4179 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4180 }
4181 
4182 static void
4183 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4184     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4185 {
4186 	int error;
4187 
4188 	bzero(kif, sizeof(*kif));
4189 
4190 	/* Set a default type to allow for empty fill_kinfo() methods. */
4191 	kif->kf_type = KF_TYPE_UNKNOWN;
4192 	kif->kf_flags = xlate_fflags(fp->f_flag);
4193 	if (rightsp != NULL)
4194 		kif->kf_cap_rights = *rightsp;
4195 	else
4196 		cap_rights_init_zero(&kif->kf_cap_rights);
4197 	kif->kf_fd = fd;
4198 	kif->kf_ref_count = refcount_load(&fp->f_count);
4199 	kif->kf_offset = foffset_get(fp);
4200 
4201 	/*
4202 	 * This may drop the filedesc lock, so the 'fp' cannot be
4203 	 * accessed after this call.
4204 	 */
4205 	error = fo_fill_kinfo(fp, kif, fdp);
4206 	if (error == 0)
4207 		kif->kf_status |= KF_ATTR_VALID;
4208 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4209 		pack_kinfo(kif);
4210 	else
4211 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4212 }
4213 
4214 static void
4215 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4216     struct kinfo_file *kif, int flags)
4217 {
4218 	int error;
4219 
4220 	bzero(kif, sizeof(*kif));
4221 
4222 	kif->kf_type = KF_TYPE_VNODE;
4223 	error = vn_fill_kinfo_vnode(vp, kif);
4224 	if (error == 0)
4225 		kif->kf_status |= KF_ATTR_VALID;
4226 	kif->kf_flags = xlate_fflags(fflags);
4227 	cap_rights_init_zero(&kif->kf_cap_rights);
4228 	kif->kf_fd = fd;
4229 	kif->kf_ref_count = -1;
4230 	kif->kf_offset = -1;
4231 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4232 		pack_kinfo(kif);
4233 	else
4234 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4235 	vrele(vp);
4236 }
4237 
4238 struct export_fd_buf {
4239 	struct filedesc		*fdp;
4240 	struct pwddesc	*pdp;
4241 	struct sbuf 		*sb;
4242 	ssize_t			remainder;
4243 	struct kinfo_file	kif;
4244 	int			flags;
4245 };
4246 
4247 static int
4248 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4249 {
4250 	struct kinfo_file *kif;
4251 
4252 	kif = &efbuf->kif;
4253 	if (efbuf->remainder != -1) {
4254 		if (efbuf->remainder < kif->kf_structsize) {
4255 			/* Terminate export. */
4256 			efbuf->remainder = 0;
4257 			return (0);
4258 		}
4259 		efbuf->remainder -= kif->kf_structsize;
4260 	}
4261 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4262 }
4263 
4264 static int
4265 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4266     struct export_fd_buf *efbuf)
4267 {
4268 	int error;
4269 
4270 	if (efbuf->remainder == 0)
4271 		return (0);
4272 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4273 	    efbuf->flags);
4274 	FILEDESC_SUNLOCK(efbuf->fdp);
4275 	error = export_kinfo_to_sb(efbuf);
4276 	FILEDESC_SLOCK(efbuf->fdp);
4277 	return (error);
4278 }
4279 
4280 static int
4281 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4282     struct export_fd_buf *efbuf)
4283 {
4284 	int error;
4285 
4286 	if (efbuf->remainder == 0)
4287 		return (0);
4288 	if (efbuf->pdp != NULL)
4289 		PWDDESC_XUNLOCK(efbuf->pdp);
4290 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4291 	error = export_kinfo_to_sb(efbuf);
4292 	if (efbuf->pdp != NULL)
4293 		PWDDESC_XLOCK(efbuf->pdp);
4294 	return (error);
4295 }
4296 
4297 /*
4298  * Store a process file descriptor information to sbuf.
4299  *
4300  * Takes a locked proc as argument, and returns with the proc unlocked.
4301  */
4302 int
4303 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4304     int flags)
4305 {
4306 	struct file *fp;
4307 	struct filedesc *fdp;
4308 	struct pwddesc *pdp;
4309 	struct export_fd_buf *efbuf;
4310 	struct vnode *cttyvp, *textvp, *tracevp;
4311 	struct pwd *pwd;
4312 	int error, i, lastfile;
4313 	cap_rights_t rights;
4314 
4315 	PROC_LOCK_ASSERT(p, MA_OWNED);
4316 
4317 	/* ktrace vnode */
4318 	tracevp = ktr_get_tracevp(p, true);
4319 	/* text vnode */
4320 	textvp = p->p_textvp;
4321 	if (textvp != NULL)
4322 		vrefact(textvp);
4323 	/* Controlling tty. */
4324 	cttyvp = NULL;
4325 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4326 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4327 		if (cttyvp != NULL)
4328 			vrefact(cttyvp);
4329 	}
4330 	fdp = fdhold(p);
4331 	pdp = pdhold(p);
4332 	PROC_UNLOCK(p);
4333 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4334 	efbuf->fdp = NULL;
4335 	efbuf->pdp = NULL;
4336 	efbuf->sb = sb;
4337 	efbuf->remainder = maxlen;
4338 	efbuf->flags = flags;
4339 	if (tracevp != NULL)
4340 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4341 		    efbuf);
4342 	if (textvp != NULL)
4343 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4344 	if (cttyvp != NULL)
4345 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4346 		    efbuf);
4347 	error = 0;
4348 	if (pdp == NULL || fdp == NULL)
4349 		goto fail;
4350 	efbuf->fdp = fdp;
4351 	efbuf->pdp = pdp;
4352 	PWDDESC_XLOCK(pdp);
4353 	pwd = pwd_hold_pwddesc(pdp);
4354 	if (pwd != NULL) {
4355 		/* working directory */
4356 		if (pwd->pwd_cdir != NULL) {
4357 			vrefact(pwd->pwd_cdir);
4358 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4359 			    FREAD, efbuf);
4360 		}
4361 		/* root directory */
4362 		if (pwd->pwd_rdir != NULL) {
4363 			vrefact(pwd->pwd_rdir);
4364 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4365 			    FREAD, efbuf);
4366 		}
4367 		/* jail directory */
4368 		if (pwd->pwd_jdir != NULL) {
4369 			vrefact(pwd->pwd_jdir);
4370 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4371 			    FREAD, efbuf);
4372 		}
4373 	}
4374 	PWDDESC_XUNLOCK(pdp);
4375 	if (pwd != NULL)
4376 		pwd_drop(pwd);
4377 	FILEDESC_SLOCK(fdp);
4378 	lastfile = fdlastfile(fdp);
4379 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4380 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4381 			continue;
4382 #ifdef CAPABILITIES
4383 		rights = *cap_rights(fdp, i);
4384 #else /* !CAPABILITIES */
4385 		rights = cap_no_rights;
4386 #endif
4387 		/*
4388 		 * Create sysctl entry.  It is OK to drop the filedesc
4389 		 * lock inside of export_file_to_sb() as we will
4390 		 * re-validate and re-evaluate its properties when the
4391 		 * loop continues.
4392 		 */
4393 		error = export_file_to_sb(fp, i, &rights, efbuf);
4394 		if (error != 0 || efbuf->remainder == 0)
4395 			break;
4396 	}
4397 	FILEDESC_SUNLOCK(fdp);
4398 fail:
4399 	if (fdp != NULL)
4400 		fddrop(fdp);
4401 	if (pdp != NULL)
4402 		pddrop(pdp);
4403 	free(efbuf, M_TEMP);
4404 	return (error);
4405 }
4406 
4407 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4408 
4409 /*
4410  * Get per-process file descriptors for use by procstat(1), et al.
4411  */
4412 static int
4413 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4414 {
4415 	struct sbuf sb;
4416 	struct proc *p;
4417 	ssize_t maxlen;
4418 	u_int namelen;
4419 	int error, error2, *name;
4420 
4421 	namelen = arg2;
4422 	if (namelen != 1)
4423 		return (EINVAL);
4424 
4425 	name = (int *)arg1;
4426 
4427 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4428 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4429 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4430 	if (error != 0) {
4431 		sbuf_delete(&sb);
4432 		return (error);
4433 	}
4434 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4435 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4436 	    KERN_FILEDESC_PACK_KINFO);
4437 	error2 = sbuf_finish(&sb);
4438 	sbuf_delete(&sb);
4439 	return (error != 0 ? error : error2);
4440 }
4441 
4442 #ifdef COMPAT_FREEBSD7
4443 #ifdef KINFO_OFILE_SIZE
4444 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4445 #endif
4446 
4447 static void
4448 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4449 {
4450 
4451 	okif->kf_structsize = sizeof(*okif);
4452 	okif->kf_type = kif->kf_type;
4453 	okif->kf_fd = kif->kf_fd;
4454 	okif->kf_ref_count = kif->kf_ref_count;
4455 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4456 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4457 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4458 	okif->kf_offset = kif->kf_offset;
4459 	if (kif->kf_type == KF_TYPE_VNODE)
4460 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4461 	else
4462 		okif->kf_vnode_type = KF_VTYPE_VNON;
4463 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4464 	if (kif->kf_type == KF_TYPE_SOCKET) {
4465 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4466 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4467 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4468 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4469 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4470 	} else {
4471 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4472 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4473 	}
4474 }
4475 
4476 static int
4477 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4478     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4479 {
4480 	int error;
4481 
4482 	vrefact(vp);
4483 	PWDDESC_XUNLOCK(pdp);
4484 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4485 	kinfo_to_okinfo(kif, okif);
4486 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4487 	PWDDESC_XLOCK(pdp);
4488 	return (error);
4489 }
4490 
4491 /*
4492  * Get per-process file descriptors for use by procstat(1), et al.
4493  */
4494 static int
4495 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4496 {
4497 	struct kinfo_ofile *okif;
4498 	struct kinfo_file *kif;
4499 	struct filedesc *fdp;
4500 	struct pwddesc *pdp;
4501 	struct pwd *pwd;
4502 	u_int namelen;
4503 	int error, i, lastfile, *name;
4504 	struct file *fp;
4505 	struct proc *p;
4506 
4507 	namelen = arg2;
4508 	if (namelen != 1)
4509 		return (EINVAL);
4510 
4511 	name = (int *)arg1;
4512 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4513 	if (error != 0)
4514 		return (error);
4515 	fdp = fdhold(p);
4516 	if (fdp != NULL)
4517 		pdp = pdhold(p);
4518 	PROC_UNLOCK(p);
4519 	if (fdp == NULL || pdp == NULL) {
4520 		if (fdp != NULL)
4521 			fddrop(fdp);
4522 		return (ENOENT);
4523 	}
4524 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4525 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4526 	PWDDESC_XLOCK(pdp);
4527 	pwd = pwd_hold_pwddesc(pdp);
4528 	if (pwd != NULL) {
4529 		if (pwd->pwd_cdir != NULL)
4530 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4531 			    okif, pdp, req);
4532 		if (pwd->pwd_rdir != NULL)
4533 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4534 			    okif, pdp, req);
4535 		if (pwd->pwd_jdir != NULL)
4536 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4537 			    okif, pdp, req);
4538 	}
4539 	PWDDESC_XUNLOCK(pdp);
4540 	if (pwd != NULL)
4541 		pwd_drop(pwd);
4542 	FILEDESC_SLOCK(fdp);
4543 	lastfile = fdlastfile(fdp);
4544 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4545 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4546 			continue;
4547 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4548 		    KERN_FILEDESC_PACK_KINFO);
4549 		FILEDESC_SUNLOCK(fdp);
4550 		kinfo_to_okinfo(kif, okif);
4551 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4552 		FILEDESC_SLOCK(fdp);
4553 		if (error)
4554 			break;
4555 	}
4556 	FILEDESC_SUNLOCK(fdp);
4557 	fddrop(fdp);
4558 	pddrop(pdp);
4559 	free(kif, M_TEMP);
4560 	free(okif, M_TEMP);
4561 	return (0);
4562 }
4563 
4564 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4565     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4566     "Process ofiledesc entries");
4567 #endif	/* COMPAT_FREEBSD7 */
4568 
4569 int
4570 vntype_to_kinfo(int vtype)
4571 {
4572 	struct {
4573 		int	vtype;
4574 		int	kf_vtype;
4575 	} vtypes_table[] = {
4576 		{ VBAD, KF_VTYPE_VBAD },
4577 		{ VBLK, KF_VTYPE_VBLK },
4578 		{ VCHR, KF_VTYPE_VCHR },
4579 		{ VDIR, KF_VTYPE_VDIR },
4580 		{ VFIFO, KF_VTYPE_VFIFO },
4581 		{ VLNK, KF_VTYPE_VLNK },
4582 		{ VNON, KF_VTYPE_VNON },
4583 		{ VREG, KF_VTYPE_VREG },
4584 		{ VSOCK, KF_VTYPE_VSOCK }
4585 	};
4586 	unsigned int i;
4587 
4588 	/*
4589 	 * Perform vtype translation.
4590 	 */
4591 	for (i = 0; i < nitems(vtypes_table); i++)
4592 		if (vtypes_table[i].vtype == vtype)
4593 			return (vtypes_table[i].kf_vtype);
4594 
4595 	return (KF_VTYPE_UNKNOWN);
4596 }
4597 
4598 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4599     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4600     "Process filedesc entries");
4601 
4602 /*
4603  * Store a process current working directory information to sbuf.
4604  *
4605  * Takes a locked proc as argument, and returns with the proc unlocked.
4606  */
4607 int
4608 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4609 {
4610 	struct pwddesc *pdp;
4611 	struct pwd *pwd;
4612 	struct export_fd_buf *efbuf;
4613 	struct vnode *cdir;
4614 	int error;
4615 
4616 	PROC_LOCK_ASSERT(p, MA_OWNED);
4617 
4618 	pdp = pdhold(p);
4619 	PROC_UNLOCK(p);
4620 	if (pdp == NULL)
4621 		return (EINVAL);
4622 
4623 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4624 	efbuf->pdp = pdp;
4625 	efbuf->sb = sb;
4626 	efbuf->remainder = maxlen;
4627 
4628 	PWDDESC_XLOCK(pdp);
4629 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4630 	cdir = pwd->pwd_cdir;
4631 	if (cdir == NULL) {
4632 		error = EINVAL;
4633 	} else {
4634 		vrefact(cdir);
4635 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4636 	}
4637 	PWDDESC_XUNLOCK(pdp);
4638 	pddrop(pdp);
4639 	free(efbuf, M_TEMP);
4640 	return (error);
4641 }
4642 
4643 /*
4644  * Get per-process current working directory.
4645  */
4646 static int
4647 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4648 {
4649 	struct sbuf sb;
4650 	struct proc *p;
4651 	ssize_t maxlen;
4652 	u_int namelen;
4653 	int error, error2, *name;
4654 
4655 	namelen = arg2;
4656 	if (namelen != 1)
4657 		return (EINVAL);
4658 
4659 	name = (int *)arg1;
4660 
4661 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4662 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4663 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4664 	if (error != 0) {
4665 		sbuf_delete(&sb);
4666 		return (error);
4667 	}
4668 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4669 	error = kern_proc_cwd_out(p, &sb, maxlen);
4670 	error2 = sbuf_finish(&sb);
4671 	sbuf_delete(&sb);
4672 	return (error != 0 ? error : error2);
4673 }
4674 
4675 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4676     sysctl_kern_proc_cwd, "Process current working directory");
4677 
4678 #ifdef DDB
4679 /*
4680  * For the purposes of debugging, generate a human-readable string for the
4681  * file type.
4682  */
4683 static const char *
4684 file_type_to_name(short type)
4685 {
4686 
4687 	switch (type) {
4688 	case 0:
4689 		return ("zero");
4690 	case DTYPE_VNODE:
4691 		return ("vnode");
4692 	case DTYPE_SOCKET:
4693 		return ("socket");
4694 	case DTYPE_PIPE:
4695 		return ("pipe");
4696 	case DTYPE_FIFO:
4697 		return ("fifo");
4698 	case DTYPE_KQUEUE:
4699 		return ("kqueue");
4700 	case DTYPE_CRYPTO:
4701 		return ("crypto");
4702 	case DTYPE_MQUEUE:
4703 		return ("mqueue");
4704 	case DTYPE_SHM:
4705 		return ("shm");
4706 	case DTYPE_SEM:
4707 		return ("ksem");
4708 	case DTYPE_PTS:
4709 		return ("pts");
4710 	case DTYPE_DEV:
4711 		return ("dev");
4712 	case DTYPE_PROCDESC:
4713 		return ("proc");
4714 	case DTYPE_EVENTFD:
4715 		return ("eventfd");
4716 	case DTYPE_LINUXTFD:
4717 		return ("ltimer");
4718 	default:
4719 		return ("unkn");
4720 	}
4721 }
4722 
4723 /*
4724  * For the purposes of debugging, identify a process (if any, perhaps one of
4725  * many) that references the passed file in its file descriptor array. Return
4726  * NULL if none.
4727  */
4728 static struct proc *
4729 file_to_first_proc(struct file *fp)
4730 {
4731 	struct filedesc *fdp;
4732 	struct proc *p;
4733 	int n;
4734 
4735 	FOREACH_PROC_IN_SYSTEM(p) {
4736 		if (p->p_state == PRS_NEW)
4737 			continue;
4738 		fdp = p->p_fd;
4739 		if (fdp == NULL)
4740 			continue;
4741 		for (n = 0; n < fdp->fd_nfiles; n++) {
4742 			if (fp == fdp->fd_ofiles[n].fde_file)
4743 				return (p);
4744 		}
4745 	}
4746 	return (NULL);
4747 }
4748 
4749 static void
4750 db_print_file(struct file *fp, int header)
4751 {
4752 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4753 	struct proc *p;
4754 
4755 	if (header)
4756 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4757 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4758 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4759 		    "FCmd");
4760 	p = file_to_first_proc(fp);
4761 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4762 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4763 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4764 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4765 
4766 #undef XPTRWIDTH
4767 }
4768 
4769 DB_SHOW_COMMAND(file, db_show_file)
4770 {
4771 	struct file *fp;
4772 
4773 	if (!have_addr) {
4774 		db_printf("usage: show file <addr>\n");
4775 		return;
4776 	}
4777 	fp = (struct file *)addr;
4778 	db_print_file(fp, 1);
4779 }
4780 
4781 DB_SHOW_COMMAND(files, db_show_files)
4782 {
4783 	struct filedesc *fdp;
4784 	struct file *fp;
4785 	struct proc *p;
4786 	int header;
4787 	int n;
4788 
4789 	header = 1;
4790 	FOREACH_PROC_IN_SYSTEM(p) {
4791 		if (p->p_state == PRS_NEW)
4792 			continue;
4793 		if ((fdp = p->p_fd) == NULL)
4794 			continue;
4795 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4796 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4797 				continue;
4798 			db_print_file(fp, header);
4799 			header = 0;
4800 		}
4801 	}
4802 }
4803 #endif
4804 
4805 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4806     &maxfilesperproc, 0, "Maximum files allowed open per process");
4807 
4808 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4809     &maxfiles, 0, "Maximum number of files");
4810 
4811 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4812     &openfiles, 0, "System-wide number of open files");
4813 
4814 /* ARGSUSED*/
4815 static void
4816 filelistinit(void *dummy)
4817 {
4818 
4819 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4820 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4821 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4822 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4823 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4824 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4825 	/*
4826 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4827 	 * the vnode zone.
4828 	 */
4829 	vfs_smr = uma_zone_get_smr(pwd_zone);
4830 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4831 }
4832 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4833 
4834 /*-------------------------------------------------------------------*/
4835 
4836 static int
4837 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4838     int flags, struct thread *td)
4839 {
4840 
4841 	return (EBADF);
4842 }
4843 
4844 static int
4845 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4846     struct thread *td)
4847 {
4848 
4849 	return (EINVAL);
4850 }
4851 
4852 static int
4853 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4854     struct thread *td)
4855 {
4856 
4857 	return (EBADF);
4858 }
4859 
4860 static int
4861 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4862     struct thread *td)
4863 {
4864 
4865 	return (0);
4866 }
4867 
4868 static int
4869 badfo_kqfilter(struct file *fp, struct knote *kn)
4870 {
4871 
4872 	return (EBADF);
4873 }
4874 
4875 static int
4876 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4877 {
4878 
4879 	return (EBADF);
4880 }
4881 
4882 static int
4883 badfo_close(struct file *fp, struct thread *td)
4884 {
4885 
4886 	return (0);
4887 }
4888 
4889 static int
4890 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4891     struct thread *td)
4892 {
4893 
4894 	return (EBADF);
4895 }
4896 
4897 static int
4898 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4899     struct thread *td)
4900 {
4901 
4902 	return (EBADF);
4903 }
4904 
4905 static int
4906 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4907     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4908     struct thread *td)
4909 {
4910 
4911 	return (EBADF);
4912 }
4913 
4914 static int
4915 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4916 {
4917 
4918 	return (0);
4919 }
4920 
4921 struct fileops badfileops = {
4922 	.fo_read = badfo_readwrite,
4923 	.fo_write = badfo_readwrite,
4924 	.fo_truncate = badfo_truncate,
4925 	.fo_ioctl = badfo_ioctl,
4926 	.fo_poll = badfo_poll,
4927 	.fo_kqfilter = badfo_kqfilter,
4928 	.fo_stat = badfo_stat,
4929 	.fo_close = badfo_close,
4930 	.fo_chmod = badfo_chmod,
4931 	.fo_chown = badfo_chown,
4932 	.fo_sendfile = badfo_sendfile,
4933 	.fo_fill_kinfo = badfo_fill_kinfo,
4934 };
4935 
4936 static int
4937 path_poll(struct file *fp, int events, struct ucred *active_cred,
4938     struct thread *td)
4939 {
4940 	return (POLLNVAL);
4941 }
4942 
4943 static int
4944 path_close(struct file *fp, struct thread *td)
4945 {
4946 	MPASS(fp->f_type == DTYPE_VNODE);
4947 	fp->f_ops = &badfileops;
4948 	vdrop(fp->f_vnode);
4949 	return (0);
4950 }
4951 
4952 struct fileops path_fileops = {
4953 	.fo_read = badfo_readwrite,
4954 	.fo_write = badfo_readwrite,
4955 	.fo_truncate = badfo_truncate,
4956 	.fo_ioctl = badfo_ioctl,
4957 	.fo_poll = path_poll,
4958 	.fo_kqfilter = vn_kqfilter_opath,
4959 	.fo_stat = vn_statfile,
4960 	.fo_close = path_close,
4961 	.fo_chmod = badfo_chmod,
4962 	.fo_chown = badfo_chown,
4963 	.fo_sendfile = badfo_sendfile,
4964 	.fo_fill_kinfo = vn_fill_kinfo,
4965 	.fo_flags = DFLAG_PASSABLE,
4966 };
4967 
4968 int
4969 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4970     int flags, struct thread *td)
4971 {
4972 
4973 	return (EOPNOTSUPP);
4974 }
4975 
4976 int
4977 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4978     struct thread *td)
4979 {
4980 
4981 	return (EINVAL);
4982 }
4983 
4984 int
4985 invfo_ioctl(struct file *fp, u_long com, void *data,
4986     struct ucred *active_cred, struct thread *td)
4987 {
4988 
4989 	return (ENOTTY);
4990 }
4991 
4992 int
4993 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4994     struct thread *td)
4995 {
4996 
4997 	return (poll_no_poll(events));
4998 }
4999 
5000 int
5001 invfo_kqfilter(struct file *fp, struct knote *kn)
5002 {
5003 
5004 	return (EINVAL);
5005 }
5006 
5007 int
5008 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5009     struct thread *td)
5010 {
5011 
5012 	return (EINVAL);
5013 }
5014 
5015 int
5016 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5017     struct thread *td)
5018 {
5019 
5020 	return (EINVAL);
5021 }
5022 
5023 int
5024 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5025     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5026     struct thread *td)
5027 {
5028 
5029 	return (EINVAL);
5030 }
5031 
5032 /*-------------------------------------------------------------------*/
5033 
5034 /*
5035  * File Descriptor pseudo-device driver (/dev/fd/).
5036  *
5037  * Opening minor device N dup()s the file (if any) connected to file
5038  * descriptor N belonging to the calling process.  Note that this driver
5039  * consists of only the ``open()'' routine, because all subsequent
5040  * references to this file will be direct to the other driver.
5041  *
5042  * XXX: we could give this one a cloning event handler if necessary.
5043  */
5044 
5045 /* ARGSUSED */
5046 static int
5047 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5048 {
5049 
5050 	/*
5051 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5052 	 * the file descriptor being sought for duplication. The error
5053 	 * return ensures that the vnode for this device will be released
5054 	 * by vn_open. Open will detect this special error and take the
5055 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5056 	 * will simply report the error.
5057 	 */
5058 	td->td_dupfd = dev2unit(dev);
5059 	return (ENODEV);
5060 }
5061 
5062 static struct cdevsw fildesc_cdevsw = {
5063 	.d_version =	D_VERSION,
5064 	.d_open =	fdopen,
5065 	.d_name =	"FD",
5066 };
5067 
5068 static void
5069 fildesc_drvinit(void *unused)
5070 {
5071 	struct cdev *dev;
5072 
5073 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5074 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5075 	make_dev_alias(dev, "stdin");
5076 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5077 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5078 	make_dev_alias(dev, "stdout");
5079 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5080 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5081 	make_dev_alias(dev, "stderr");
5082 }
5083 
5084 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5085