xref: /freebsd/sys/kern/kern_descrip.c (revision e2eeea75eb8b6dd50c1298067a0655880d186734)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98     "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 
102 MALLOC_DECLARE(M_FADVISE);
103 
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
107 VFS_SMR_DECLARE;
108 
109 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
110 		    struct thread *td, int holdleaders);
111 static int	fd_first_free(struct filedesc *fdp, int low, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the last used fd.
217  *
218  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219  * Otherwise use fdlastfile.
220  */
221 int
222 fdlastfile_single(struct filedesc *fdp)
223 {
224 	NDSLOTTYPE *map = fdp->fd_map;
225 	int off, minoff;
226 
227 	off = NDSLOT(fdp->fd_nfiles - 1);
228 	for (minoff = NDSLOT(0); off >= minoff; --off)
229 		if (map[off] != 0)
230 			return (off * NDENTRIES + flsl(map[off]) - 1);
231 	return (-1);
232 }
233 
234 int
235 fdlastfile(struct filedesc *fdp)
236 {
237 
238 	FILEDESC_LOCK_ASSERT(fdp);
239 	return (fdlastfile_single(fdp));
240 }
241 
242 static int
243 fdisused(struct filedesc *fdp, int fd)
244 {
245 
246 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 
249 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
250 }
251 
252 /*
253  * Mark a file descriptor as used.
254  */
255 static void
256 fdused_init(struct filedesc *fdp, int fd)
257 {
258 
259 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 
261 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
262 }
263 
264 static void
265 fdused(struct filedesc *fdp, int fd)
266 {
267 
268 	FILEDESC_XLOCK_ASSERT(fdp);
269 
270 	fdused_init(fdp, fd);
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile++;
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	fde = &fdp->fd_ofiles[fd];
311 #ifdef CAPABILITIES
312 	seqc_write_begin(&fde->fde_seqc);
313 #endif
314 	fde->fde_file = NULL;
315 #ifdef CAPABILITIES
316 	seqc_write_end(&fde->fde_seqc);
317 #endif
318 	fdefree_last(fde);
319 	fdunused(fdp, fd);
320 }
321 
322 /*
323  * System calls on descriptors.
324  */
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
327 	int	dummy;
328 };
329 #endif
330 /* ARGSUSED */
331 int
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
333 {
334 #ifdef	RACCT
335 	uint64_t lim;
336 #endif
337 
338 	td->td_retval[0] = getmaxfd(td);
339 #ifdef	RACCT
340 	PROC_LOCK(td->td_proc);
341 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 	PROC_UNLOCK(td->td_proc);
343 	if (lim < td->td_retval[0])
344 		td->td_retval[0] = lim;
345 #endif
346 	return (0);
347 }
348 
349 /*
350  * Duplicate a file descriptor to a particular value.
351  *
352  * Note: keep in mind that a potential race condition exists when closing
353  * descriptors from a shared descriptor table (via rfork).
354  */
355 #ifndef _SYS_SYSPROTO_H_
356 struct dup2_args {
357 	u_int	from;
358 	u_int	to;
359 };
360 #endif
361 /* ARGSUSED */
362 int
363 sys_dup2(struct thread *td, struct dup2_args *uap)
364 {
365 
366 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
367 }
368 
369 /*
370  * Duplicate a file descriptor.
371  */
372 #ifndef _SYS_SYSPROTO_H_
373 struct dup_args {
374 	u_int	fd;
375 };
376 #endif
377 /* ARGSUSED */
378 int
379 sys_dup(struct thread *td, struct dup_args *uap)
380 {
381 
382 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
383 }
384 
385 /*
386  * The file control system call.
387  */
388 #ifndef _SYS_SYSPROTO_H_
389 struct fcntl_args {
390 	int	fd;
391 	int	cmd;
392 	long	arg;
393 };
394 #endif
395 /* ARGSUSED */
396 int
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
398 {
399 
400 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
401 }
402 
403 int
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
405 {
406 	struct flock fl;
407 	struct __oflock ofl;
408 	intptr_t arg1;
409 	int error, newcmd;
410 
411 	error = 0;
412 	newcmd = cmd;
413 	switch (cmd) {
414 	case F_OGETLK:
415 	case F_OSETLK:
416 	case F_OSETLKW:
417 		/*
418 		 * Convert old flock structure to new.
419 		 */
420 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 		fl.l_start = ofl.l_start;
422 		fl.l_len = ofl.l_len;
423 		fl.l_pid = ofl.l_pid;
424 		fl.l_type = ofl.l_type;
425 		fl.l_whence = ofl.l_whence;
426 		fl.l_sysid = 0;
427 
428 		switch (cmd) {
429 		case F_OGETLK:
430 			newcmd = F_GETLK;
431 			break;
432 		case F_OSETLK:
433 			newcmd = F_SETLK;
434 			break;
435 		case F_OSETLKW:
436 			newcmd = F_SETLKW;
437 			break;
438 		}
439 		arg1 = (intptr_t)&fl;
440 		break;
441 	case F_GETLK:
442 	case F_SETLK:
443 	case F_SETLKW:
444 	case F_SETLK_REMOTE:
445 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 		arg1 = (intptr_t)&fl;
447 		break;
448 	default:
449 		arg1 = arg;
450 		break;
451 	}
452 	if (error)
453 		return (error);
454 	error = kern_fcntl(td, fd, newcmd, arg1);
455 	if (error)
456 		return (error);
457 	if (cmd == F_OGETLK) {
458 		ofl.l_start = fl.l_start;
459 		ofl.l_len = fl.l_len;
460 		ofl.l_pid = fl.l_pid;
461 		ofl.l_type = fl.l_type;
462 		ofl.l_whence = fl.l_whence;
463 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 	} else if (cmd == F_GETLK) {
465 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
466 	}
467 	return (error);
468 }
469 
470 int
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
472 {
473 	struct filedesc *fdp;
474 	struct flock *flp;
475 	struct file *fp, *fp2;
476 	struct filedescent *fde;
477 	struct proc *p;
478 	struct vnode *vp;
479 	struct mount *mp;
480 	int error, flg, seals, tmp;
481 	uint64_t bsize;
482 	off_t foffset;
483 
484 	error = 0;
485 	flg = F_POSIX;
486 	p = td->td_proc;
487 	fdp = p->p_fd;
488 
489 	AUDIT_ARG_FD(cmd);
490 	AUDIT_ARG_CMD(cmd);
491 	switch (cmd) {
492 	case F_DUPFD:
493 		tmp = arg;
494 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
495 		break;
496 
497 	case F_DUPFD_CLOEXEC:
498 		tmp = arg;
499 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
500 		break;
501 
502 	case F_DUP2FD:
503 		tmp = arg;
504 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
505 		break;
506 
507 	case F_DUP2FD_CLOEXEC:
508 		tmp = arg;
509 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
510 		break;
511 
512 	case F_GETFD:
513 		error = EBADF;
514 		FILEDESC_SLOCK(fdp);
515 		fde = fdeget_locked(fdp, fd);
516 		if (fde != NULL) {
517 			td->td_retval[0] =
518 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
519 			error = 0;
520 		}
521 		FILEDESC_SUNLOCK(fdp);
522 		break;
523 
524 	case F_SETFD:
525 		error = EBADF;
526 		FILEDESC_XLOCK(fdp);
527 		fde = fdeget_locked(fdp, fd);
528 		if (fde != NULL) {
529 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
531 			error = 0;
532 		}
533 		FILEDESC_XUNLOCK(fdp);
534 		break;
535 
536 	case F_GETFL:
537 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
538 		if (error != 0)
539 			break;
540 		td->td_retval[0] = OFLAGS(fp->f_flag);
541 		fdrop(fp, td);
542 		break;
543 
544 	case F_SETFL:
545 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
546 		if (error != 0)
547 			break;
548 		do {
549 			tmp = flg = fp->f_flag;
550 			tmp &= ~FCNTLFLAGS;
551 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
552 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
553 		tmp = fp->f_flag & FNONBLOCK;
554 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
555 		if (error != 0) {
556 			fdrop(fp, td);
557 			break;
558 		}
559 		tmp = fp->f_flag & FASYNC;
560 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
561 		if (error == 0) {
562 			fdrop(fp, td);
563 			break;
564 		}
565 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
566 		tmp = 0;
567 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
568 		fdrop(fp, td);
569 		break;
570 
571 	case F_GETOWN:
572 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
573 		if (error != 0)
574 			break;
575 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
576 		if (error == 0)
577 			td->td_retval[0] = tmp;
578 		fdrop(fp, td);
579 		break;
580 
581 	case F_SETOWN:
582 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
583 		if (error != 0)
584 			break;
585 		tmp = arg;
586 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
587 		fdrop(fp, td);
588 		break;
589 
590 	case F_SETLK_REMOTE:
591 		error = priv_check(td, PRIV_NFS_LOCKD);
592 		if (error != 0)
593 			return (error);
594 		flg = F_REMOTE;
595 		goto do_setlk;
596 
597 	case F_SETLKW:
598 		flg |= F_WAIT;
599 		/* FALLTHROUGH F_SETLK */
600 
601 	case F_SETLK:
602 	do_setlk:
603 		flp = (struct flock *)arg;
604 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
605 			error = EINVAL;
606 			break;
607 		}
608 
609 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
610 		if (error != 0)
611 			break;
612 		if (fp->f_type != DTYPE_VNODE) {
613 			error = EBADF;
614 			fdrop(fp, td);
615 			break;
616 		}
617 
618 		if (flp->l_whence == SEEK_CUR) {
619 			foffset = foffset_get(fp);
620 			if (foffset < 0 ||
621 			    (flp->l_start > 0 &&
622 			     foffset > OFF_MAX - flp->l_start)) {
623 				error = EOVERFLOW;
624 				fdrop(fp, td);
625 				break;
626 			}
627 			flp->l_start += foffset;
628 		}
629 
630 		vp = fp->f_vnode;
631 		switch (flp->l_type) {
632 		case F_RDLCK:
633 			if ((fp->f_flag & FREAD) == 0) {
634 				error = EBADF;
635 				break;
636 			}
637 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
638 				PROC_LOCK(p->p_leader);
639 				p->p_leader->p_flag |= P_ADVLOCK;
640 				PROC_UNLOCK(p->p_leader);
641 			}
642 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
643 			    flp, flg);
644 			break;
645 		case F_WRLCK:
646 			if ((fp->f_flag & FWRITE) == 0) {
647 				error = EBADF;
648 				break;
649 			}
650 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 				PROC_LOCK(p->p_leader);
652 				p->p_leader->p_flag |= P_ADVLOCK;
653 				PROC_UNLOCK(p->p_leader);
654 			}
655 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
656 			    flp, flg);
657 			break;
658 		case F_UNLCK:
659 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
660 			    flp, flg);
661 			break;
662 		case F_UNLCKSYS:
663 			if (flg != F_REMOTE) {
664 				error = EINVAL;
665 				break;
666 			}
667 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
668 			    F_UNLCKSYS, flp, flg);
669 			break;
670 		default:
671 			error = EINVAL;
672 			break;
673 		}
674 		if (error != 0 || flp->l_type == F_UNLCK ||
675 		    flp->l_type == F_UNLCKSYS) {
676 			fdrop(fp, td);
677 			break;
678 		}
679 
680 		/*
681 		 * Check for a race with close.
682 		 *
683 		 * The vnode is now advisory locked (or unlocked, but this case
684 		 * is not really important) as the caller requested.
685 		 * We had to drop the filedesc lock, so we need to recheck if
686 		 * the descriptor is still valid, because if it was closed
687 		 * in the meantime we need to remove advisory lock from the
688 		 * vnode - close on any descriptor leading to an advisory
689 		 * locked vnode, removes that lock.
690 		 * We will return 0 on purpose in that case, as the result of
691 		 * successful advisory lock might have been externally visible
692 		 * already. This is fine - effectively we pretend to the caller
693 		 * that the closing thread was a bit slower and that the
694 		 * advisory lock succeeded before the close.
695 		 */
696 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
697 		if (error != 0) {
698 			fdrop(fp, td);
699 			break;
700 		}
701 		if (fp != fp2) {
702 			flp->l_whence = SEEK_SET;
703 			flp->l_start = 0;
704 			flp->l_len = 0;
705 			flp->l_type = F_UNLCK;
706 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
707 			    F_UNLCK, flp, F_POSIX);
708 		}
709 		fdrop(fp, td);
710 		fdrop(fp2, td);
711 		break;
712 
713 	case F_GETLK:
714 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
715 		if (error != 0)
716 			break;
717 		if (fp->f_type != DTYPE_VNODE) {
718 			error = EBADF;
719 			fdrop(fp, td);
720 			break;
721 		}
722 		flp = (struct flock *)arg;
723 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
724 		    flp->l_type != F_UNLCK) {
725 			error = EINVAL;
726 			fdrop(fp, td);
727 			break;
728 		}
729 		if (flp->l_whence == SEEK_CUR) {
730 			foffset = foffset_get(fp);
731 			if ((flp->l_start > 0 &&
732 			    foffset > OFF_MAX - flp->l_start) ||
733 			    (flp->l_start < 0 &&
734 			    foffset < OFF_MIN - flp->l_start)) {
735 				error = EOVERFLOW;
736 				fdrop(fp, td);
737 				break;
738 			}
739 			flp->l_start += foffset;
740 		}
741 		vp = fp->f_vnode;
742 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
743 		    F_POSIX);
744 		fdrop(fp, td);
745 		break;
746 
747 	case F_ADD_SEALS:
748 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
749 		if (error != 0)
750 			break;
751 		error = fo_add_seals(fp, arg);
752 		fdrop(fp, td);
753 		break;
754 
755 	case F_GET_SEALS:
756 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
757 		if (error != 0)
758 			break;
759 		if (fo_get_seals(fp, &seals) == 0)
760 			td->td_retval[0] = seals;
761 		else
762 			error = EINVAL;
763 		fdrop(fp, td);
764 		break;
765 
766 	case F_RDAHEAD:
767 		arg = arg ? 128 * 1024: 0;
768 		/* FALLTHROUGH */
769 	case F_READAHEAD:
770 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
771 		if (error != 0)
772 			break;
773 		if (fp->f_type != DTYPE_VNODE) {
774 			fdrop(fp, td);
775 			error = EBADF;
776 			break;
777 		}
778 		vp = fp->f_vnode;
779 		if (vp->v_type != VREG) {
780 			fdrop(fp, td);
781 			error = ENOTTY;
782 			break;
783 		}
784 
785 		/*
786 		 * Exclusive lock synchronizes against f_seqcount reads and
787 		 * writes in sequential_heuristic().
788 		 */
789 		error = vn_lock(vp, LK_EXCLUSIVE);
790 		if (error != 0) {
791 			fdrop(fp, td);
792 			break;
793 		}
794 		if (arg >= 0) {
795 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
796 			arg = MIN(arg, INT_MAX - bsize + 1);
797 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
798 			    (arg + bsize - 1) / bsize);
799 			atomic_set_int(&fp->f_flag, FRDAHEAD);
800 		} else {
801 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
802 		}
803 		VOP_UNLOCK(vp);
804 		fdrop(fp, td);
805 		break;
806 
807 	case F_ISUNIONSTACK:
808 		/*
809 		 * Check if the vnode is part of a union stack (either the
810 		 * "union" flag from mount(2) or unionfs).
811 		 *
812 		 * Prior to introduction of this op libc's readdir would call
813 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
814 		 * data just to check fs name and a mount flag.
815 		 *
816 		 * Fixing the code to handle everything in the kernel instead
817 		 * is a non-trivial endeavor and has low priority, thus this
818 		 * horrible kludge facilitates the current behavior in a much
819 		 * cheaper manner until someone(tm) sorts this out.
820 		 */
821 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
822 		if (error != 0)
823 			break;
824 		if (fp->f_type != DTYPE_VNODE) {
825 			fdrop(fp, td);
826 			error = EBADF;
827 			break;
828 		}
829 		vp = fp->f_vnode;
830 		/*
831 		 * Since we don't prevent dooming the vnode even non-null mp
832 		 * found can become immediately stale. This is tolerable since
833 		 * mount points are type-stable (providing safe memory access)
834 		 * and any vfs op on this vnode going forward will return an
835 		 * error (meaning return value in this case is meaningless).
836 		 */
837 		mp = atomic_load_ptr(&vp->v_mount);
838 		if (__predict_false(mp == NULL)) {
839 			fdrop(fp, td);
840 			error = EBADF;
841 			break;
842 		}
843 		td->td_retval[0] = 0;
844 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
845 		    mp->mnt_flag & MNT_UNION)
846 			td->td_retval[0] = 1;
847 		fdrop(fp, td);
848 		break;
849 
850 	default:
851 		error = EINVAL;
852 		break;
853 	}
854 	return (error);
855 }
856 
857 static int
858 getmaxfd(struct thread *td)
859 {
860 
861 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
862 }
863 
864 /*
865  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
866  */
867 int
868 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
869 {
870 	struct filedesc *fdp;
871 	struct filedescent *oldfde, *newfde;
872 	struct proc *p;
873 	struct file *delfp;
874 	u_long *oioctls, *nioctls;
875 	int error, maxfd;
876 
877 	p = td->td_proc;
878 	fdp = p->p_fd;
879 	oioctls = NULL;
880 
881 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
882 	MPASS(mode < FDDUP_LASTMODE);
883 
884 	AUDIT_ARG_FD(old);
885 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
886 
887 	/*
888 	 * Verify we have a valid descriptor to dup from and possibly to
889 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
890 	 * return EINVAL when the new descriptor is out of bounds.
891 	 */
892 	if (old < 0)
893 		return (EBADF);
894 	if (new < 0)
895 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
896 	maxfd = getmaxfd(td);
897 	if (new >= maxfd)
898 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
899 
900 	error = EBADF;
901 	FILEDESC_XLOCK(fdp);
902 	if (fget_locked(fdp, old) == NULL)
903 		goto unlock;
904 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
905 		td->td_retval[0] = new;
906 		if (flags & FDDUP_FLAG_CLOEXEC)
907 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
908 		error = 0;
909 		goto unlock;
910 	}
911 
912 	oldfde = &fdp->fd_ofiles[old];
913 	if (!fhold(oldfde->fde_file))
914 		goto unlock;
915 
916 	/*
917 	 * If the caller specified a file descriptor, make sure the file
918 	 * table is large enough to hold it, and grab it.  Otherwise, just
919 	 * allocate a new descriptor the usual way.
920 	 */
921 	switch (mode) {
922 	case FDDUP_NORMAL:
923 	case FDDUP_FCNTL:
924 		if ((error = fdalloc(td, new, &new)) != 0) {
925 			fdrop(oldfde->fde_file, td);
926 			goto unlock;
927 		}
928 		break;
929 	case FDDUP_MUSTREPLACE:
930 		/* Target file descriptor must exist. */
931 		if (fget_locked(fdp, new) == NULL) {
932 			fdrop(oldfde->fde_file, td);
933 			goto unlock;
934 		}
935 		break;
936 	case FDDUP_FIXED:
937 		if (new >= fdp->fd_nfiles) {
938 			/*
939 			 * The resource limits are here instead of e.g.
940 			 * fdalloc(), because the file descriptor table may be
941 			 * shared between processes, so we can't really use
942 			 * racct_add()/racct_sub().  Instead of counting the
943 			 * number of actually allocated descriptors, just put
944 			 * the limit on the size of the file descriptor table.
945 			 */
946 #ifdef RACCT
947 			if (RACCT_ENABLED()) {
948 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
949 				if (error != 0) {
950 					error = EMFILE;
951 					fdrop(oldfde->fde_file, td);
952 					goto unlock;
953 				}
954 			}
955 #endif
956 			fdgrowtable_exp(fdp, new + 1);
957 		}
958 		if (!fdisused(fdp, new))
959 			fdused(fdp, new);
960 		break;
961 	default:
962 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
963 	}
964 
965 	KASSERT(old != new, ("new fd is same as old"));
966 
967 	newfde = &fdp->fd_ofiles[new];
968 	delfp = newfde->fde_file;
969 
970 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
971 
972 	/*
973 	 * Duplicate the source descriptor.
974 	 */
975 #ifdef CAPABILITIES
976 	seqc_write_begin(&newfde->fde_seqc);
977 #endif
978 	oioctls = filecaps_free_prep(&newfde->fde_caps);
979 	memcpy(newfde, oldfde, fde_change_size);
980 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
981 	    nioctls);
982 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
983 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
984 	else
985 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
986 #ifdef CAPABILITIES
987 	seqc_write_end(&newfde->fde_seqc);
988 #endif
989 	td->td_retval[0] = new;
990 
991 	error = 0;
992 
993 	if (delfp != NULL) {
994 		(void) closefp(fdp, new, delfp, td, 1);
995 		FILEDESC_UNLOCK_ASSERT(fdp);
996 	} else {
997 unlock:
998 		FILEDESC_XUNLOCK(fdp);
999 	}
1000 
1001 	filecaps_free_finish(oioctls);
1002 	return (error);
1003 }
1004 
1005 static void
1006 sigiofree(struct sigio *sigio)
1007 {
1008 	crfree(sigio->sio_ucred);
1009 	free(sigio, M_SIGIO);
1010 }
1011 
1012 static struct sigio *
1013 funsetown_locked(struct sigio *sigio)
1014 {
1015 	struct proc *p;
1016 	struct pgrp *pg;
1017 
1018 	SIGIO_ASSERT_LOCKED();
1019 
1020 	if (sigio == NULL)
1021 		return (NULL);
1022 	*(sigio->sio_myref) = NULL;
1023 	if (sigio->sio_pgid < 0) {
1024 		pg = sigio->sio_pgrp;
1025 		PGRP_LOCK(pg);
1026 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1027 		    sigio, sio_pgsigio);
1028 		PGRP_UNLOCK(pg);
1029 	} else {
1030 		p = sigio->sio_proc;
1031 		PROC_LOCK(p);
1032 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1033 		    sigio, sio_pgsigio);
1034 		PROC_UNLOCK(p);
1035 	}
1036 	return (sigio);
1037 }
1038 
1039 /*
1040  * If sigio is on the list associated with a process or process group,
1041  * disable signalling from the device, remove sigio from the list and
1042  * free sigio.
1043  */
1044 void
1045 funsetown(struct sigio **sigiop)
1046 {
1047 	struct sigio *sigio;
1048 
1049 	/* Racy check, consumers must provide synchronization. */
1050 	if (*sigiop == NULL)
1051 		return;
1052 
1053 	SIGIO_LOCK();
1054 	sigio = funsetown_locked(*sigiop);
1055 	SIGIO_UNLOCK();
1056 	if (sigio != NULL)
1057 		sigiofree(sigio);
1058 }
1059 
1060 /*
1061  * Free a list of sigio structures.  The caller must ensure that new sigio
1062  * structures cannot be added after this point.  For process groups this is
1063  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1064  * as an interlock.
1065  */
1066 void
1067 funsetownlst(struct sigiolst *sigiolst)
1068 {
1069 	struct proc *p;
1070 	struct pgrp *pg;
1071 	struct sigio *sigio, *tmp;
1072 
1073 	/* Racy check. */
1074 	sigio = SLIST_FIRST(sigiolst);
1075 	if (sigio == NULL)
1076 		return;
1077 
1078 	p = NULL;
1079 	pg = NULL;
1080 
1081 	SIGIO_LOCK();
1082 	sigio = SLIST_FIRST(sigiolst);
1083 	if (sigio == NULL) {
1084 		SIGIO_UNLOCK();
1085 		return;
1086 	}
1087 
1088 	/*
1089 	 * Every entry of the list should belong to a single proc or pgrp.
1090 	 */
1091 	if (sigio->sio_pgid < 0) {
1092 		pg = sigio->sio_pgrp;
1093 		sx_assert(&proctree_lock, SX_XLOCKED);
1094 		PGRP_LOCK(pg);
1095 	} else /* if (sigio->sio_pgid > 0) */ {
1096 		p = sigio->sio_proc;
1097 		PROC_LOCK(p);
1098 		KASSERT((p->p_flag & P_WEXIT) != 0,
1099 		    ("%s: process %p is not exiting", __func__, p));
1100 	}
1101 
1102 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1103 		*sigio->sio_myref = NULL;
1104 		if (pg != NULL) {
1105 			KASSERT(sigio->sio_pgid < 0,
1106 			    ("Proc sigio in pgrp sigio list"));
1107 			KASSERT(sigio->sio_pgrp == pg,
1108 			    ("Bogus pgrp in sigio list"));
1109 		} else /* if (p != NULL) */ {
1110 			KASSERT(sigio->sio_pgid > 0,
1111 			    ("Pgrp sigio in proc sigio list"));
1112 			KASSERT(sigio->sio_proc == p,
1113 			    ("Bogus proc in sigio list"));
1114 		}
1115 	}
1116 
1117 	if (pg != NULL)
1118 		PGRP_UNLOCK(pg);
1119 	else
1120 		PROC_UNLOCK(p);
1121 	SIGIO_UNLOCK();
1122 
1123 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1124 		sigiofree(sigio);
1125 }
1126 
1127 /*
1128  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1129  *
1130  * After permission checking, add a sigio structure to the sigio list for
1131  * the process or process group.
1132  */
1133 int
1134 fsetown(pid_t pgid, struct sigio **sigiop)
1135 {
1136 	struct proc *proc;
1137 	struct pgrp *pgrp;
1138 	struct sigio *osigio, *sigio;
1139 	int ret;
1140 
1141 	if (pgid == 0) {
1142 		funsetown(sigiop);
1143 		return (0);
1144 	}
1145 
1146 	ret = 0;
1147 
1148 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1149 	sigio->sio_pgid = pgid;
1150 	sigio->sio_ucred = crhold(curthread->td_ucred);
1151 	sigio->sio_myref = sigiop;
1152 
1153 	sx_slock(&proctree_lock);
1154 	SIGIO_LOCK();
1155 	osigio = funsetown_locked(*sigiop);
1156 	if (pgid > 0) {
1157 		proc = pfind(pgid);
1158 		if (proc == NULL) {
1159 			ret = ESRCH;
1160 			goto fail;
1161 		}
1162 
1163 		/*
1164 		 * Policy - Don't allow a process to FSETOWN a process
1165 		 * in another session.
1166 		 *
1167 		 * Remove this test to allow maximum flexibility or
1168 		 * restrict FSETOWN to the current process or process
1169 		 * group for maximum safety.
1170 		 */
1171 		if (proc->p_session != curthread->td_proc->p_session) {
1172 			PROC_UNLOCK(proc);
1173 			ret = EPERM;
1174 			goto fail;
1175 		}
1176 
1177 		sigio->sio_proc = proc;
1178 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1179 		PROC_UNLOCK(proc);
1180 	} else /* if (pgid < 0) */ {
1181 		pgrp = pgfind(-pgid);
1182 		if (pgrp == NULL) {
1183 			ret = ESRCH;
1184 			goto fail;
1185 		}
1186 
1187 		/*
1188 		 * Policy - Don't allow a process to FSETOWN a process
1189 		 * in another session.
1190 		 *
1191 		 * Remove this test to allow maximum flexibility or
1192 		 * restrict FSETOWN to the current process or process
1193 		 * group for maximum safety.
1194 		 */
1195 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1196 			PGRP_UNLOCK(pgrp);
1197 			ret = EPERM;
1198 			goto fail;
1199 		}
1200 
1201 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1202 		sigio->sio_pgrp = pgrp;
1203 		PGRP_UNLOCK(pgrp);
1204 	}
1205 	sx_sunlock(&proctree_lock);
1206 	*sigiop = sigio;
1207 	SIGIO_UNLOCK();
1208 	if (osigio != NULL)
1209 		sigiofree(osigio);
1210 	return (0);
1211 
1212 fail:
1213 	SIGIO_UNLOCK();
1214 	sx_sunlock(&proctree_lock);
1215 	sigiofree(sigio);
1216 	if (osigio != NULL)
1217 		sigiofree(osigio);
1218 	return (ret);
1219 }
1220 
1221 /*
1222  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1223  */
1224 pid_t
1225 fgetown(struct sigio **sigiop)
1226 {
1227 	pid_t pgid;
1228 
1229 	SIGIO_LOCK();
1230 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1231 	SIGIO_UNLOCK();
1232 	return (pgid);
1233 }
1234 
1235 /*
1236  * Function drops the filedesc lock on return.
1237  */
1238 static int
1239 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1240     int holdleaders)
1241 {
1242 	int error;
1243 
1244 	FILEDESC_XLOCK_ASSERT(fdp);
1245 
1246 	if (holdleaders) {
1247 		if (td->td_proc->p_fdtol != NULL) {
1248 			/*
1249 			 * Ask fdfree() to sleep to ensure that all relevant
1250 			 * process leaders can be traversed in closef().
1251 			 */
1252 			fdp->fd_holdleaderscount++;
1253 		} else {
1254 			holdleaders = 0;
1255 		}
1256 	}
1257 
1258 	/*
1259 	 * We now hold the fp reference that used to be owned by the
1260 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1261 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1262 	 * added, and deleteing a knote for the new fd.
1263 	 */
1264 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1265 		knote_fdclose(td, fd);
1266 
1267 	/*
1268 	 * We need to notify mqueue if the object is of type mqueue.
1269 	 */
1270 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1271 		mq_fdclose(td, fd, fp);
1272 	FILEDESC_XUNLOCK(fdp);
1273 
1274 	error = closef(fp, td);
1275 	if (holdleaders) {
1276 		FILEDESC_XLOCK(fdp);
1277 		fdp->fd_holdleaderscount--;
1278 		if (fdp->fd_holdleaderscount == 0 &&
1279 		    fdp->fd_holdleaderswakeup != 0) {
1280 			fdp->fd_holdleaderswakeup = 0;
1281 			wakeup(&fdp->fd_holdleaderscount);
1282 		}
1283 		FILEDESC_XUNLOCK(fdp);
1284 	}
1285 	return (error);
1286 }
1287 
1288 /*
1289  * Close a file descriptor.
1290  */
1291 #ifndef _SYS_SYSPROTO_H_
1292 struct close_args {
1293 	int     fd;
1294 };
1295 #endif
1296 /* ARGSUSED */
1297 int
1298 sys_close(struct thread *td, struct close_args *uap)
1299 {
1300 
1301 	return (kern_close(td, uap->fd));
1302 }
1303 
1304 int
1305 kern_close(struct thread *td, int fd)
1306 {
1307 	struct filedesc *fdp;
1308 	struct file *fp;
1309 
1310 	fdp = td->td_proc->p_fd;
1311 
1312 	AUDIT_SYSCLOSE(td, fd);
1313 
1314 	FILEDESC_XLOCK(fdp);
1315 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1316 		FILEDESC_XUNLOCK(fdp);
1317 		return (EBADF);
1318 	}
1319 	fdfree(fdp, fd);
1320 
1321 	/* closefp() drops the FILEDESC lock for us. */
1322 	return (closefp(fdp, fd, fp, td, 1));
1323 }
1324 
1325 int
1326 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1327 {
1328 	struct filedesc *fdp;
1329 	int fd, ret, lastfile;
1330 
1331 	ret = 0;
1332 	fdp = td->td_proc->p_fd;
1333 	FILEDESC_SLOCK(fdp);
1334 
1335 	/*
1336 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1337 	 * open should not be a usage error.  From a close_range() perspective,
1338 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1339 	 * be a usage error as all fd above 3 are in-fact already closed.
1340 	 */
1341 	if (highfd < lowfd) {
1342 		ret = EINVAL;
1343 		goto out;
1344 	}
1345 
1346 	/*
1347 	 * If lastfile == -1, we're dealing with either a fresh file
1348 	 * table or one in which every fd has been closed.  Just return
1349 	 * successful; there's nothing left to do.
1350 	 */
1351 	lastfile = fdlastfile(fdp);
1352 	if (lastfile == -1)
1353 		goto out;
1354 	/* Clamped to [lowfd, lastfile] */
1355 	highfd = MIN(highfd, lastfile);
1356 	for (fd = lowfd; fd <= highfd; fd++) {
1357 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1358 			FILEDESC_SUNLOCK(fdp);
1359 			(void)kern_close(td, fd);
1360 			FILEDESC_SLOCK(fdp);
1361 		}
1362 	}
1363 out:
1364 	FILEDESC_SUNLOCK(fdp);
1365 	return (ret);
1366 }
1367 
1368 #ifndef _SYS_SYSPROTO_H_
1369 struct close_range_args {
1370 	u_int	lowfd;
1371 	u_int	highfd;
1372 	int	flags;
1373 };
1374 #endif
1375 int
1376 sys_close_range(struct thread *td, struct close_range_args *uap)
1377 {
1378 
1379 	/* No flags currently defined */
1380 	if (uap->flags != 0)
1381 		return (EINVAL);
1382 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1383 }
1384 
1385 #ifdef COMPAT_FREEBSD12
1386 /*
1387  * Close open file descriptors.
1388  */
1389 #ifndef _SYS_SYSPROTO_H_
1390 struct freebsd12_closefrom_args {
1391 	int	lowfd;
1392 };
1393 #endif
1394 /* ARGSUSED */
1395 int
1396 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1397 {
1398 	u_int lowfd;
1399 
1400 	AUDIT_ARG_FD(uap->lowfd);
1401 
1402 	/*
1403 	 * Treat negative starting file descriptor values identical to
1404 	 * closefrom(0) which closes all files.
1405 	 */
1406 	lowfd = MAX(0, uap->lowfd);
1407 	return (kern_close_range(td, lowfd, ~0U));
1408 }
1409 #endif	/* COMPAT_FREEBSD12 */
1410 
1411 #if defined(COMPAT_43)
1412 /*
1413  * Return status information about a file descriptor.
1414  */
1415 #ifndef _SYS_SYSPROTO_H_
1416 struct ofstat_args {
1417 	int	fd;
1418 	struct	ostat *sb;
1419 };
1420 #endif
1421 /* ARGSUSED */
1422 int
1423 ofstat(struct thread *td, struct ofstat_args *uap)
1424 {
1425 	struct ostat oub;
1426 	struct stat ub;
1427 	int error;
1428 
1429 	error = kern_fstat(td, uap->fd, &ub);
1430 	if (error == 0) {
1431 		cvtstat(&ub, &oub);
1432 		error = copyout(&oub, uap->sb, sizeof(oub));
1433 	}
1434 	return (error);
1435 }
1436 #endif /* COMPAT_43 */
1437 
1438 #if defined(COMPAT_FREEBSD11)
1439 int
1440 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1441 {
1442 	struct stat sb;
1443 	struct freebsd11_stat osb;
1444 	int error;
1445 
1446 	error = kern_fstat(td, uap->fd, &sb);
1447 	if (error != 0)
1448 		return (error);
1449 	error = freebsd11_cvtstat(&sb, &osb);
1450 	if (error == 0)
1451 		error = copyout(&osb, uap->sb, sizeof(osb));
1452 	return (error);
1453 }
1454 #endif	/* COMPAT_FREEBSD11 */
1455 
1456 /*
1457  * Return status information about a file descriptor.
1458  */
1459 #ifndef _SYS_SYSPROTO_H_
1460 struct fstat_args {
1461 	int	fd;
1462 	struct	stat *sb;
1463 };
1464 #endif
1465 /* ARGSUSED */
1466 int
1467 sys_fstat(struct thread *td, struct fstat_args *uap)
1468 {
1469 	struct stat ub;
1470 	int error;
1471 
1472 	error = kern_fstat(td, uap->fd, &ub);
1473 	if (error == 0)
1474 		error = copyout(&ub, uap->sb, sizeof(ub));
1475 	return (error);
1476 }
1477 
1478 int
1479 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1480 {
1481 	struct file *fp;
1482 	int error;
1483 
1484 	AUDIT_ARG_FD(fd);
1485 
1486 	error = fget(td, fd, &cap_fstat_rights, &fp);
1487 	if (__predict_false(error != 0))
1488 		return (error);
1489 
1490 	AUDIT_ARG_FILE(td->td_proc, fp);
1491 
1492 	error = fo_stat(fp, sbp, td->td_ucred, td);
1493 	fdrop(fp, td);
1494 #ifdef __STAT_TIME_T_EXT
1495 	sbp->st_atim_ext = 0;
1496 	sbp->st_mtim_ext = 0;
1497 	sbp->st_ctim_ext = 0;
1498 	sbp->st_btim_ext = 0;
1499 #endif
1500 #ifdef KTRACE
1501 	if (KTRPOINT(td, KTR_STRUCT))
1502 		ktrstat_error(sbp, error);
1503 #endif
1504 	return (error);
1505 }
1506 
1507 #if defined(COMPAT_FREEBSD11)
1508 /*
1509  * Return status information about a file descriptor.
1510  */
1511 #ifndef _SYS_SYSPROTO_H_
1512 struct freebsd11_nfstat_args {
1513 	int	fd;
1514 	struct	nstat *sb;
1515 };
1516 #endif
1517 /* ARGSUSED */
1518 int
1519 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1520 {
1521 	struct nstat nub;
1522 	struct stat ub;
1523 	int error;
1524 
1525 	error = kern_fstat(td, uap->fd, &ub);
1526 	if (error == 0) {
1527 		freebsd11_cvtnstat(&ub, &nub);
1528 		error = copyout(&nub, uap->sb, sizeof(nub));
1529 	}
1530 	return (error);
1531 }
1532 #endif /* COMPAT_FREEBSD11 */
1533 
1534 /*
1535  * Return pathconf information about a file descriptor.
1536  */
1537 #ifndef _SYS_SYSPROTO_H_
1538 struct fpathconf_args {
1539 	int	fd;
1540 	int	name;
1541 };
1542 #endif
1543 /* ARGSUSED */
1544 int
1545 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1546 {
1547 	long value;
1548 	int error;
1549 
1550 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1551 	if (error == 0)
1552 		td->td_retval[0] = value;
1553 	return (error);
1554 }
1555 
1556 int
1557 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1558 {
1559 	struct file *fp;
1560 	struct vnode *vp;
1561 	int error;
1562 
1563 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1564 	if (error != 0)
1565 		return (error);
1566 
1567 	if (name == _PC_ASYNC_IO) {
1568 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1569 		goto out;
1570 	}
1571 	vp = fp->f_vnode;
1572 	if (vp != NULL) {
1573 		vn_lock(vp, LK_SHARED | LK_RETRY);
1574 		error = VOP_PATHCONF(vp, name, valuep);
1575 		VOP_UNLOCK(vp);
1576 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1577 		if (name != _PC_PIPE_BUF) {
1578 			error = EINVAL;
1579 		} else {
1580 			*valuep = PIPE_BUF;
1581 			error = 0;
1582 		}
1583 	} else {
1584 		error = EOPNOTSUPP;
1585 	}
1586 out:
1587 	fdrop(fp, td);
1588 	return (error);
1589 }
1590 
1591 /*
1592  * Copy filecaps structure allocating memory for ioctls array if needed.
1593  *
1594  * The last parameter indicates whether the fdtable is locked. If it is not and
1595  * ioctls are encountered, copying fails and the caller must lock the table.
1596  *
1597  * Note that if the table was not locked, the caller has to check the relevant
1598  * sequence counter to determine whether the operation was successful.
1599  */
1600 bool
1601 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1602 {
1603 	size_t size;
1604 
1605 	if (src->fc_ioctls != NULL && !locked)
1606 		return (false);
1607 	memcpy(dst, src, sizeof(*src));
1608 	if (src->fc_ioctls == NULL)
1609 		return (true);
1610 
1611 	KASSERT(src->fc_nioctls > 0,
1612 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1613 
1614 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1615 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1616 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1617 	return (true);
1618 }
1619 
1620 static u_long *
1621 filecaps_copy_prep(const struct filecaps *src)
1622 {
1623 	u_long *ioctls;
1624 	size_t size;
1625 
1626 	if (__predict_true(src->fc_ioctls == NULL))
1627 		return (NULL);
1628 
1629 	KASSERT(src->fc_nioctls > 0,
1630 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1631 
1632 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1633 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1634 	return (ioctls);
1635 }
1636 
1637 static void
1638 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1639     u_long *ioctls)
1640 {
1641 	size_t size;
1642 
1643 	*dst = *src;
1644 	if (__predict_true(src->fc_ioctls == NULL)) {
1645 		MPASS(ioctls == NULL);
1646 		return;
1647 	}
1648 
1649 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1650 	dst->fc_ioctls = ioctls;
1651 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1652 }
1653 
1654 /*
1655  * Move filecaps structure to the new place and clear the old place.
1656  */
1657 void
1658 filecaps_move(struct filecaps *src, struct filecaps *dst)
1659 {
1660 
1661 	*dst = *src;
1662 	bzero(src, sizeof(*src));
1663 }
1664 
1665 /*
1666  * Fill the given filecaps structure with full rights.
1667  */
1668 static void
1669 filecaps_fill(struct filecaps *fcaps)
1670 {
1671 
1672 	CAP_ALL(&fcaps->fc_rights);
1673 	fcaps->fc_ioctls = NULL;
1674 	fcaps->fc_nioctls = -1;
1675 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1676 }
1677 
1678 /*
1679  * Free memory allocated within filecaps structure.
1680  */
1681 void
1682 filecaps_free(struct filecaps *fcaps)
1683 {
1684 
1685 	free(fcaps->fc_ioctls, M_FILECAPS);
1686 	bzero(fcaps, sizeof(*fcaps));
1687 }
1688 
1689 static u_long *
1690 filecaps_free_prep(struct filecaps *fcaps)
1691 {
1692 	u_long *ioctls;
1693 
1694 	ioctls = fcaps->fc_ioctls;
1695 	bzero(fcaps, sizeof(*fcaps));
1696 	return (ioctls);
1697 }
1698 
1699 static void
1700 filecaps_free_finish(u_long *ioctls)
1701 {
1702 
1703 	free(ioctls, M_FILECAPS);
1704 }
1705 
1706 /*
1707  * Validate the given filecaps structure.
1708  */
1709 static void
1710 filecaps_validate(const struct filecaps *fcaps, const char *func)
1711 {
1712 
1713 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1714 	    ("%s: invalid rights", func));
1715 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1716 	    ("%s: invalid fcntls", func));
1717 	KASSERT(fcaps->fc_fcntls == 0 ||
1718 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1719 	    ("%s: fcntls without CAP_FCNTL", func));
1720 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1721 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1722 	    ("%s: invalid ioctls", func));
1723 	KASSERT(fcaps->fc_nioctls == 0 ||
1724 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1725 	    ("%s: ioctls without CAP_IOCTL", func));
1726 }
1727 
1728 static void
1729 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1730 {
1731 	int nfd1;
1732 
1733 	FILEDESC_XLOCK_ASSERT(fdp);
1734 
1735 	nfd1 = fdp->fd_nfiles * 2;
1736 	if (nfd1 < nfd)
1737 		nfd1 = nfd;
1738 	fdgrowtable(fdp, nfd1);
1739 }
1740 
1741 /*
1742  * Grow the file table to accommodate (at least) nfd descriptors.
1743  */
1744 static void
1745 fdgrowtable(struct filedesc *fdp, int nfd)
1746 {
1747 	struct filedesc0 *fdp0;
1748 	struct freetable *ft;
1749 	struct fdescenttbl *ntable;
1750 	struct fdescenttbl *otable;
1751 	int nnfiles, onfiles;
1752 	NDSLOTTYPE *nmap, *omap;
1753 
1754 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1755 
1756 	/* save old values */
1757 	onfiles = fdp->fd_nfiles;
1758 	otable = fdp->fd_files;
1759 	omap = fdp->fd_map;
1760 
1761 	/* compute the size of the new table */
1762 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1763 	if (nnfiles <= onfiles)
1764 		/* the table is already large enough */
1765 		return;
1766 
1767 	/*
1768 	 * Allocate a new table.  We need enough space for the number of
1769 	 * entries, file entries themselves and the struct freetable we will use
1770 	 * when we decommission the table and place it on the freelist.
1771 	 * We place the struct freetable in the middle so we don't have
1772 	 * to worry about padding.
1773 	 */
1774 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1775 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1776 	    sizeof(struct freetable),
1777 	    M_FILEDESC, M_ZERO | M_WAITOK);
1778 	/* copy the old data */
1779 	ntable->fdt_nfiles = nnfiles;
1780 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1781 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1782 
1783 	/*
1784 	 * Allocate a new map only if the old is not large enough.  It will
1785 	 * grow at a slower rate than the table as it can map more
1786 	 * entries than the table can hold.
1787 	 */
1788 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1789 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1790 		    M_ZERO | M_WAITOK);
1791 		/* copy over the old data and update the pointer */
1792 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1793 		fdp->fd_map = nmap;
1794 	}
1795 
1796 	/*
1797 	 * Make sure that ntable is correctly initialized before we replace
1798 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1799 	 * data.
1800 	 */
1801 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1802 
1803 	/*
1804 	 * Do not free the old file table, as some threads may still
1805 	 * reference entries within it.  Instead, place it on a freelist
1806 	 * which will be processed when the struct filedesc is released.
1807 	 *
1808 	 * Note that if onfiles == NDFILE, we're dealing with the original
1809 	 * static allocation contained within (struct filedesc0 *)fdp,
1810 	 * which must not be freed.
1811 	 */
1812 	if (onfiles > NDFILE) {
1813 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1814 		fdp0 = (struct filedesc0 *)fdp;
1815 		ft->ft_table = otable;
1816 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1817 	}
1818 	/*
1819 	 * The map does not have the same possibility of threads still
1820 	 * holding references to it.  So always free it as long as it
1821 	 * does not reference the original static allocation.
1822 	 */
1823 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1824 		free(omap, M_FILEDESC);
1825 }
1826 
1827 /*
1828  * Allocate a file descriptor for the process.
1829  */
1830 int
1831 fdalloc(struct thread *td, int minfd, int *result)
1832 {
1833 	struct proc *p = td->td_proc;
1834 	struct filedesc *fdp = p->p_fd;
1835 	int fd, maxfd, allocfd;
1836 #ifdef RACCT
1837 	int error;
1838 #endif
1839 
1840 	FILEDESC_XLOCK_ASSERT(fdp);
1841 
1842 	if (fdp->fd_freefile > minfd)
1843 		minfd = fdp->fd_freefile;
1844 
1845 	maxfd = getmaxfd(td);
1846 
1847 	/*
1848 	 * Search the bitmap for a free descriptor starting at minfd.
1849 	 * If none is found, grow the file table.
1850 	 */
1851 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1852 	if (__predict_false(fd >= maxfd))
1853 		return (EMFILE);
1854 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1855 		allocfd = min(fd * 2, maxfd);
1856 #ifdef RACCT
1857 		if (RACCT_ENABLED()) {
1858 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1859 			if (error != 0)
1860 				return (EMFILE);
1861 		}
1862 #endif
1863 		/*
1864 		 * fd is already equal to first free descriptor >= minfd, so
1865 		 * we only need to grow the table and we are done.
1866 		 */
1867 		fdgrowtable_exp(fdp, allocfd);
1868 	}
1869 
1870 	/*
1871 	 * Perform some sanity checks, then mark the file descriptor as
1872 	 * used and return it to the caller.
1873 	 */
1874 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1875 	    ("invalid descriptor %d", fd));
1876 	KASSERT(!fdisused(fdp, fd),
1877 	    ("fd_first_free() returned non-free descriptor"));
1878 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1879 	    ("file descriptor isn't free"));
1880 	fdused(fdp, fd);
1881 	*result = fd;
1882 	return (0);
1883 }
1884 
1885 /*
1886  * Allocate n file descriptors for the process.
1887  */
1888 int
1889 fdallocn(struct thread *td, int minfd, int *fds, int n)
1890 {
1891 	struct proc *p = td->td_proc;
1892 	struct filedesc *fdp = p->p_fd;
1893 	int i;
1894 
1895 	FILEDESC_XLOCK_ASSERT(fdp);
1896 
1897 	for (i = 0; i < n; i++)
1898 		if (fdalloc(td, 0, &fds[i]) != 0)
1899 			break;
1900 
1901 	if (i < n) {
1902 		for (i--; i >= 0; i--)
1903 			fdunused(fdp, fds[i]);
1904 		return (EMFILE);
1905 	}
1906 
1907 	return (0);
1908 }
1909 
1910 /*
1911  * Create a new open file structure and allocate a file descriptor for the
1912  * process that refers to it.  We add one reference to the file for the
1913  * descriptor table and one reference for resultfp. This is to prevent us
1914  * being preempted and the entry in the descriptor table closed after we
1915  * release the FILEDESC lock.
1916  */
1917 int
1918 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1919     struct filecaps *fcaps)
1920 {
1921 	struct file *fp;
1922 	int error, fd;
1923 
1924 	error = falloc_noinstall(td, &fp);
1925 	if (error)
1926 		return (error);		/* no reference held on error */
1927 
1928 	error = finstall(td, fp, &fd, flags, fcaps);
1929 	if (error) {
1930 		fdrop(fp, td);		/* one reference (fp only) */
1931 		return (error);
1932 	}
1933 
1934 	if (resultfp != NULL)
1935 		*resultfp = fp;		/* copy out result */
1936 	else
1937 		fdrop(fp, td);		/* release local reference */
1938 
1939 	if (resultfd != NULL)
1940 		*resultfd = fd;
1941 
1942 	return (0);
1943 }
1944 
1945 /*
1946  * Create a new open file structure without allocating a file descriptor.
1947  */
1948 int
1949 falloc_noinstall(struct thread *td, struct file **resultfp)
1950 {
1951 	struct file *fp;
1952 	int maxuserfiles = maxfiles - (maxfiles / 20);
1953 	int openfiles_new;
1954 	static struct timeval lastfail;
1955 	static int curfail;
1956 
1957 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1958 
1959 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1960 	if ((openfiles_new >= maxuserfiles &&
1961 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1962 	    openfiles_new >= maxfiles) {
1963 		atomic_subtract_int(&openfiles, 1);
1964 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1965 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1966 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1967 		}
1968 		return (ENFILE);
1969 	}
1970 	fp = uma_zalloc(file_zone, M_WAITOK);
1971 	bzero(fp, sizeof(*fp));
1972 	refcount_init(&fp->f_count, 1);
1973 	fp->f_cred = crhold(td->td_ucred);
1974 	fp->f_ops = &badfileops;
1975 	*resultfp = fp;
1976 	return (0);
1977 }
1978 
1979 /*
1980  * Install a file in a file descriptor table.
1981  */
1982 void
1983 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1984     struct filecaps *fcaps)
1985 {
1986 	struct filedescent *fde;
1987 
1988 	MPASS(fp != NULL);
1989 	if (fcaps != NULL)
1990 		filecaps_validate(fcaps, __func__);
1991 	FILEDESC_XLOCK_ASSERT(fdp);
1992 
1993 	fde = &fdp->fd_ofiles[fd];
1994 #ifdef CAPABILITIES
1995 	seqc_write_begin(&fde->fde_seqc);
1996 #endif
1997 	fde->fde_file = fp;
1998 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1999 	if (fcaps != NULL)
2000 		filecaps_move(fcaps, &fde->fde_caps);
2001 	else
2002 		filecaps_fill(&fde->fde_caps);
2003 #ifdef CAPABILITIES
2004 	seqc_write_end(&fde->fde_seqc);
2005 #endif
2006 }
2007 
2008 int
2009 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2010     struct filecaps *fcaps)
2011 {
2012 	struct filedesc *fdp = td->td_proc->p_fd;
2013 	int error;
2014 
2015 	MPASS(fd != NULL);
2016 
2017 	if (!fhold(fp))
2018 		return (EBADF);
2019 	FILEDESC_XLOCK(fdp);
2020 	error = fdalloc(td, 0, fd);
2021 	if (__predict_false(error != 0)) {
2022 		FILEDESC_XUNLOCK(fdp);
2023 		fdrop(fp, td);
2024 		return (error);
2025 	}
2026 	_finstall(fdp, fp, *fd, flags, fcaps);
2027 	FILEDESC_XUNLOCK(fdp);
2028 	return (0);
2029 }
2030 
2031 /*
2032  * Build a new filedesc structure from another.
2033  *
2034  * If fdp is not NULL, return with it shared locked.
2035  */
2036 struct filedesc *
2037 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2038 {
2039 	struct filedesc0 *newfdp0;
2040 	struct filedesc *newfdp;
2041 
2042 	if (prepfiles)
2043 		MPASS(lastfile != NULL);
2044 	else
2045 		MPASS(lastfile == NULL);
2046 
2047 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2048 	newfdp = &newfdp0->fd_fd;
2049 
2050 	/* Create the file descriptor table. */
2051 	FILEDESC_LOCK_INIT(newfdp);
2052 	refcount_init(&newfdp->fd_refcnt, 1);
2053 	refcount_init(&newfdp->fd_holdcnt, 1);
2054 	newfdp->fd_map = newfdp0->fd_dmap;
2055 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2056 	newfdp->fd_files->fdt_nfiles = NDFILE;
2057 
2058 	if (fdp == NULL)
2059 		return (newfdp);
2060 
2061 	FILEDESC_SLOCK(fdp);
2062 	if (!prepfiles) {
2063 		FILEDESC_SUNLOCK(fdp);
2064 		return (newfdp);
2065 	}
2066 
2067 	for (;;) {
2068 		*lastfile = fdlastfile(fdp);
2069 		if (*lastfile < newfdp->fd_nfiles)
2070 			break;
2071 		FILEDESC_SUNLOCK(fdp);
2072 		fdgrowtable(newfdp, *lastfile + 1);
2073 		FILEDESC_SLOCK(fdp);
2074 	}
2075 
2076 	return (newfdp);
2077 }
2078 
2079 /*
2080  * Build a pwddesc structure from another.
2081  * Copy the current, root, and jail root vnode references.
2082  *
2083  * If pdp is not NULL, return with it shared locked.
2084  */
2085 struct pwddesc *
2086 pdinit(struct pwddesc *pdp, bool keeplock)
2087 {
2088 	struct pwddesc *newpdp;
2089 	struct pwd *newpwd;
2090 
2091 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2092 
2093 	PWDDESC_LOCK_INIT(newpdp);
2094 	refcount_init(&newpdp->pd_refcount, 1);
2095 	newpdp->pd_cmask = CMASK;
2096 
2097 	if (pdp == NULL) {
2098 		newpwd = pwd_alloc();
2099 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2100 		return (newpdp);
2101 	}
2102 
2103 	PWDDESC_XLOCK(pdp);
2104 	newpwd = pwd_hold_pwddesc(pdp);
2105 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2106 	if (!keeplock)
2107 		PWDDESC_XUNLOCK(pdp);
2108 	return (newpdp);
2109 }
2110 
2111 static struct filedesc *
2112 fdhold(struct proc *p)
2113 {
2114 	struct filedesc *fdp;
2115 
2116 	PROC_LOCK_ASSERT(p, MA_OWNED);
2117 	fdp = p->p_fd;
2118 	if (fdp != NULL)
2119 		refcount_acquire(&fdp->fd_holdcnt);
2120 	return (fdp);
2121 }
2122 
2123 static struct pwddesc *
2124 pdhold(struct proc *p)
2125 {
2126 	struct pwddesc *pdp;
2127 
2128 	PROC_LOCK_ASSERT(p, MA_OWNED);
2129 	pdp = p->p_pd;
2130 	if (pdp != NULL)
2131 		refcount_acquire(&pdp->pd_refcount);
2132 	return (pdp);
2133 }
2134 
2135 static void
2136 fddrop(struct filedesc *fdp)
2137 {
2138 
2139 	if (fdp->fd_holdcnt > 1) {
2140 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2141 			return;
2142 	}
2143 
2144 	FILEDESC_LOCK_DESTROY(fdp);
2145 	uma_zfree(filedesc0_zone, fdp);
2146 }
2147 
2148 static void
2149 pddrop(struct pwddesc *pdp)
2150 {
2151 	struct pwd *pwd;
2152 
2153 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2154 		return;
2155 
2156 	PWDDESC_XLOCK(pdp);
2157 	if (refcount_release(&pdp->pd_refcount) == 0) {
2158 		PWDDESC_XUNLOCK(pdp);
2159 		return;
2160 	}
2161 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2162 	pwd_set(pdp, NULL);
2163 	PWDDESC_XUNLOCK(pdp);
2164 	pwd_drop(pwd);
2165 
2166 	PWDDESC_LOCK_DESTROY(pdp);
2167 	free(pdp, M_PWDDESC);
2168 }
2169 
2170 /*
2171  * Share a filedesc structure.
2172  */
2173 struct filedesc *
2174 fdshare(struct filedesc *fdp)
2175 {
2176 
2177 	refcount_acquire(&fdp->fd_refcnt);
2178 	return (fdp);
2179 }
2180 
2181 /*
2182  * Share a pwddesc structure.
2183  */
2184 struct pwddesc *
2185 pdshare(struct pwddesc *pdp)
2186 {
2187 	refcount_acquire(&pdp->pd_refcount);
2188 	return (pdp);
2189 }
2190 
2191 /*
2192  * Unshare a filedesc structure, if necessary by making a copy
2193  */
2194 void
2195 fdunshare(struct thread *td)
2196 {
2197 	struct filedesc *tmp;
2198 	struct proc *p = td->td_proc;
2199 
2200 	if (p->p_fd->fd_refcnt == 1)
2201 		return;
2202 
2203 	tmp = fdcopy(p->p_fd);
2204 	fdescfree(td);
2205 	p->p_fd = tmp;
2206 }
2207 
2208 /*
2209  * Unshare a pwddesc structure.
2210  */
2211 void
2212 pdunshare(struct thread *td)
2213 {
2214 	struct pwddesc *pdp;
2215 	struct proc *p;
2216 
2217 	p = td->td_proc;
2218 	/* Not shared. */
2219 	if (p->p_pd->pd_refcount == 1)
2220 		return;
2221 
2222 	pdp = pdcopy(p->p_pd);
2223 	pdescfree(td);
2224 	p->p_pd = pdp;
2225 }
2226 
2227 void
2228 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2229 {
2230 
2231 	fdescfree(td);
2232 	td->td_proc->p_fd = fdp;
2233 }
2234 
2235 /*
2236  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2237  * this is to ease callers, not catch errors.
2238  */
2239 struct filedesc *
2240 fdcopy(struct filedesc *fdp)
2241 {
2242 	struct filedesc *newfdp;
2243 	struct filedescent *nfde, *ofde;
2244 	int i, lastfile;
2245 
2246 	MPASS(fdp != NULL);
2247 
2248 	newfdp = fdinit(fdp, true, &lastfile);
2249 	/* copy all passable descriptors (i.e. not kqueue) */
2250 	newfdp->fd_freefile = -1;
2251 	for (i = 0; i <= lastfile; ++i) {
2252 		ofde = &fdp->fd_ofiles[i];
2253 		if (ofde->fde_file == NULL ||
2254 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2255 		    !fhold(ofde->fde_file)) {
2256 			if (newfdp->fd_freefile == -1)
2257 				newfdp->fd_freefile = i;
2258 			continue;
2259 		}
2260 		nfde = &newfdp->fd_ofiles[i];
2261 		*nfde = *ofde;
2262 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2263 		fdused_init(newfdp, i);
2264 	}
2265 	if (newfdp->fd_freefile == -1)
2266 		newfdp->fd_freefile = i;
2267 	FILEDESC_SUNLOCK(fdp);
2268 	return (newfdp);
2269 }
2270 
2271 /*
2272  * Copy a pwddesc structure.
2273  */
2274 struct pwddesc *
2275 pdcopy(struct pwddesc *pdp)
2276 {
2277 	struct pwddesc *newpdp;
2278 
2279 	MPASS(pdp != NULL);
2280 
2281 	newpdp = pdinit(pdp, true);
2282 	newpdp->pd_cmask = pdp->pd_cmask;
2283 	PWDDESC_XUNLOCK(pdp);
2284 	return (newpdp);
2285 }
2286 
2287 /*
2288  * Copies a filedesc structure, while remapping all file descriptors
2289  * stored inside using a translation table.
2290  *
2291  * File descriptors are copied over to the new file descriptor table,
2292  * regardless of whether the close-on-exec flag is set.
2293  */
2294 int
2295 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2296     struct filedesc **ret)
2297 {
2298 	struct filedesc *newfdp;
2299 	struct filedescent *nfde, *ofde;
2300 	int error, i, lastfile;
2301 
2302 	MPASS(fdp != NULL);
2303 
2304 	newfdp = fdinit(fdp, true, &lastfile);
2305 	if (nfds > lastfile + 1) {
2306 		/* New table cannot be larger than the old one. */
2307 		error = E2BIG;
2308 		goto bad;
2309 	}
2310 	/* Copy all passable descriptors (i.e. not kqueue). */
2311 	newfdp->fd_freefile = nfds;
2312 	for (i = 0; i < nfds; ++i) {
2313 		if (fds[i] < 0 || fds[i] > lastfile) {
2314 			/* File descriptor out of bounds. */
2315 			error = EBADF;
2316 			goto bad;
2317 		}
2318 		ofde = &fdp->fd_ofiles[fds[i]];
2319 		if (ofde->fde_file == NULL) {
2320 			/* Unused file descriptor. */
2321 			error = EBADF;
2322 			goto bad;
2323 		}
2324 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2325 			/* File descriptor cannot be passed. */
2326 			error = EINVAL;
2327 			goto bad;
2328 		}
2329 		if (!fhold(ofde->fde_file)) {
2330 			error = EBADF;
2331 			goto bad;
2332 		}
2333 		nfde = &newfdp->fd_ofiles[i];
2334 		*nfde = *ofde;
2335 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2336 		fdused_init(newfdp, i);
2337 	}
2338 	FILEDESC_SUNLOCK(fdp);
2339 	*ret = newfdp;
2340 	return (0);
2341 bad:
2342 	FILEDESC_SUNLOCK(fdp);
2343 	fdescfree_remapped(newfdp);
2344 	return (error);
2345 }
2346 
2347 /*
2348  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2349  * one of processes using it exits) and the table used to be shared.
2350  */
2351 static void
2352 fdclearlocks(struct thread *td)
2353 {
2354 	struct filedesc *fdp;
2355 	struct filedesc_to_leader *fdtol;
2356 	struct flock lf;
2357 	struct file *fp;
2358 	struct proc *p;
2359 	struct vnode *vp;
2360 	int i, lastfile;
2361 
2362 	p = td->td_proc;
2363 	fdp = p->p_fd;
2364 	fdtol = p->p_fdtol;
2365 	MPASS(fdtol != NULL);
2366 
2367 	FILEDESC_XLOCK(fdp);
2368 	KASSERT(fdtol->fdl_refcount > 0,
2369 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2370 	    fdtol->fdl_refcount));
2371 	if (fdtol->fdl_refcount == 1 &&
2372 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2373 		lastfile = fdlastfile(fdp);
2374 		for (i = 0; i <= lastfile; i++) {
2375 			fp = fdp->fd_ofiles[i].fde_file;
2376 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2377 			    !fhold(fp))
2378 				continue;
2379 			FILEDESC_XUNLOCK(fdp);
2380 			lf.l_whence = SEEK_SET;
2381 			lf.l_start = 0;
2382 			lf.l_len = 0;
2383 			lf.l_type = F_UNLCK;
2384 			vp = fp->f_vnode;
2385 			(void) VOP_ADVLOCK(vp,
2386 			    (caddr_t)p->p_leader, F_UNLCK,
2387 			    &lf, F_POSIX);
2388 			FILEDESC_XLOCK(fdp);
2389 			fdrop(fp, td);
2390 		}
2391 	}
2392 retry:
2393 	if (fdtol->fdl_refcount == 1) {
2394 		if (fdp->fd_holdleaderscount > 0 &&
2395 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2396 			/*
2397 			 * close() or kern_dup() has cleared a reference
2398 			 * in a shared file descriptor table.
2399 			 */
2400 			fdp->fd_holdleaderswakeup = 1;
2401 			sx_sleep(&fdp->fd_holdleaderscount,
2402 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2403 			goto retry;
2404 		}
2405 		if (fdtol->fdl_holdcount > 0) {
2406 			/*
2407 			 * Ensure that fdtol->fdl_leader remains
2408 			 * valid in closef().
2409 			 */
2410 			fdtol->fdl_wakeup = 1;
2411 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2412 			    "fdlhold", 0);
2413 			goto retry;
2414 		}
2415 	}
2416 	fdtol->fdl_refcount--;
2417 	if (fdtol->fdl_refcount == 0 &&
2418 	    fdtol->fdl_holdcount == 0) {
2419 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2420 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2421 	} else
2422 		fdtol = NULL;
2423 	p->p_fdtol = NULL;
2424 	FILEDESC_XUNLOCK(fdp);
2425 	if (fdtol != NULL)
2426 		free(fdtol, M_FILEDESC_TO_LEADER);
2427 }
2428 
2429 /*
2430  * Release a filedesc structure.
2431  */
2432 static void
2433 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2434 {
2435 	struct filedesc0 *fdp0;
2436 	struct freetable *ft, *tft;
2437 	struct filedescent *fde;
2438 	struct file *fp;
2439 	int i, lastfile;
2440 
2441 	lastfile = fdlastfile_single(fdp);
2442 	for (i = 0; i <= lastfile; i++) {
2443 		fde = &fdp->fd_ofiles[i];
2444 		fp = fde->fde_file;
2445 		if (fp != NULL) {
2446 			fdefree_last(fde);
2447 			if (needclose)
2448 				(void) closef(fp, td);
2449 			else
2450 				fdrop(fp, td);
2451 		}
2452 	}
2453 
2454 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2455 		free(fdp->fd_map, M_FILEDESC);
2456 	if (fdp->fd_nfiles > NDFILE)
2457 		free(fdp->fd_files, M_FILEDESC);
2458 
2459 	fdp0 = (struct filedesc0 *)fdp;
2460 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2461 		free(ft->ft_table, M_FILEDESC);
2462 
2463 	fddrop(fdp);
2464 }
2465 
2466 void
2467 fdescfree(struct thread *td)
2468 {
2469 	struct proc *p;
2470 	struct filedesc *fdp;
2471 
2472 	p = td->td_proc;
2473 	fdp = p->p_fd;
2474 	MPASS(fdp != NULL);
2475 
2476 #ifdef RACCT
2477 	if (RACCT_ENABLED())
2478 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2479 #endif
2480 
2481 	if (p->p_fdtol != NULL)
2482 		fdclearlocks(td);
2483 
2484 	PROC_LOCK(p);
2485 	p->p_fd = NULL;
2486 	PROC_UNLOCK(p);
2487 
2488 	if (refcount_release(&fdp->fd_refcnt) == 0)
2489 		return;
2490 
2491 	fdescfree_fds(td, fdp, 1);
2492 }
2493 
2494 void
2495 pdescfree(struct thread *td)
2496 {
2497 	struct proc *p;
2498 	struct pwddesc *pdp;
2499 
2500 	p = td->td_proc;
2501 	pdp = p->p_pd;
2502 	MPASS(pdp != NULL);
2503 
2504 	PROC_LOCK(p);
2505 	p->p_pd = NULL;
2506 	PROC_UNLOCK(p);
2507 
2508 	pddrop(pdp);
2509 }
2510 
2511 void
2512 fdescfree_remapped(struct filedesc *fdp)
2513 {
2514 	fdescfree_fds(curthread, fdp, 0);
2515 }
2516 
2517 /*
2518  * For setugid programs, we don't want to people to use that setugidness
2519  * to generate error messages which write to a file which otherwise would
2520  * otherwise be off-limits to the process.  We check for filesystems where
2521  * the vnode can change out from under us after execve (like [lin]procfs).
2522  *
2523  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2524  * sufficient.  We also don't check for setugidness since we know we are.
2525  */
2526 static bool
2527 is_unsafe(struct file *fp)
2528 {
2529 	struct vnode *vp;
2530 
2531 	if (fp->f_type != DTYPE_VNODE)
2532 		return (false);
2533 
2534 	vp = fp->f_vnode;
2535 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2536 }
2537 
2538 /*
2539  * Make this setguid thing safe, if at all possible.
2540  */
2541 void
2542 fdsetugidsafety(struct thread *td)
2543 {
2544 	struct filedesc *fdp;
2545 	struct file *fp;
2546 	int i;
2547 
2548 	fdp = td->td_proc->p_fd;
2549 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2550 	MPASS(fdp->fd_nfiles >= 3);
2551 	for (i = 0; i <= 2; i++) {
2552 		fp = fdp->fd_ofiles[i].fde_file;
2553 		if (fp != NULL && is_unsafe(fp)) {
2554 			FILEDESC_XLOCK(fdp);
2555 			knote_fdclose(td, i);
2556 			/*
2557 			 * NULL-out descriptor prior to close to avoid
2558 			 * a race while close blocks.
2559 			 */
2560 			fdfree(fdp, i);
2561 			FILEDESC_XUNLOCK(fdp);
2562 			(void) closef(fp, td);
2563 		}
2564 	}
2565 }
2566 
2567 /*
2568  * If a specific file object occupies a specific file descriptor, close the
2569  * file descriptor entry and drop a reference on the file object.  This is a
2570  * convenience function to handle a subsequent error in a function that calls
2571  * falloc() that handles the race that another thread might have closed the
2572  * file descriptor out from under the thread creating the file object.
2573  */
2574 void
2575 fdclose(struct thread *td, struct file *fp, int idx)
2576 {
2577 	struct filedesc *fdp = td->td_proc->p_fd;
2578 
2579 	FILEDESC_XLOCK(fdp);
2580 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2581 		fdfree(fdp, idx);
2582 		FILEDESC_XUNLOCK(fdp);
2583 		fdrop(fp, td);
2584 	} else
2585 		FILEDESC_XUNLOCK(fdp);
2586 }
2587 
2588 /*
2589  * Close any files on exec?
2590  */
2591 void
2592 fdcloseexec(struct thread *td)
2593 {
2594 	struct filedesc *fdp;
2595 	struct filedescent *fde;
2596 	struct file *fp;
2597 	int i, lastfile;
2598 
2599 	fdp = td->td_proc->p_fd;
2600 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2601 	lastfile = fdlastfile_single(fdp);
2602 	for (i = 0; i <= lastfile; i++) {
2603 		fde = &fdp->fd_ofiles[i];
2604 		fp = fde->fde_file;
2605 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2606 		    (fde->fde_flags & UF_EXCLOSE))) {
2607 			FILEDESC_XLOCK(fdp);
2608 			fdfree(fdp, i);
2609 			(void) closefp(fdp, i, fp, td, 0);
2610 			FILEDESC_UNLOCK_ASSERT(fdp);
2611 		}
2612 	}
2613 }
2614 
2615 /*
2616  * It is unsafe for set[ug]id processes to be started with file
2617  * descriptors 0..2 closed, as these descriptors are given implicit
2618  * significance in the Standard C library.  fdcheckstd() will create a
2619  * descriptor referencing /dev/null for each of stdin, stdout, and
2620  * stderr that is not already open.
2621  */
2622 int
2623 fdcheckstd(struct thread *td)
2624 {
2625 	struct filedesc *fdp;
2626 	register_t save;
2627 	int i, error, devnull;
2628 
2629 	fdp = td->td_proc->p_fd;
2630 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2631 	MPASS(fdp->fd_nfiles >= 3);
2632 	devnull = -1;
2633 	for (i = 0; i <= 2; i++) {
2634 		if (fdp->fd_ofiles[i].fde_file != NULL)
2635 			continue;
2636 
2637 		save = td->td_retval[0];
2638 		if (devnull != -1) {
2639 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2640 		} else {
2641 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2642 			    UIO_SYSSPACE, O_RDWR, 0);
2643 			if (error == 0) {
2644 				devnull = td->td_retval[0];
2645 				KASSERT(devnull == i, ("we didn't get our fd"));
2646 			}
2647 		}
2648 		td->td_retval[0] = save;
2649 		if (error != 0)
2650 			return (error);
2651 	}
2652 	return (0);
2653 }
2654 
2655 /*
2656  * Internal form of close.  Decrement reference count on file structure.
2657  * Note: td may be NULL when closing a file that was being passed in a
2658  * message.
2659  */
2660 int
2661 closef(struct file *fp, struct thread *td)
2662 {
2663 	struct vnode *vp;
2664 	struct flock lf;
2665 	struct filedesc_to_leader *fdtol;
2666 	struct filedesc *fdp;
2667 
2668 	/*
2669 	 * POSIX record locking dictates that any close releases ALL
2670 	 * locks owned by this process.  This is handled by setting
2671 	 * a flag in the unlock to free ONLY locks obeying POSIX
2672 	 * semantics, and not to free BSD-style file locks.
2673 	 * If the descriptor was in a message, POSIX-style locks
2674 	 * aren't passed with the descriptor, and the thread pointer
2675 	 * will be NULL.  Callers should be careful only to pass a
2676 	 * NULL thread pointer when there really is no owning
2677 	 * context that might have locks, or the locks will be
2678 	 * leaked.
2679 	 */
2680 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2681 		vp = fp->f_vnode;
2682 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2683 			lf.l_whence = SEEK_SET;
2684 			lf.l_start = 0;
2685 			lf.l_len = 0;
2686 			lf.l_type = F_UNLCK;
2687 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2688 			    F_UNLCK, &lf, F_POSIX);
2689 		}
2690 		fdtol = td->td_proc->p_fdtol;
2691 		if (fdtol != NULL) {
2692 			/*
2693 			 * Handle special case where file descriptor table is
2694 			 * shared between multiple process leaders.
2695 			 */
2696 			fdp = td->td_proc->p_fd;
2697 			FILEDESC_XLOCK(fdp);
2698 			for (fdtol = fdtol->fdl_next;
2699 			    fdtol != td->td_proc->p_fdtol;
2700 			    fdtol = fdtol->fdl_next) {
2701 				if ((fdtol->fdl_leader->p_flag &
2702 				    P_ADVLOCK) == 0)
2703 					continue;
2704 				fdtol->fdl_holdcount++;
2705 				FILEDESC_XUNLOCK(fdp);
2706 				lf.l_whence = SEEK_SET;
2707 				lf.l_start = 0;
2708 				lf.l_len = 0;
2709 				lf.l_type = F_UNLCK;
2710 				vp = fp->f_vnode;
2711 				(void) VOP_ADVLOCK(vp,
2712 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2713 				    F_POSIX);
2714 				FILEDESC_XLOCK(fdp);
2715 				fdtol->fdl_holdcount--;
2716 				if (fdtol->fdl_holdcount == 0 &&
2717 				    fdtol->fdl_wakeup != 0) {
2718 					fdtol->fdl_wakeup = 0;
2719 					wakeup(fdtol);
2720 				}
2721 			}
2722 			FILEDESC_XUNLOCK(fdp);
2723 		}
2724 	}
2725 	return (fdrop(fp, td));
2726 }
2727 
2728 /*
2729  * Initialize the file pointer with the specified properties.
2730  *
2731  * The ops are set with release semantics to be certain that the flags, type,
2732  * and data are visible when ops is.  This is to prevent ops methods from being
2733  * called with bad data.
2734  */
2735 void
2736 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2737 {
2738 	fp->f_data = data;
2739 	fp->f_flag = flag;
2740 	fp->f_type = type;
2741 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2742 }
2743 
2744 void
2745 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2746 {
2747 	fp->f_seqcount[UIO_READ] = 1;
2748 	fp->f_seqcount[UIO_WRITE] = 1;
2749 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2750 	    data, ops);
2751 }
2752 
2753 int
2754 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2755     struct file **fpp, struct filecaps *havecapsp)
2756 {
2757 	struct filedescent *fde;
2758 	int error;
2759 
2760 	FILEDESC_LOCK_ASSERT(fdp);
2761 
2762 	fde = fdeget_locked(fdp, fd);
2763 	if (fde == NULL) {
2764 		error = EBADF;
2765 		goto out;
2766 	}
2767 
2768 #ifdef CAPABILITIES
2769 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2770 	if (error != 0)
2771 		goto out;
2772 #endif
2773 
2774 	if (havecapsp != NULL)
2775 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2776 
2777 	*fpp = fde->fde_file;
2778 
2779 	error = 0;
2780 out:
2781 	return (error);
2782 }
2783 
2784 int
2785 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2786     struct file **fpp, struct filecaps *havecapsp)
2787 {
2788 	struct filedesc *fdp = td->td_proc->p_fd;
2789 	int error;
2790 #ifndef CAPABILITIES
2791 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2792 	if (havecapsp != NULL && error == 0)
2793 		filecaps_fill(havecapsp);
2794 #else
2795 	struct file *fp;
2796 	seqc_t seq;
2797 
2798 	*fpp = NULL;
2799 	for (;;) {
2800 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2801 		if (error != 0)
2802 			return (error);
2803 
2804 		if (havecapsp != NULL) {
2805 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2806 			    havecapsp, false)) {
2807 				fdrop(fp, td);
2808 				goto get_locked;
2809 			}
2810 		}
2811 
2812 		if (!fd_modified(fdp, fd, seq))
2813 			break;
2814 		fdrop(fp, td);
2815 	}
2816 
2817 	*fpp = fp;
2818 	return (0);
2819 
2820 get_locked:
2821 	FILEDESC_SLOCK(fdp);
2822 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2823 	if (error == 0 && !fhold(*fpp))
2824 		error = EBADF;
2825 	FILEDESC_SUNLOCK(fdp);
2826 #endif
2827 	return (error);
2828 }
2829 
2830 #ifdef CAPABILITIES
2831 int
2832 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2833 {
2834 	const struct filedescent *fde;
2835 	const struct fdescenttbl *fdt;
2836 	struct filedesc *fdp;
2837 	struct file *fp;
2838 	struct vnode *vp;
2839 	const cap_rights_t *haverights;
2840 	cap_rights_t rights;
2841 	seqc_t seq;
2842 
2843 	VFS_SMR_ASSERT_ENTERED();
2844 
2845 	rights = *ndp->ni_rightsneeded;
2846 	cap_rights_set_one(&rights, CAP_LOOKUP);
2847 
2848 	fdp = curproc->p_fd;
2849 	fdt = fdp->fd_files;
2850 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2851 		return (EBADF);
2852 	seq = seqc_read_any(fd_seqc(fdt, fd));
2853 	if (__predict_false(seqc_in_modify(seq)))
2854 		return (EAGAIN);
2855 	fde = &fdt->fdt_ofiles[fd];
2856 	haverights = cap_rights_fde_inline(fde);
2857 	fp = fde->fde_file;
2858 	if (__predict_false(fp == NULL))
2859 		return (EAGAIN);
2860 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2861 		return (EAGAIN);
2862 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2863 	vp = fp->f_vnode;
2864 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2865 		return (EAGAIN);
2866 	}
2867 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2868 		return (EAGAIN);
2869 	}
2870 	/*
2871 	 * Use an acquire barrier to force re-reading of fdt so it is
2872 	 * refreshed for verification.
2873 	 */
2874 	atomic_thread_fence_acq();
2875 	fdt = fdp->fd_files;
2876 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2877 		return (EAGAIN);
2878 	/*
2879 	 * If file descriptor doesn't have all rights,
2880 	 * all lookups relative to it must also be
2881 	 * strictly relative.
2882 	 *
2883 	 * Not yet supported by fast path.
2884 	 */
2885 	CAP_ALL(&rights);
2886 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2887 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2888 	    ndp->ni_filecaps.fc_nioctls != -1) {
2889 #ifdef notyet
2890 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2891 #else
2892 		return (EAGAIN);
2893 #endif
2894 	}
2895 	*vpp = vp;
2896 	return (0);
2897 }
2898 #else
2899 int
2900 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2901 {
2902 	const struct fdescenttbl *fdt;
2903 	struct filedesc *fdp;
2904 	struct file *fp;
2905 	struct vnode *vp;
2906 
2907 	VFS_SMR_ASSERT_ENTERED();
2908 
2909 	fdp = curproc->p_fd;
2910 	fdt = fdp->fd_files;
2911 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2912 		return (EBADF);
2913 	fp = fdt->fdt_ofiles[fd].fde_file;
2914 	if (__predict_false(fp == NULL))
2915 		return (EAGAIN);
2916 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2917 	vp = fp->f_vnode;
2918 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2919 		return (EAGAIN);
2920 	}
2921 	/*
2922 	 * Use an acquire barrier to force re-reading of fdt so it is
2923 	 * refreshed for verification.
2924 	 */
2925 	atomic_thread_fence_acq();
2926 	fdt = fdp->fd_files;
2927 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2928 		return (EAGAIN);
2929 	filecaps_fill(&ndp->ni_filecaps);
2930 	*vpp = vp;
2931 	return (0);
2932 }
2933 #endif
2934 
2935 int
2936 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2937     struct file **fpp, seqc_t *seqp)
2938 {
2939 #ifdef CAPABILITIES
2940 	const struct filedescent *fde;
2941 #endif
2942 	const struct fdescenttbl *fdt;
2943 	struct file *fp;
2944 #ifdef CAPABILITIES
2945 	seqc_t seq;
2946 	cap_rights_t haverights;
2947 	int error;
2948 #endif
2949 
2950 	fdt = fdp->fd_files;
2951 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2952 		return (EBADF);
2953 	/*
2954 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2955 	 * never raising a refcount above 0.  To accomplish this we have
2956 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2957 	 * must be re-verified once we acquire a reference to be certain
2958 	 * that the identity is still correct and we did not lose a race
2959 	 * due to preemption.
2960 	 */
2961 	for (;;) {
2962 #ifdef CAPABILITIES
2963 		seq = seqc_read(fd_seqc(fdt, fd));
2964 		fde = &fdt->fdt_ofiles[fd];
2965 		haverights = *cap_rights_fde_inline(fde);
2966 		fp = fde->fde_file;
2967 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2968 			continue;
2969 #else
2970 		fp = fdt->fdt_ofiles[fd].fde_file;
2971 #endif
2972 		if (fp == NULL)
2973 			return (EBADF);
2974 #ifdef CAPABILITIES
2975 		error = cap_check_inline(&haverights, needrightsp);
2976 		if (error != 0)
2977 			return (error);
2978 #endif
2979 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
2980 			/*
2981 			 * The count was found either saturated or zero.
2982 			 * This re-read is not any more racy than using the
2983 			 * return value from fcmpset.
2984 			 */
2985 			if (refcount_load(&fp->f_count) != 0)
2986 				return (EBADF);
2987 			/*
2988 			 * Force a reload. Other thread could reallocate the
2989 			 * table before this fd was closed, so it is possible
2990 			 * that there is a stale fp pointer in cached version.
2991 			 */
2992 			fdt = atomic_load_ptr(&fdp->fd_files);
2993 			continue;
2994 		}
2995 		/*
2996 		 * Use an acquire barrier to force re-reading of fdt so it is
2997 		 * refreshed for verification.
2998 		 */
2999 		atomic_thread_fence_acq();
3000 		fdt = fdp->fd_files;
3001 #ifdef	CAPABILITIES
3002 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3003 #else
3004 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3005 #endif
3006 			break;
3007 		fdrop(fp, curthread);
3008 	}
3009 	*fpp = fp;
3010 	if (seqp != NULL) {
3011 #ifdef CAPABILITIES
3012 		*seqp = seq;
3013 #endif
3014 	}
3015 	return (0);
3016 }
3017 
3018 /*
3019  * See the comments in fget_unlocked_seq for an explanation of how this works.
3020  *
3021  * This is a simplified variant which bails out to the aforementioned routine
3022  * if anything goes wrong. In practice this only happens when userspace is
3023  * racing with itself.
3024  */
3025 int
3026 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3027     struct file **fpp)
3028 {
3029 #ifdef CAPABILITIES
3030 	const struct filedescent *fde;
3031 #endif
3032 	const struct fdescenttbl *fdt;
3033 	struct file *fp;
3034 #ifdef CAPABILITIES
3035 	seqc_t seq;
3036 	const cap_rights_t *haverights;
3037 #endif
3038 
3039 	fdt = fdp->fd_files;
3040 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3041 		return (EBADF);
3042 #ifdef CAPABILITIES
3043 	seq = seqc_read_any(fd_seqc(fdt, fd));
3044 	if (__predict_false(seqc_in_modify(seq)))
3045 		goto out_fallback;
3046 	fde = &fdt->fdt_ofiles[fd];
3047 	haverights = cap_rights_fde_inline(fde);
3048 	fp = fde->fde_file;
3049 #else
3050 	fp = fdt->fdt_ofiles[fd].fde_file;
3051 #endif
3052 	if (__predict_false(fp == NULL))
3053 		goto out_fallback;
3054 #ifdef CAPABILITIES
3055 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3056 		goto out_fallback;
3057 #endif
3058 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3059 		goto out_fallback;
3060 
3061 	/*
3062 	 * Use an acquire barrier to force re-reading of fdt so it is
3063 	 * refreshed for verification.
3064 	 */
3065 	atomic_thread_fence_acq();
3066 	fdt = fdp->fd_files;
3067 #ifdef	CAPABILITIES
3068 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3069 #else
3070 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3071 #endif
3072 		goto out_fdrop;
3073 	*fpp = fp;
3074 	return (0);
3075 out_fdrop:
3076 	fdrop(fp, curthread);
3077 out_fallback:
3078 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3079 }
3080 
3081 /*
3082  * Extract the file pointer associated with the specified descriptor for the
3083  * current user process.
3084  *
3085  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3086  * returned.
3087  *
3088  * File's rights will be checked against the capability rights mask.
3089  *
3090  * If an error occurred the non-zero error is returned and *fpp is set to
3091  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3092  * responsible for fdrop().
3093  */
3094 static __inline int
3095 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3096     cap_rights_t *needrightsp)
3097 {
3098 	struct filedesc *fdp;
3099 	struct file *fp;
3100 	int error;
3101 
3102 	*fpp = NULL;
3103 	fdp = td->td_proc->p_fd;
3104 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3105 	if (__predict_false(error != 0))
3106 		return (error);
3107 	if (__predict_false(fp->f_ops == &badfileops)) {
3108 		fdrop(fp, td);
3109 		return (EBADF);
3110 	}
3111 
3112 	/*
3113 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3114 	 */
3115 	error = 0;
3116 	switch (flags) {
3117 	case FREAD:
3118 	case FWRITE:
3119 		if ((fp->f_flag & flags) == 0)
3120 			error = EBADF;
3121 		break;
3122 	case FEXEC:
3123 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3124 		    ((fp->f_flag & FWRITE) != 0))
3125 			error = EBADF;
3126 		break;
3127 	case 0:
3128 		break;
3129 	default:
3130 		KASSERT(0, ("wrong flags"));
3131 	}
3132 
3133 	if (error != 0) {
3134 		fdrop(fp, td);
3135 		return (error);
3136 	}
3137 
3138 	*fpp = fp;
3139 	return (0);
3140 }
3141 
3142 int
3143 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3144 {
3145 
3146 	return (_fget(td, fd, fpp, 0, rightsp));
3147 }
3148 
3149 int
3150 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3151     struct file **fpp)
3152 {
3153 	int error;
3154 #ifndef CAPABILITIES
3155 	error = _fget(td, fd, fpp, 0, rightsp);
3156 	if (maxprotp != NULL)
3157 		*maxprotp = VM_PROT_ALL;
3158 	return (error);
3159 #else
3160 	cap_rights_t fdrights;
3161 	struct filedesc *fdp;
3162 	struct file *fp;
3163 	seqc_t seq;
3164 
3165 	*fpp = NULL;
3166 	fdp = td->td_proc->p_fd;
3167 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3168 	for (;;) {
3169 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3170 		if (__predict_false(error != 0))
3171 			return (error);
3172 		if (__predict_false(fp->f_ops == &badfileops)) {
3173 			fdrop(fp, td);
3174 			return (EBADF);
3175 		}
3176 		if (maxprotp != NULL)
3177 			fdrights = *cap_rights(fdp, fd);
3178 		if (!fd_modified(fdp, fd, seq))
3179 			break;
3180 		fdrop(fp, td);
3181 	}
3182 
3183 	/*
3184 	 * If requested, convert capability rights to access flags.
3185 	 */
3186 	if (maxprotp != NULL)
3187 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3188 	*fpp = fp;
3189 	return (0);
3190 #endif
3191 }
3192 
3193 int
3194 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3195 {
3196 
3197 	return (_fget(td, fd, fpp, FREAD, rightsp));
3198 }
3199 
3200 int
3201 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3202 {
3203 
3204 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3205 }
3206 
3207 int
3208 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3209     struct file **fpp)
3210 {
3211 	struct filedesc *fdp = td->td_proc->p_fd;
3212 #ifndef CAPABILITIES
3213 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3214 #else
3215 	struct file *fp;
3216 	int error;
3217 	seqc_t seq;
3218 
3219 	*fpp = NULL;
3220 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3221 	for (;;) {
3222 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3223 		if (error != 0)
3224 			return (error);
3225 		error = cap_fcntl_check(fdp, fd, needfcntl);
3226 		if (!fd_modified(fdp, fd, seq))
3227 			break;
3228 		fdrop(fp, td);
3229 	}
3230 	if (error != 0) {
3231 		fdrop(fp, td);
3232 		return (error);
3233 	}
3234 	*fpp = fp;
3235 	return (0);
3236 #endif
3237 }
3238 
3239 /*
3240  * Like fget() but loads the underlying vnode, or returns an error if the
3241  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3242  * never have VM objects.  The returned vnode will be vref()'d.
3243  *
3244  * XXX: what about the unused flags ?
3245  */
3246 static __inline int
3247 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3248     struct vnode **vpp)
3249 {
3250 	struct file *fp;
3251 	int error;
3252 
3253 	*vpp = NULL;
3254 	error = _fget(td, fd, &fp, flags, needrightsp);
3255 	if (error != 0)
3256 		return (error);
3257 	if (fp->f_vnode == NULL) {
3258 		error = EINVAL;
3259 	} else {
3260 		*vpp = fp->f_vnode;
3261 		vrefact(*vpp);
3262 	}
3263 	fdrop(fp, td);
3264 
3265 	return (error);
3266 }
3267 
3268 int
3269 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3270 {
3271 
3272 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3273 }
3274 
3275 int
3276 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3277     struct filecaps *havecaps, struct vnode **vpp)
3278 {
3279 	struct filecaps caps;
3280 	struct file *fp;
3281 	int error;
3282 
3283 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3284 	if (error != 0)
3285 		return (error);
3286 	if (fp->f_ops == &badfileops) {
3287 		error = EBADF;
3288 		goto out;
3289 	}
3290 	if (fp->f_vnode == NULL) {
3291 		error = EINVAL;
3292 		goto out;
3293 	}
3294 
3295 	*havecaps = caps;
3296 	*vpp = fp->f_vnode;
3297 	vrefact(*vpp);
3298 	fdrop(fp, td);
3299 
3300 	return (0);
3301 out:
3302 	filecaps_free(&caps);
3303 	fdrop(fp, td);
3304 	return (error);
3305 }
3306 
3307 int
3308 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3309 {
3310 
3311 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3312 }
3313 
3314 int
3315 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3316 {
3317 
3318 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3319 }
3320 
3321 #ifdef notyet
3322 int
3323 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3324     struct vnode **vpp)
3325 {
3326 
3327 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3328 }
3329 #endif
3330 
3331 /*
3332  * Handle the last reference to a file being closed.
3333  *
3334  * Without the noinline attribute clang keeps inlining the func thorough this
3335  * file when fdrop is used.
3336  */
3337 int __noinline
3338 _fdrop(struct file *fp, struct thread *td)
3339 {
3340 	int error;
3341 #ifdef INVARIANTS
3342 	int count;
3343 
3344 	count = refcount_load(&fp->f_count);
3345 	if (count != 0)
3346 		panic("fdrop: fp %p count %d", fp, count);
3347 #endif
3348 	error = fo_close(fp, td);
3349 	atomic_subtract_int(&openfiles, 1);
3350 	crfree(fp->f_cred);
3351 	free(fp->f_advice, M_FADVISE);
3352 	uma_zfree(file_zone, fp);
3353 
3354 	return (error);
3355 }
3356 
3357 /*
3358  * Apply an advisory lock on a file descriptor.
3359  *
3360  * Just attempt to get a record lock of the requested type on the entire file
3361  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3362  */
3363 #ifndef _SYS_SYSPROTO_H_
3364 struct flock_args {
3365 	int	fd;
3366 	int	how;
3367 };
3368 #endif
3369 /* ARGSUSED */
3370 int
3371 sys_flock(struct thread *td, struct flock_args *uap)
3372 {
3373 	struct file *fp;
3374 	struct vnode *vp;
3375 	struct flock lf;
3376 	int error;
3377 
3378 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3379 	if (error != 0)
3380 		return (error);
3381 	if (fp->f_type != DTYPE_VNODE) {
3382 		fdrop(fp, td);
3383 		return (EOPNOTSUPP);
3384 	}
3385 
3386 	vp = fp->f_vnode;
3387 	lf.l_whence = SEEK_SET;
3388 	lf.l_start = 0;
3389 	lf.l_len = 0;
3390 	if (uap->how & LOCK_UN) {
3391 		lf.l_type = F_UNLCK;
3392 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3393 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3394 		goto done2;
3395 	}
3396 	if (uap->how & LOCK_EX)
3397 		lf.l_type = F_WRLCK;
3398 	else if (uap->how & LOCK_SH)
3399 		lf.l_type = F_RDLCK;
3400 	else {
3401 		error = EBADF;
3402 		goto done2;
3403 	}
3404 	atomic_set_int(&fp->f_flag, FHASLOCK);
3405 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3406 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3407 done2:
3408 	fdrop(fp, td);
3409 	return (error);
3410 }
3411 /*
3412  * Duplicate the specified descriptor to a free descriptor.
3413  */
3414 int
3415 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3416     int openerror, int *indxp)
3417 {
3418 	struct filedescent *newfde, *oldfde;
3419 	struct file *fp;
3420 	u_long *ioctls;
3421 	int error, indx;
3422 
3423 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3424 	    ("unexpected error %d in %s", openerror, __func__));
3425 
3426 	/*
3427 	 * If the to-be-dup'd fd number is greater than the allowed number
3428 	 * of file descriptors, or the fd to be dup'd has already been
3429 	 * closed, then reject.
3430 	 */
3431 	FILEDESC_XLOCK(fdp);
3432 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3433 		FILEDESC_XUNLOCK(fdp);
3434 		return (EBADF);
3435 	}
3436 
3437 	error = fdalloc(td, 0, &indx);
3438 	if (error != 0) {
3439 		FILEDESC_XUNLOCK(fdp);
3440 		return (error);
3441 	}
3442 
3443 	/*
3444 	 * There are two cases of interest here.
3445 	 *
3446 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3447 	 *
3448 	 * For ENXIO steal away the file structure from (dfd) and store it in
3449 	 * (indx).  (dfd) is effectively closed by this operation.
3450 	 */
3451 	switch (openerror) {
3452 	case ENODEV:
3453 		/*
3454 		 * Check that the mode the file is being opened for is a
3455 		 * subset of the mode of the existing descriptor.
3456 		 */
3457 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3458 			fdunused(fdp, indx);
3459 			FILEDESC_XUNLOCK(fdp);
3460 			return (EACCES);
3461 		}
3462 		if (!fhold(fp)) {
3463 			fdunused(fdp, indx);
3464 			FILEDESC_XUNLOCK(fdp);
3465 			return (EBADF);
3466 		}
3467 		newfde = &fdp->fd_ofiles[indx];
3468 		oldfde = &fdp->fd_ofiles[dfd];
3469 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3470 #ifdef CAPABILITIES
3471 		seqc_write_begin(&newfde->fde_seqc);
3472 #endif
3473 		memcpy(newfde, oldfde, fde_change_size);
3474 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3475 		    ioctls);
3476 #ifdef CAPABILITIES
3477 		seqc_write_end(&newfde->fde_seqc);
3478 #endif
3479 		break;
3480 	case ENXIO:
3481 		/*
3482 		 * Steal away the file pointer from dfd and stuff it into indx.
3483 		 */
3484 		newfde = &fdp->fd_ofiles[indx];
3485 		oldfde = &fdp->fd_ofiles[dfd];
3486 #ifdef CAPABILITIES
3487 		seqc_write_begin(&newfde->fde_seqc);
3488 #endif
3489 		memcpy(newfde, oldfde, fde_change_size);
3490 		oldfde->fde_file = NULL;
3491 		fdunused(fdp, dfd);
3492 #ifdef CAPABILITIES
3493 		seqc_write_end(&newfde->fde_seqc);
3494 #endif
3495 		break;
3496 	}
3497 	FILEDESC_XUNLOCK(fdp);
3498 	*indxp = indx;
3499 	return (0);
3500 }
3501 
3502 /*
3503  * This sysctl determines if we will allow a process to chroot(2) if it
3504  * has a directory open:
3505  *	0: disallowed for all processes.
3506  *	1: allowed for processes that were not already chroot(2)'ed.
3507  *	2: allowed for all processes.
3508  */
3509 
3510 static int chroot_allow_open_directories = 1;
3511 
3512 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3513     &chroot_allow_open_directories, 0,
3514     "Allow a process to chroot(2) if it has a directory open");
3515 
3516 /*
3517  * Helper function for raised chroot(2) security function:  Refuse if
3518  * any filedescriptors are open directories.
3519  */
3520 static int
3521 chroot_refuse_vdir_fds(struct filedesc *fdp)
3522 {
3523 	struct vnode *vp;
3524 	struct file *fp;
3525 	int fd, lastfile;
3526 
3527 	FILEDESC_LOCK_ASSERT(fdp);
3528 
3529 	lastfile = fdlastfile(fdp);
3530 	for (fd = 0; fd <= lastfile; fd++) {
3531 		fp = fget_locked(fdp, fd);
3532 		if (fp == NULL)
3533 			continue;
3534 		if (fp->f_type == DTYPE_VNODE) {
3535 			vp = fp->f_vnode;
3536 			if (vp->v_type == VDIR)
3537 				return (EPERM);
3538 		}
3539 	}
3540 	return (0);
3541 }
3542 
3543 static void
3544 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3545 {
3546 
3547 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3548 		vrefact(oldpwd->pwd_cdir);
3549 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3550 	}
3551 
3552 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3553 		vrefact(oldpwd->pwd_rdir);
3554 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3555 	}
3556 
3557 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3558 		vrefact(oldpwd->pwd_jdir);
3559 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3560 	}
3561 }
3562 
3563 struct pwd *
3564 pwd_hold_pwddesc(struct pwddesc *pdp)
3565 {
3566 	struct pwd *pwd;
3567 
3568 	PWDDESC_ASSERT_XLOCKED(pdp);
3569 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3570 	if (pwd != NULL)
3571 		refcount_acquire(&pwd->pwd_refcount);
3572 	return (pwd);
3573 }
3574 
3575 bool
3576 pwd_hold_smr(struct pwd *pwd)
3577 {
3578 
3579 	MPASS(pwd != NULL);
3580 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3581 		return (true);
3582 	}
3583 	return (false);
3584 }
3585 
3586 struct pwd *
3587 pwd_hold(struct thread *td)
3588 {
3589 	struct pwddesc *pdp;
3590 	struct pwd *pwd;
3591 
3592 	pdp = td->td_proc->p_pd;
3593 
3594 	vfs_smr_enter();
3595 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3596 	if (pwd_hold_smr(pwd)) {
3597 		vfs_smr_exit();
3598 		return (pwd);
3599 	}
3600 	vfs_smr_exit();
3601 	PWDDESC_XLOCK(pdp);
3602 	pwd = pwd_hold_pwddesc(pdp);
3603 	MPASS(pwd != NULL);
3604 	PWDDESC_XUNLOCK(pdp);
3605 	return (pwd);
3606 }
3607 
3608 struct pwd *
3609 pwd_get_smr(void)
3610 {
3611 	struct pwd *pwd;
3612 
3613 	pwd = vfs_smr_entered_load(&curproc->p_pd->pd_pwd);
3614 	MPASS(pwd != NULL);
3615 	return (pwd);
3616 }
3617 
3618 static struct pwd *
3619 pwd_alloc(void)
3620 {
3621 	struct pwd *pwd;
3622 
3623 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3624 	bzero(pwd, sizeof(*pwd));
3625 	refcount_init(&pwd->pwd_refcount, 1);
3626 	return (pwd);
3627 }
3628 
3629 void
3630 pwd_drop(struct pwd *pwd)
3631 {
3632 
3633 	if (!refcount_release(&pwd->pwd_refcount))
3634 		return;
3635 
3636 	if (pwd->pwd_cdir != NULL)
3637 		vrele(pwd->pwd_cdir);
3638 	if (pwd->pwd_rdir != NULL)
3639 		vrele(pwd->pwd_rdir);
3640 	if (pwd->pwd_jdir != NULL)
3641 		vrele(pwd->pwd_jdir);
3642 	uma_zfree_smr(pwd_zone, pwd);
3643 }
3644 
3645 /*
3646 * Common routine for kern_chroot() and jail_attach().  The caller is
3647 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3648 * authorize this operation.
3649 */
3650 int
3651 pwd_chroot(struct thread *td, struct vnode *vp)
3652 {
3653 	struct pwddesc *pdp;
3654 	struct filedesc *fdp;
3655 	struct pwd *newpwd, *oldpwd;
3656 	int error;
3657 
3658 	fdp = td->td_proc->p_fd;
3659 	pdp = td->td_proc->p_pd;
3660 	newpwd = pwd_alloc();
3661 	FILEDESC_SLOCK(fdp);
3662 	PWDDESC_XLOCK(pdp);
3663 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3664 	if (chroot_allow_open_directories == 0 ||
3665 	    (chroot_allow_open_directories == 1 &&
3666 	    oldpwd->pwd_rdir != rootvnode)) {
3667 		error = chroot_refuse_vdir_fds(fdp);
3668 		FILEDESC_SUNLOCK(fdp);
3669 		if (error != 0) {
3670 			PWDDESC_XUNLOCK(pdp);
3671 			pwd_drop(newpwd);
3672 			return (error);
3673 		}
3674 	} else {
3675 		FILEDESC_SUNLOCK(fdp);
3676 	}
3677 
3678 	vrefact(vp);
3679 	newpwd->pwd_rdir = vp;
3680 	if (oldpwd->pwd_jdir == NULL) {
3681 		vrefact(vp);
3682 		newpwd->pwd_jdir = vp;
3683 	}
3684 	pwd_fill(oldpwd, newpwd);
3685 	pwd_set(pdp, newpwd);
3686 	PWDDESC_XUNLOCK(pdp);
3687 	pwd_drop(oldpwd);
3688 	return (0);
3689 }
3690 
3691 void
3692 pwd_chdir(struct thread *td, struct vnode *vp)
3693 {
3694 	struct pwddesc *pdp;
3695 	struct pwd *newpwd, *oldpwd;
3696 
3697 	VNPASS(vp->v_usecount > 0, vp);
3698 
3699 	newpwd = pwd_alloc();
3700 	pdp = td->td_proc->p_pd;
3701 	PWDDESC_XLOCK(pdp);
3702 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3703 	newpwd->pwd_cdir = vp;
3704 	pwd_fill(oldpwd, newpwd);
3705 	pwd_set(pdp, newpwd);
3706 	PWDDESC_XUNLOCK(pdp);
3707 	pwd_drop(oldpwd);
3708 }
3709 
3710 void
3711 pwd_ensure_dirs(void)
3712 {
3713 	struct pwddesc *pdp;
3714 	struct pwd *oldpwd, *newpwd;
3715 
3716 	pdp = curproc->p_pd;
3717 	PWDDESC_XLOCK(pdp);
3718 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3719 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3720 		PWDDESC_XUNLOCK(pdp);
3721 		return;
3722 	}
3723 	PWDDESC_XUNLOCK(pdp);
3724 
3725 	newpwd = pwd_alloc();
3726 	PWDDESC_XLOCK(pdp);
3727 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3728 	pwd_fill(oldpwd, newpwd);
3729 	if (newpwd->pwd_cdir == NULL) {
3730 		vrefact(rootvnode);
3731 		newpwd->pwd_cdir = rootvnode;
3732 	}
3733 	if (newpwd->pwd_rdir == NULL) {
3734 		vrefact(rootvnode);
3735 		newpwd->pwd_rdir = rootvnode;
3736 	}
3737 	pwd_set(pdp, newpwd);
3738 	PWDDESC_XUNLOCK(pdp);
3739 	pwd_drop(oldpwd);
3740 }
3741 
3742 void
3743 pwd_set_rootvnode(void)
3744 {
3745 	struct pwddesc *pdp;
3746 	struct pwd *oldpwd, *newpwd;
3747 
3748 	pdp = curproc->p_pd;
3749 
3750 	newpwd = pwd_alloc();
3751 	PWDDESC_XLOCK(pdp);
3752 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3753 	vrefact(rootvnode);
3754 	newpwd->pwd_cdir = rootvnode;
3755 	vrefact(rootvnode);
3756 	newpwd->pwd_rdir = rootvnode;
3757 	pwd_fill(oldpwd, newpwd);
3758 	pwd_set(pdp, newpwd);
3759 	PWDDESC_XUNLOCK(pdp);
3760 	pwd_drop(oldpwd);
3761 }
3762 
3763 /*
3764  * Scan all active processes and prisons to see if any of them have a current
3765  * or root directory of `olddp'. If so, replace them with the new mount point.
3766  */
3767 void
3768 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3769 {
3770 	struct pwddesc *pdp;
3771 	struct pwd *newpwd, *oldpwd;
3772 	struct prison *pr;
3773 	struct proc *p;
3774 	int nrele;
3775 
3776 	if (vrefcnt(olddp) == 1)
3777 		return;
3778 	nrele = 0;
3779 	newpwd = pwd_alloc();
3780 	sx_slock(&allproc_lock);
3781 	FOREACH_PROC_IN_SYSTEM(p) {
3782 		PROC_LOCK(p);
3783 		pdp = pdhold(p);
3784 		PROC_UNLOCK(p);
3785 		if (pdp == NULL)
3786 			continue;
3787 		PWDDESC_XLOCK(pdp);
3788 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3789 		if (oldpwd == NULL ||
3790 		    (oldpwd->pwd_cdir != olddp &&
3791 		    oldpwd->pwd_rdir != olddp &&
3792 		    oldpwd->pwd_jdir != olddp)) {
3793 			PWDDESC_XUNLOCK(pdp);
3794 			pddrop(pdp);
3795 			continue;
3796 		}
3797 		if (oldpwd->pwd_cdir == olddp) {
3798 			vrefact(newdp);
3799 			newpwd->pwd_cdir = newdp;
3800 		}
3801 		if (oldpwd->pwd_rdir == olddp) {
3802 			vrefact(newdp);
3803 			newpwd->pwd_rdir = newdp;
3804 		}
3805 		if (oldpwd->pwd_jdir == olddp) {
3806 			vrefact(newdp);
3807 			newpwd->pwd_jdir = newdp;
3808 		}
3809 		pwd_fill(oldpwd, newpwd);
3810 		pwd_set(pdp, newpwd);
3811 		PWDDESC_XUNLOCK(pdp);
3812 		pwd_drop(oldpwd);
3813 		pddrop(pdp);
3814 		newpwd = pwd_alloc();
3815 	}
3816 	sx_sunlock(&allproc_lock);
3817 	pwd_drop(newpwd);
3818 	if (rootvnode == olddp) {
3819 		vrefact(newdp);
3820 		rootvnode = newdp;
3821 		nrele++;
3822 	}
3823 	mtx_lock(&prison0.pr_mtx);
3824 	if (prison0.pr_root == olddp) {
3825 		vrefact(newdp);
3826 		prison0.pr_root = newdp;
3827 		nrele++;
3828 	}
3829 	mtx_unlock(&prison0.pr_mtx);
3830 	sx_slock(&allprison_lock);
3831 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3832 		mtx_lock(&pr->pr_mtx);
3833 		if (pr->pr_root == olddp) {
3834 			vrefact(newdp);
3835 			pr->pr_root = newdp;
3836 			nrele++;
3837 		}
3838 		mtx_unlock(&pr->pr_mtx);
3839 	}
3840 	sx_sunlock(&allprison_lock);
3841 	while (nrele--)
3842 		vrele(olddp);
3843 }
3844 
3845 struct filedesc_to_leader *
3846 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3847 {
3848 	struct filedesc_to_leader *fdtol;
3849 
3850 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3851 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3852 	fdtol->fdl_refcount = 1;
3853 	fdtol->fdl_holdcount = 0;
3854 	fdtol->fdl_wakeup = 0;
3855 	fdtol->fdl_leader = leader;
3856 	if (old != NULL) {
3857 		FILEDESC_XLOCK(fdp);
3858 		fdtol->fdl_next = old->fdl_next;
3859 		fdtol->fdl_prev = old;
3860 		old->fdl_next = fdtol;
3861 		fdtol->fdl_next->fdl_prev = fdtol;
3862 		FILEDESC_XUNLOCK(fdp);
3863 	} else {
3864 		fdtol->fdl_next = fdtol;
3865 		fdtol->fdl_prev = fdtol;
3866 	}
3867 	return (fdtol);
3868 }
3869 
3870 static int
3871 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3872 {
3873 	NDSLOTTYPE *map;
3874 	struct filedesc *fdp;
3875 	int count, off, minoff;
3876 
3877 	if (*(int *)arg1 != 0)
3878 		return (EINVAL);
3879 
3880 	fdp = curproc->p_fd;
3881 	count = 0;
3882 	FILEDESC_SLOCK(fdp);
3883 	map = fdp->fd_map;
3884 	off = NDSLOT(fdp->fd_nfiles - 1);
3885 	for (minoff = NDSLOT(0); off >= minoff; --off)
3886 		count += bitcountl(map[off]);
3887 	FILEDESC_SUNLOCK(fdp);
3888 
3889 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3890 }
3891 
3892 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3893     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3894     "Number of open file descriptors");
3895 
3896 /*
3897  * Get file structures globally.
3898  */
3899 static int
3900 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3901 {
3902 	struct xfile xf;
3903 	struct filedesc *fdp;
3904 	struct file *fp;
3905 	struct proc *p;
3906 	int error, n, lastfile;
3907 
3908 	error = sysctl_wire_old_buffer(req, 0);
3909 	if (error != 0)
3910 		return (error);
3911 	if (req->oldptr == NULL) {
3912 		n = 0;
3913 		sx_slock(&allproc_lock);
3914 		FOREACH_PROC_IN_SYSTEM(p) {
3915 			PROC_LOCK(p);
3916 			if (p->p_state == PRS_NEW) {
3917 				PROC_UNLOCK(p);
3918 				continue;
3919 			}
3920 			fdp = fdhold(p);
3921 			PROC_UNLOCK(p);
3922 			if (fdp == NULL)
3923 				continue;
3924 			/* overestimates sparse tables. */
3925 			n += fdp->fd_nfiles;
3926 			fddrop(fdp);
3927 		}
3928 		sx_sunlock(&allproc_lock);
3929 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3930 	}
3931 	error = 0;
3932 	bzero(&xf, sizeof(xf));
3933 	xf.xf_size = sizeof(xf);
3934 	sx_slock(&allproc_lock);
3935 	FOREACH_PROC_IN_SYSTEM(p) {
3936 		PROC_LOCK(p);
3937 		if (p->p_state == PRS_NEW) {
3938 			PROC_UNLOCK(p);
3939 			continue;
3940 		}
3941 		if (p_cansee(req->td, p) != 0) {
3942 			PROC_UNLOCK(p);
3943 			continue;
3944 		}
3945 		xf.xf_pid = p->p_pid;
3946 		xf.xf_uid = p->p_ucred->cr_uid;
3947 		fdp = fdhold(p);
3948 		PROC_UNLOCK(p);
3949 		if (fdp == NULL)
3950 			continue;
3951 		FILEDESC_SLOCK(fdp);
3952 		lastfile = fdlastfile(fdp);
3953 		for (n = 0; fdp->fd_refcnt > 0 && n <= lastfile; ++n) {
3954 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3955 				continue;
3956 			xf.xf_fd = n;
3957 			xf.xf_file = (uintptr_t)fp;
3958 			xf.xf_data = (uintptr_t)fp->f_data;
3959 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3960 			xf.xf_type = (uintptr_t)fp->f_type;
3961 			xf.xf_count = refcount_load(&fp->f_count);
3962 			xf.xf_msgcount = 0;
3963 			xf.xf_offset = foffset_get(fp);
3964 			xf.xf_flag = fp->f_flag;
3965 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3966 			if (error)
3967 				break;
3968 		}
3969 		FILEDESC_SUNLOCK(fdp);
3970 		fddrop(fdp);
3971 		if (error)
3972 			break;
3973 	}
3974 	sx_sunlock(&allproc_lock);
3975 	return (error);
3976 }
3977 
3978 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3979     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3980 
3981 #ifdef KINFO_FILE_SIZE
3982 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3983 #endif
3984 
3985 static int
3986 xlate_fflags(int fflags)
3987 {
3988 	static const struct {
3989 		int	fflag;
3990 		int	kf_fflag;
3991 	} fflags_table[] = {
3992 		{ FAPPEND, KF_FLAG_APPEND },
3993 		{ FASYNC, KF_FLAG_ASYNC },
3994 		{ FFSYNC, KF_FLAG_FSYNC },
3995 		{ FHASLOCK, KF_FLAG_HASLOCK },
3996 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3997 		{ FREAD, KF_FLAG_READ },
3998 		{ FWRITE, KF_FLAG_WRITE },
3999 		{ O_CREAT, KF_FLAG_CREAT },
4000 		{ O_DIRECT, KF_FLAG_DIRECT },
4001 		{ O_EXCL, KF_FLAG_EXCL },
4002 		{ O_EXEC, KF_FLAG_EXEC },
4003 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4004 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4005 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4006 		{ O_TRUNC, KF_FLAG_TRUNC }
4007 	};
4008 	unsigned int i;
4009 	int kflags;
4010 
4011 	kflags = 0;
4012 	for (i = 0; i < nitems(fflags_table); i++)
4013 		if (fflags & fflags_table[i].fflag)
4014 			kflags |=  fflags_table[i].kf_fflag;
4015 	return (kflags);
4016 }
4017 
4018 /* Trim unused data from kf_path by truncating the structure size. */
4019 void
4020 pack_kinfo(struct kinfo_file *kif)
4021 {
4022 
4023 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4024 	    strlen(kif->kf_path) + 1;
4025 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4026 }
4027 
4028 static void
4029 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4030     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4031 {
4032 	int error;
4033 
4034 	bzero(kif, sizeof(*kif));
4035 
4036 	/* Set a default type to allow for empty fill_kinfo() methods. */
4037 	kif->kf_type = KF_TYPE_UNKNOWN;
4038 	kif->kf_flags = xlate_fflags(fp->f_flag);
4039 	if (rightsp != NULL)
4040 		kif->kf_cap_rights = *rightsp;
4041 	else
4042 		cap_rights_init_zero(&kif->kf_cap_rights);
4043 	kif->kf_fd = fd;
4044 	kif->kf_ref_count = refcount_load(&fp->f_count);
4045 	kif->kf_offset = foffset_get(fp);
4046 
4047 	/*
4048 	 * This may drop the filedesc lock, so the 'fp' cannot be
4049 	 * accessed after this call.
4050 	 */
4051 	error = fo_fill_kinfo(fp, kif, fdp);
4052 	if (error == 0)
4053 		kif->kf_status |= KF_ATTR_VALID;
4054 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4055 		pack_kinfo(kif);
4056 	else
4057 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4058 }
4059 
4060 static void
4061 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4062     struct kinfo_file *kif, int flags)
4063 {
4064 	int error;
4065 
4066 	bzero(kif, sizeof(*kif));
4067 
4068 	kif->kf_type = KF_TYPE_VNODE;
4069 	error = vn_fill_kinfo_vnode(vp, kif);
4070 	if (error == 0)
4071 		kif->kf_status |= KF_ATTR_VALID;
4072 	kif->kf_flags = xlate_fflags(fflags);
4073 	cap_rights_init_zero(&kif->kf_cap_rights);
4074 	kif->kf_fd = fd;
4075 	kif->kf_ref_count = -1;
4076 	kif->kf_offset = -1;
4077 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4078 		pack_kinfo(kif);
4079 	else
4080 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4081 	vrele(vp);
4082 }
4083 
4084 struct export_fd_buf {
4085 	struct filedesc		*fdp;
4086 	struct pwddesc	*pdp;
4087 	struct sbuf 		*sb;
4088 	ssize_t			remainder;
4089 	struct kinfo_file	kif;
4090 	int			flags;
4091 };
4092 
4093 static int
4094 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4095 {
4096 	struct kinfo_file *kif;
4097 
4098 	kif = &efbuf->kif;
4099 	if (efbuf->remainder != -1) {
4100 		if (efbuf->remainder < kif->kf_structsize) {
4101 			/* Terminate export. */
4102 			efbuf->remainder = 0;
4103 			return (0);
4104 		}
4105 		efbuf->remainder -= kif->kf_structsize;
4106 	}
4107 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4108 }
4109 
4110 static int
4111 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4112     struct export_fd_buf *efbuf)
4113 {
4114 	int error;
4115 
4116 	if (efbuf->remainder == 0)
4117 		return (0);
4118 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4119 	    efbuf->flags);
4120 	FILEDESC_SUNLOCK(efbuf->fdp);
4121 	error = export_kinfo_to_sb(efbuf);
4122 	FILEDESC_SLOCK(efbuf->fdp);
4123 	return (error);
4124 }
4125 
4126 static int
4127 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4128     struct export_fd_buf *efbuf)
4129 {
4130 	int error;
4131 
4132 	if (efbuf->remainder == 0)
4133 		return (0);
4134 	if (efbuf->pdp != NULL)
4135 		PWDDESC_XUNLOCK(efbuf->pdp);
4136 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4137 	error = export_kinfo_to_sb(efbuf);
4138 	if (efbuf->pdp != NULL)
4139 		PWDDESC_XLOCK(efbuf->pdp);
4140 	return (error);
4141 }
4142 
4143 /*
4144  * Store a process file descriptor information to sbuf.
4145  *
4146  * Takes a locked proc as argument, and returns with the proc unlocked.
4147  */
4148 int
4149 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4150     int flags)
4151 {
4152 	struct file *fp;
4153 	struct filedesc *fdp;
4154 	struct pwddesc *pdp;
4155 	struct export_fd_buf *efbuf;
4156 	struct vnode *cttyvp, *textvp, *tracevp;
4157 	struct pwd *pwd;
4158 	int error, i, lastfile;
4159 	cap_rights_t rights;
4160 
4161 	PROC_LOCK_ASSERT(p, MA_OWNED);
4162 
4163 	/* ktrace vnode */
4164 	tracevp = p->p_tracevp;
4165 	if (tracevp != NULL)
4166 		vrefact(tracevp);
4167 	/* text vnode */
4168 	textvp = p->p_textvp;
4169 	if (textvp != NULL)
4170 		vrefact(textvp);
4171 	/* Controlling tty. */
4172 	cttyvp = NULL;
4173 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4174 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4175 		if (cttyvp != NULL)
4176 			vrefact(cttyvp);
4177 	}
4178 	fdp = fdhold(p);
4179 	pdp = pdhold(p);
4180 	PROC_UNLOCK(p);
4181 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4182 	efbuf->fdp = NULL;
4183 	efbuf->pdp = NULL;
4184 	efbuf->sb = sb;
4185 	efbuf->remainder = maxlen;
4186 	efbuf->flags = flags;
4187 	if (tracevp != NULL)
4188 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4189 		    efbuf);
4190 	if (textvp != NULL)
4191 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4192 	if (cttyvp != NULL)
4193 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4194 		    efbuf);
4195 	error = 0;
4196 	if (pdp == NULL || fdp == NULL)
4197 		goto fail;
4198 	efbuf->fdp = fdp;
4199 	efbuf->pdp = pdp;
4200 	PWDDESC_XLOCK(pdp);
4201 	pwd = pwd_hold_pwddesc(pdp);
4202 	if (pwd != NULL) {
4203 		/* working directory */
4204 		if (pwd->pwd_cdir != NULL) {
4205 			vrefact(pwd->pwd_cdir);
4206 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4207 		}
4208 		/* root directory */
4209 		if (pwd->pwd_rdir != NULL) {
4210 			vrefact(pwd->pwd_rdir);
4211 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4212 		}
4213 		/* jail directory */
4214 		if (pwd->pwd_jdir != NULL) {
4215 			vrefact(pwd->pwd_jdir);
4216 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4217 		}
4218 	}
4219 	PWDDESC_XUNLOCK(pdp);
4220 	if (pwd != NULL)
4221 		pwd_drop(pwd);
4222 	FILEDESC_SLOCK(fdp);
4223 	lastfile = fdlastfile(fdp);
4224 	for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4225 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4226 			continue;
4227 #ifdef CAPABILITIES
4228 		rights = *cap_rights(fdp, i);
4229 #else /* !CAPABILITIES */
4230 		rights = cap_no_rights;
4231 #endif
4232 		/*
4233 		 * Create sysctl entry.  It is OK to drop the filedesc
4234 		 * lock inside of export_file_to_sb() as we will
4235 		 * re-validate and re-evaluate its properties when the
4236 		 * loop continues.
4237 		 */
4238 		error = export_file_to_sb(fp, i, &rights, efbuf);
4239 		if (error != 0 || efbuf->remainder == 0)
4240 			break;
4241 	}
4242 	FILEDESC_SUNLOCK(fdp);
4243 fail:
4244 	if (fdp != NULL)
4245 		fddrop(fdp);
4246 	if (pdp != NULL)
4247 		pddrop(pdp);
4248 	free(efbuf, M_TEMP);
4249 	return (error);
4250 }
4251 
4252 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4253 
4254 /*
4255  * Get per-process file descriptors for use by procstat(1), et al.
4256  */
4257 static int
4258 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4259 {
4260 	struct sbuf sb;
4261 	struct proc *p;
4262 	ssize_t maxlen;
4263 	int error, error2, *name;
4264 
4265 	name = (int *)arg1;
4266 
4267 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4268 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4269 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4270 	if (error != 0) {
4271 		sbuf_delete(&sb);
4272 		return (error);
4273 	}
4274 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4275 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4276 	    KERN_FILEDESC_PACK_KINFO);
4277 	error2 = sbuf_finish(&sb);
4278 	sbuf_delete(&sb);
4279 	return (error != 0 ? error : error2);
4280 }
4281 
4282 #ifdef COMPAT_FREEBSD7
4283 #ifdef KINFO_OFILE_SIZE
4284 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4285 #endif
4286 
4287 static void
4288 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4289 {
4290 
4291 	okif->kf_structsize = sizeof(*okif);
4292 	okif->kf_type = kif->kf_type;
4293 	okif->kf_fd = kif->kf_fd;
4294 	okif->kf_ref_count = kif->kf_ref_count;
4295 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4296 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4297 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4298 	okif->kf_offset = kif->kf_offset;
4299 	if (kif->kf_type == KF_TYPE_VNODE)
4300 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4301 	else
4302 		okif->kf_vnode_type = KF_VTYPE_VNON;
4303 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4304 	if (kif->kf_type == KF_TYPE_SOCKET) {
4305 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4306 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4307 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4308 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4309 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4310 	} else {
4311 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4312 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4313 	}
4314 }
4315 
4316 static int
4317 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4318     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4319 {
4320 	int error;
4321 
4322 	vrefact(vp);
4323 	PWDDESC_XUNLOCK(pdp);
4324 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4325 	kinfo_to_okinfo(kif, okif);
4326 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4327 	PWDDESC_XLOCK(pdp);
4328 	return (error);
4329 }
4330 
4331 /*
4332  * Get per-process file descriptors for use by procstat(1), et al.
4333  */
4334 static int
4335 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4336 {
4337 	struct kinfo_ofile *okif;
4338 	struct kinfo_file *kif;
4339 	struct filedesc *fdp;
4340 	struct pwddesc *pdp;
4341 	struct pwd *pwd;
4342 	int error, i, lastfile, *name;
4343 	struct file *fp;
4344 	struct proc *p;
4345 
4346 	name = (int *)arg1;
4347 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4348 	if (error != 0)
4349 		return (error);
4350 	fdp = fdhold(p);
4351 	if (fdp != NULL)
4352 		pdp = pdhold(p);
4353 	PROC_UNLOCK(p);
4354 	if (fdp == NULL || pdp == NULL) {
4355 		if (fdp != NULL)
4356 			fddrop(fdp);
4357 		return (ENOENT);
4358 	}
4359 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4360 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4361 	PWDDESC_XLOCK(pdp);
4362 	pwd = pwd_hold_pwddesc(pdp);
4363 	if (pwd != NULL) {
4364 		if (pwd->pwd_cdir != NULL)
4365 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4366 			    okif, pdp, req);
4367 		if (pwd->pwd_rdir != NULL)
4368 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4369 			    okif, pdp, req);
4370 		if (pwd->pwd_jdir != NULL)
4371 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4372 			    okif, pdp, req);
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; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4380 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4381 			continue;
4382 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4383 		    KERN_FILEDESC_PACK_KINFO);
4384 		FILEDESC_SUNLOCK(fdp);
4385 		kinfo_to_okinfo(kif, okif);
4386 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4387 		FILEDESC_SLOCK(fdp);
4388 		if (error)
4389 			break;
4390 	}
4391 	FILEDESC_SUNLOCK(fdp);
4392 	fddrop(fdp);
4393 	pddrop(pdp);
4394 	free(kif, M_TEMP);
4395 	free(okif, M_TEMP);
4396 	return (0);
4397 }
4398 
4399 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4400     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4401     "Process ofiledesc entries");
4402 #endif	/* COMPAT_FREEBSD7 */
4403 
4404 int
4405 vntype_to_kinfo(int vtype)
4406 {
4407 	struct {
4408 		int	vtype;
4409 		int	kf_vtype;
4410 	} vtypes_table[] = {
4411 		{ VBAD, KF_VTYPE_VBAD },
4412 		{ VBLK, KF_VTYPE_VBLK },
4413 		{ VCHR, KF_VTYPE_VCHR },
4414 		{ VDIR, KF_VTYPE_VDIR },
4415 		{ VFIFO, KF_VTYPE_VFIFO },
4416 		{ VLNK, KF_VTYPE_VLNK },
4417 		{ VNON, KF_VTYPE_VNON },
4418 		{ VREG, KF_VTYPE_VREG },
4419 		{ VSOCK, KF_VTYPE_VSOCK }
4420 	};
4421 	unsigned int i;
4422 
4423 	/*
4424 	 * Perform vtype translation.
4425 	 */
4426 	for (i = 0; i < nitems(vtypes_table); i++)
4427 		if (vtypes_table[i].vtype == vtype)
4428 			return (vtypes_table[i].kf_vtype);
4429 
4430 	return (KF_VTYPE_UNKNOWN);
4431 }
4432 
4433 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4434     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4435     "Process filedesc entries");
4436 
4437 /*
4438  * Store a process current working directory information to sbuf.
4439  *
4440  * Takes a locked proc as argument, and returns with the proc unlocked.
4441  */
4442 int
4443 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4444 {
4445 	struct pwddesc *pdp;
4446 	struct pwd *pwd;
4447 	struct export_fd_buf *efbuf;
4448 	struct vnode *cdir;
4449 	int error;
4450 
4451 	PROC_LOCK_ASSERT(p, MA_OWNED);
4452 
4453 	pdp = pdhold(p);
4454 	PROC_UNLOCK(p);
4455 	if (pdp == NULL)
4456 		return (EINVAL);
4457 
4458 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4459 	efbuf->pdp = pdp;
4460 	efbuf->sb = sb;
4461 	efbuf->remainder = maxlen;
4462 
4463 	PWDDESC_XLOCK(pdp);
4464 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4465 	cdir = pwd->pwd_cdir;
4466 	if (cdir == NULL) {
4467 		error = EINVAL;
4468 	} else {
4469 		vrefact(cdir);
4470 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4471 	}
4472 	PWDDESC_XUNLOCK(pdp);
4473 	pddrop(pdp);
4474 	free(efbuf, M_TEMP);
4475 	return (error);
4476 }
4477 
4478 /*
4479  * Get per-process current working directory.
4480  */
4481 static int
4482 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4483 {
4484 	struct sbuf sb;
4485 	struct proc *p;
4486 	ssize_t maxlen;
4487 	int error, error2, *name;
4488 
4489 	name = (int *)arg1;
4490 
4491 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4492 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4493 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4494 	if (error != 0) {
4495 		sbuf_delete(&sb);
4496 		return (error);
4497 	}
4498 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4499 	error = kern_proc_cwd_out(p, &sb, maxlen);
4500 	error2 = sbuf_finish(&sb);
4501 	sbuf_delete(&sb);
4502 	return (error != 0 ? error : error2);
4503 }
4504 
4505 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4506     sysctl_kern_proc_cwd, "Process current working directory");
4507 
4508 #ifdef DDB
4509 /*
4510  * For the purposes of debugging, generate a human-readable string for the
4511  * file type.
4512  */
4513 static const char *
4514 file_type_to_name(short type)
4515 {
4516 
4517 	switch (type) {
4518 	case 0:
4519 		return ("zero");
4520 	case DTYPE_VNODE:
4521 		return ("vnode");
4522 	case DTYPE_SOCKET:
4523 		return ("socket");
4524 	case DTYPE_PIPE:
4525 		return ("pipe");
4526 	case DTYPE_FIFO:
4527 		return ("fifo");
4528 	case DTYPE_KQUEUE:
4529 		return ("kqueue");
4530 	case DTYPE_CRYPTO:
4531 		return ("crypto");
4532 	case DTYPE_MQUEUE:
4533 		return ("mqueue");
4534 	case DTYPE_SHM:
4535 		return ("shm");
4536 	case DTYPE_SEM:
4537 		return ("ksem");
4538 	case DTYPE_PTS:
4539 		return ("pts");
4540 	case DTYPE_DEV:
4541 		return ("dev");
4542 	case DTYPE_PROCDESC:
4543 		return ("proc");
4544 	case DTYPE_LINUXEFD:
4545 		return ("levent");
4546 	case DTYPE_LINUXTFD:
4547 		return ("ltimer");
4548 	default:
4549 		return ("unkn");
4550 	}
4551 }
4552 
4553 /*
4554  * For the purposes of debugging, identify a process (if any, perhaps one of
4555  * many) that references the passed file in its file descriptor array. Return
4556  * NULL if none.
4557  */
4558 static struct proc *
4559 file_to_first_proc(struct file *fp)
4560 {
4561 	struct filedesc *fdp;
4562 	struct proc *p;
4563 	int n;
4564 
4565 	FOREACH_PROC_IN_SYSTEM(p) {
4566 		if (p->p_state == PRS_NEW)
4567 			continue;
4568 		fdp = p->p_fd;
4569 		if (fdp == NULL)
4570 			continue;
4571 		for (n = 0; n < fdp->fd_nfiles; n++) {
4572 			if (fp == fdp->fd_ofiles[n].fde_file)
4573 				return (p);
4574 		}
4575 	}
4576 	return (NULL);
4577 }
4578 
4579 static void
4580 db_print_file(struct file *fp, int header)
4581 {
4582 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4583 	struct proc *p;
4584 
4585 	if (header)
4586 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4587 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4588 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4589 		    "FCmd");
4590 	p = file_to_first_proc(fp);
4591 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4592 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4593 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4594 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4595 
4596 #undef XPTRWIDTH
4597 }
4598 
4599 DB_SHOW_COMMAND(file, db_show_file)
4600 {
4601 	struct file *fp;
4602 
4603 	if (!have_addr) {
4604 		db_printf("usage: show file <addr>\n");
4605 		return;
4606 	}
4607 	fp = (struct file *)addr;
4608 	db_print_file(fp, 1);
4609 }
4610 
4611 DB_SHOW_COMMAND(files, db_show_files)
4612 {
4613 	struct filedesc *fdp;
4614 	struct file *fp;
4615 	struct proc *p;
4616 	int header;
4617 	int n;
4618 
4619 	header = 1;
4620 	FOREACH_PROC_IN_SYSTEM(p) {
4621 		if (p->p_state == PRS_NEW)
4622 			continue;
4623 		if ((fdp = p->p_fd) == NULL)
4624 			continue;
4625 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4626 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4627 				continue;
4628 			db_print_file(fp, header);
4629 			header = 0;
4630 		}
4631 	}
4632 }
4633 #endif
4634 
4635 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4636     &maxfilesperproc, 0, "Maximum files allowed open per process");
4637 
4638 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4639     &maxfiles, 0, "Maximum number of files");
4640 
4641 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4642     &openfiles, 0, "System-wide number of open files");
4643 
4644 /* ARGSUSED*/
4645 static void
4646 filelistinit(void *dummy)
4647 {
4648 
4649 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4650 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4651 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4652 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4653 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4654 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4655 	/*
4656 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4657 	 * the vnode zone.
4658 	 */
4659 	vfs_smr = uma_zone_get_smr(pwd_zone);
4660 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4661 }
4662 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4663 
4664 /*-------------------------------------------------------------------*/
4665 
4666 static int
4667 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4668     int flags, struct thread *td)
4669 {
4670 
4671 	return (EBADF);
4672 }
4673 
4674 static int
4675 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4676     struct thread *td)
4677 {
4678 
4679 	return (EINVAL);
4680 }
4681 
4682 static int
4683 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4684     struct thread *td)
4685 {
4686 
4687 	return (EBADF);
4688 }
4689 
4690 static int
4691 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4692     struct thread *td)
4693 {
4694 
4695 	return (0);
4696 }
4697 
4698 static int
4699 badfo_kqfilter(struct file *fp, struct knote *kn)
4700 {
4701 
4702 	return (EBADF);
4703 }
4704 
4705 static int
4706 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4707     struct thread *td)
4708 {
4709 
4710 	return (EBADF);
4711 }
4712 
4713 static int
4714 badfo_close(struct file *fp, struct thread *td)
4715 {
4716 
4717 	return (0);
4718 }
4719 
4720 static int
4721 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4722     struct thread *td)
4723 {
4724 
4725 	return (EBADF);
4726 }
4727 
4728 static int
4729 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4730     struct thread *td)
4731 {
4732 
4733 	return (EBADF);
4734 }
4735 
4736 static int
4737 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4738     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4739     struct thread *td)
4740 {
4741 
4742 	return (EBADF);
4743 }
4744 
4745 static int
4746 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4747 {
4748 
4749 	return (0);
4750 }
4751 
4752 struct fileops badfileops = {
4753 	.fo_read = badfo_readwrite,
4754 	.fo_write = badfo_readwrite,
4755 	.fo_truncate = badfo_truncate,
4756 	.fo_ioctl = badfo_ioctl,
4757 	.fo_poll = badfo_poll,
4758 	.fo_kqfilter = badfo_kqfilter,
4759 	.fo_stat = badfo_stat,
4760 	.fo_close = badfo_close,
4761 	.fo_chmod = badfo_chmod,
4762 	.fo_chown = badfo_chown,
4763 	.fo_sendfile = badfo_sendfile,
4764 	.fo_fill_kinfo = badfo_fill_kinfo,
4765 };
4766 
4767 int
4768 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4769     int flags, struct thread *td)
4770 {
4771 
4772 	return (EOPNOTSUPP);
4773 }
4774 
4775 int
4776 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4777     struct thread *td)
4778 {
4779 
4780 	return (EINVAL);
4781 }
4782 
4783 int
4784 invfo_ioctl(struct file *fp, u_long com, void *data,
4785     struct ucred *active_cred, struct thread *td)
4786 {
4787 
4788 	return (ENOTTY);
4789 }
4790 
4791 int
4792 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4793     struct thread *td)
4794 {
4795 
4796 	return (poll_no_poll(events));
4797 }
4798 
4799 int
4800 invfo_kqfilter(struct file *fp, struct knote *kn)
4801 {
4802 
4803 	return (EINVAL);
4804 }
4805 
4806 int
4807 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4808     struct thread *td)
4809 {
4810 
4811 	return (EINVAL);
4812 }
4813 
4814 int
4815 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4816     struct thread *td)
4817 {
4818 
4819 	return (EINVAL);
4820 }
4821 
4822 int
4823 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4824     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4825     struct thread *td)
4826 {
4827 
4828 	return (EINVAL);
4829 }
4830 
4831 /*-------------------------------------------------------------------*/
4832 
4833 /*
4834  * File Descriptor pseudo-device driver (/dev/fd/).
4835  *
4836  * Opening minor device N dup()s the file (if any) connected to file
4837  * descriptor N belonging to the calling process.  Note that this driver
4838  * consists of only the ``open()'' routine, because all subsequent
4839  * references to this file will be direct to the other driver.
4840  *
4841  * XXX: we could give this one a cloning event handler if necessary.
4842  */
4843 
4844 /* ARGSUSED */
4845 static int
4846 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4847 {
4848 
4849 	/*
4850 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4851 	 * the file descriptor being sought for duplication. The error
4852 	 * return ensures that the vnode for this device will be released
4853 	 * by vn_open. Open will detect this special error and take the
4854 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4855 	 * will simply report the error.
4856 	 */
4857 	td->td_dupfd = dev2unit(dev);
4858 	return (ENODEV);
4859 }
4860 
4861 static struct cdevsw fildesc_cdevsw = {
4862 	.d_version =	D_VERSION,
4863 	.d_open =	fdopen,
4864 	.d_name =	"FD",
4865 };
4866 
4867 static void
4868 fildesc_drvinit(void *unused)
4869 {
4870 	struct cdev *dev;
4871 
4872 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4873 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4874 	make_dev_alias(dev, "stdin");
4875 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4876 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4877 	make_dev_alias(dev, "stdout");
4878 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4879 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4880 	make_dev_alias(dev, "stderr");
4881 }
4882 
4883 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4884