xref: /freebsd/sys/kern/kern_descrip.c (revision e91d723ad446b5429318b24f4578a4f7a160f65e)
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/stat.h>
73 #include <sys/sx.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/unistd.h>
78 #include <sys/user.h>
79 #include <sys/vnode.h>
80 #ifdef KTRACE
81 #include <sys/ktrace.h>
82 #endif
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
95     "file desc to leader structures");
96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
98 
99 MALLOC_DECLARE(M_FADVISE);
100 
101 static __read_mostly uma_zone_t file_zone;
102 static __read_mostly uma_zone_t filedesc0_zone;
103 
104 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
105 		    struct thread *td, int holdleaders);
106 static int	fd_first_free(struct filedesc *fdp, int low, int size);
107 static int	fd_last_used(struct filedesc *fdp, int size);
108 static void	fdgrowtable(struct filedesc *fdp, int nfd);
109 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void	fdunused(struct filedesc *fdp, int fd);
111 static void	fdused(struct filedesc *fdp, int fd);
112 static int	getmaxfd(struct thread *td);
113 static u_long	*filecaps_copy_prep(const struct filecaps *src);
114 static void	filecaps_copy_finish(const struct filecaps *src,
115 		    struct filecaps *dst, u_long *ioctls);
116 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
117 static void	filecaps_free_finish(u_long *ioctls);
118 
119 /*
120  * Each process has:
121  *
122  * - An array of open file descriptors (fd_ofiles)
123  * - An array of file flags (fd_ofileflags)
124  * - A bitmap recording which descriptors are in use (fd_map)
125  *
126  * A process starts out with NDFILE descriptors.  The value of NDFILE has
127  * been selected based the historical limit of 20 open files, and an
128  * assumption that the majority of processes, especially short-lived
129  * processes like shells, will never need more.
130  *
131  * If this initial allocation is exhausted, a larger descriptor table and
132  * map are allocated dynamically, and the pointers in the process's struct
133  * filedesc are updated to point to those.  This is repeated every time
134  * the process runs out of file descriptors (provided it hasn't hit its
135  * resource limit).
136  *
137  * Since threads may hold references to individual descriptor table
138  * entries, the tables are never freed.  Instead, they are placed on a
139  * linked list and freed only when the struct filedesc is released.
140  */
141 #define NDFILE		20
142 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
143 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x)	((x) / NDENTRIES)
145 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
147 
148 /*
149  * SLIST entry used to keep track of ofiles which must be reclaimed when
150  * the process exits.
151  */
152 struct freetable {
153 	struct fdescenttbl *ft_table;
154 	SLIST_ENTRY(freetable) ft_next;
155 };
156 
157 /*
158  * Initial allocation: a filedesc structure + the head of SLIST used to
159  * keep track of old ofiles + enough space for NDFILE descriptors.
160  */
161 
162 struct fdescenttbl0 {
163 	int	fdt_nfiles;
164 	struct	filedescent fdt_ofiles[NDFILE];
165 };
166 
167 struct filedesc0 {
168 	struct filedesc fd_fd;
169 	SLIST_HEAD(, freetable) fd_free;
170 	struct	fdescenttbl0 fd_dfiles;
171 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
172 };
173 
174 /*
175  * Descriptor management.
176  */
177 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
178 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
180 
181 /*
182  * If low >= size, just return low. Otherwise find the first zero bit in the
183  * given bitmap, starting at low and not exceeding size - 1. Return size if
184  * not found.
185  */
186 static int
187 fd_first_free(struct filedesc *fdp, int low, int size)
188 {
189 	NDSLOTTYPE *map = fdp->fd_map;
190 	NDSLOTTYPE mask;
191 	int off, maxoff;
192 
193 	if (low >= size)
194 		return (low);
195 
196 	off = NDSLOT(low);
197 	if (low % NDENTRIES) {
198 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
199 		if ((mask &= ~map[off]) != 0UL)
200 			return (off * NDENTRIES + ffsl(mask) - 1);
201 		++off;
202 	}
203 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
204 		if (map[off] != ~0UL)
205 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
206 	return (size);
207 }
208 
209 /*
210  * Find the highest non-zero bit in the given bitmap, starting at 0 and
211  * not exceeding size - 1. Return -1 if not found.
212  */
213 static int
214 fd_last_used(struct filedesc *fdp, int size)
215 {
216 	NDSLOTTYPE *map = fdp->fd_map;
217 	NDSLOTTYPE mask;
218 	int off, minoff;
219 
220 	off = NDSLOT(size);
221 	if (size % NDENTRIES) {
222 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
223 		if ((mask &= map[off]) != 0)
224 			return (off * NDENTRIES + flsl(mask) - 1);
225 		--off;
226 	}
227 	for (minoff = NDSLOT(0); off >= minoff; --off)
228 		if (map[off] != 0)
229 			return (off * NDENTRIES + flsl(map[off]) - 1);
230 	return (-1);
231 }
232 
233 static int
234 fdisused(struct filedesc *fdp, int fd)
235 {
236 
237 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
238 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
239 
240 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
241 }
242 
243 /*
244  * Mark a file descriptor as used.
245  */
246 static void
247 fdused_init(struct filedesc *fdp, int fd)
248 {
249 
250 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
251 
252 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
253 }
254 
255 static void
256 fdused(struct filedesc *fdp, int fd)
257 {
258 
259 	FILEDESC_XLOCK_ASSERT(fdp);
260 
261 	fdused_init(fdp, fd);
262 	if (fd > fdp->fd_lastfile)
263 		fdp->fd_lastfile = fd;
264 	if (fd == fdp->fd_freefile)
265 		fdp->fd_freefile++;
266 }
267 
268 /*
269  * Mark a file descriptor as unused.
270  */
271 static void
272 fdunused(struct filedesc *fdp, int fd)
273 {
274 
275 	FILEDESC_XLOCK_ASSERT(fdp);
276 
277 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
278 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
279 	    ("fd=%d is still in use", fd));
280 
281 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
282 	if (fd < fdp->fd_freefile)
283 		fdp->fd_freefile = fd;
284 	if (fd == fdp->fd_lastfile)
285 		fdp->fd_lastfile = fd_last_used(fdp, fd);
286 }
287 
288 /*
289  * Free a file descriptor.
290  *
291  * Avoid some work if fdp is about to be destroyed.
292  */
293 static inline void
294 fdefree_last(struct filedescent *fde)
295 {
296 
297 	filecaps_free(&fde->fde_caps);
298 }
299 
300 static inline void
301 fdfree(struct filedesc *fdp, int fd)
302 {
303 	struct filedescent *fde;
304 
305 	fde = &fdp->fd_ofiles[fd];
306 #ifdef CAPABILITIES
307 	seqc_write_begin(&fde->fde_seqc);
308 #endif
309 	fde->fde_file = NULL;
310 #ifdef CAPABILITIES
311 	seqc_write_end(&fde->fde_seqc);
312 #endif
313 	fdefree_last(fde);
314 	fdunused(fdp, fd);
315 }
316 
317 void
318 pwd_ensure_dirs(void)
319 {
320 	struct filedesc *fdp;
321 
322 	fdp = curproc->p_fd;
323 	FILEDESC_XLOCK(fdp);
324 	if (fdp->fd_cdir == NULL) {
325 		fdp->fd_cdir = rootvnode;
326 		vrefact(rootvnode);
327 	}
328 	if (fdp->fd_rdir == NULL) {
329 		fdp->fd_rdir = rootvnode;
330 		vrefact(rootvnode);
331 	}
332 	FILEDESC_XUNLOCK(fdp);
333 }
334 
335 /*
336  * System calls on descriptors.
337  */
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
340 	int	dummy;
341 };
342 #endif
343 /* ARGSUSED */
344 int
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
346 {
347 #ifdef	RACCT
348 	uint64_t lim;
349 #endif
350 
351 	td->td_retval[0] = getmaxfd(td);
352 #ifdef	RACCT
353 	PROC_LOCK(td->td_proc);
354 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
355 	PROC_UNLOCK(td->td_proc);
356 	if (lim < td->td_retval[0])
357 		td->td_retval[0] = lim;
358 #endif
359 	return (0);
360 }
361 
362 /*
363  * Duplicate a file descriptor to a particular value.
364  *
365  * Note: keep in mind that a potential race condition exists when closing
366  * descriptors from a shared descriptor table (via rfork).
367  */
368 #ifndef _SYS_SYSPROTO_H_
369 struct dup2_args {
370 	u_int	from;
371 	u_int	to;
372 };
373 #endif
374 /* ARGSUSED */
375 int
376 sys_dup2(struct thread *td, struct dup2_args *uap)
377 {
378 
379 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
380 }
381 
382 /*
383  * Duplicate a file descriptor.
384  */
385 #ifndef _SYS_SYSPROTO_H_
386 struct dup_args {
387 	u_int	fd;
388 };
389 #endif
390 /* ARGSUSED */
391 int
392 sys_dup(struct thread *td, struct dup_args *uap)
393 {
394 
395 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
396 }
397 
398 /*
399  * The file control system call.
400  */
401 #ifndef _SYS_SYSPROTO_H_
402 struct fcntl_args {
403 	int	fd;
404 	int	cmd;
405 	long	arg;
406 };
407 #endif
408 /* ARGSUSED */
409 int
410 sys_fcntl(struct thread *td, struct fcntl_args *uap)
411 {
412 
413 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
414 }
415 
416 int
417 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
418 {
419 	struct flock fl;
420 	struct __oflock ofl;
421 	intptr_t arg1;
422 	int error, newcmd;
423 
424 	error = 0;
425 	newcmd = cmd;
426 	switch (cmd) {
427 	case F_OGETLK:
428 	case F_OSETLK:
429 	case F_OSETLKW:
430 		/*
431 		 * Convert old flock structure to new.
432 		 */
433 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
434 		fl.l_start = ofl.l_start;
435 		fl.l_len = ofl.l_len;
436 		fl.l_pid = ofl.l_pid;
437 		fl.l_type = ofl.l_type;
438 		fl.l_whence = ofl.l_whence;
439 		fl.l_sysid = 0;
440 
441 		switch (cmd) {
442 		case F_OGETLK:
443 			newcmd = F_GETLK;
444 			break;
445 		case F_OSETLK:
446 			newcmd = F_SETLK;
447 			break;
448 		case F_OSETLKW:
449 			newcmd = F_SETLKW;
450 			break;
451 		}
452 		arg1 = (intptr_t)&fl;
453 		break;
454 	case F_GETLK:
455 	case F_SETLK:
456 	case F_SETLKW:
457 	case F_SETLK_REMOTE:
458 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
459 		arg1 = (intptr_t)&fl;
460 		break;
461 	default:
462 		arg1 = arg;
463 		break;
464 	}
465 	if (error)
466 		return (error);
467 	error = kern_fcntl(td, fd, newcmd, arg1);
468 	if (error)
469 		return (error);
470 	if (cmd == F_OGETLK) {
471 		ofl.l_start = fl.l_start;
472 		ofl.l_len = fl.l_len;
473 		ofl.l_pid = fl.l_pid;
474 		ofl.l_type = fl.l_type;
475 		ofl.l_whence = fl.l_whence;
476 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
477 	} else if (cmd == F_GETLK) {
478 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
479 	}
480 	return (error);
481 }
482 
483 int
484 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
485 {
486 	struct filedesc *fdp;
487 	struct flock *flp;
488 	struct file *fp, *fp2;
489 	struct filedescent *fde;
490 	struct proc *p;
491 	struct vnode *vp;
492 	int error, flg, tmp;
493 	uint64_t bsize;
494 	off_t foffset;
495 
496 	error = 0;
497 	flg = F_POSIX;
498 	p = td->td_proc;
499 	fdp = p->p_fd;
500 
501 	AUDIT_ARG_FD(cmd);
502 	AUDIT_ARG_CMD(cmd);
503 	switch (cmd) {
504 	case F_DUPFD:
505 		tmp = arg;
506 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
507 		break;
508 
509 	case F_DUPFD_CLOEXEC:
510 		tmp = arg;
511 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
512 		break;
513 
514 	case F_DUP2FD:
515 		tmp = arg;
516 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
517 		break;
518 
519 	case F_DUP2FD_CLOEXEC:
520 		tmp = arg;
521 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
522 		break;
523 
524 	case F_GETFD:
525 		error = EBADF;
526 		FILEDESC_SLOCK(fdp);
527 		fde = fdeget_locked(fdp, fd);
528 		if (fde != NULL) {
529 			td->td_retval[0] =
530 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
531 			error = 0;
532 		}
533 		FILEDESC_SUNLOCK(fdp);
534 		break;
535 
536 	case F_SETFD:
537 		error = EBADF;
538 		FILEDESC_XLOCK(fdp);
539 		fde = fdeget_locked(fdp, fd);
540 		if (fde != NULL) {
541 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
542 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
543 			error = 0;
544 		}
545 		FILEDESC_XUNLOCK(fdp);
546 		break;
547 
548 	case F_GETFL:
549 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
550 		if (error != 0)
551 			break;
552 		td->td_retval[0] = OFLAGS(fp->f_flag);
553 		fdrop(fp, td);
554 		break;
555 
556 	case F_SETFL:
557 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
558 		if (error != 0)
559 			break;
560 		do {
561 			tmp = flg = fp->f_flag;
562 			tmp &= ~FCNTLFLAGS;
563 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
564 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
565 		tmp = fp->f_flag & FNONBLOCK;
566 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
567 		if (error != 0) {
568 			fdrop(fp, td);
569 			break;
570 		}
571 		tmp = fp->f_flag & FASYNC;
572 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
573 		if (error == 0) {
574 			fdrop(fp, td);
575 			break;
576 		}
577 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
578 		tmp = 0;
579 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
580 		fdrop(fp, td);
581 		break;
582 
583 	case F_GETOWN:
584 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
585 		if (error != 0)
586 			break;
587 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
588 		if (error == 0)
589 			td->td_retval[0] = tmp;
590 		fdrop(fp, td);
591 		break;
592 
593 	case F_SETOWN:
594 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
595 		if (error != 0)
596 			break;
597 		tmp = arg;
598 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
599 		fdrop(fp, td);
600 		break;
601 
602 	case F_SETLK_REMOTE:
603 		error = priv_check(td, PRIV_NFS_LOCKD);
604 		if (error != 0)
605 			return (error);
606 		flg = F_REMOTE;
607 		goto do_setlk;
608 
609 	case F_SETLKW:
610 		flg |= F_WAIT;
611 		/* FALLTHROUGH F_SETLK */
612 
613 	case F_SETLK:
614 	do_setlk:
615 		flp = (struct flock *)arg;
616 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
617 			error = EINVAL;
618 			break;
619 		}
620 
621 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
622 		if (error != 0)
623 			break;
624 		if (fp->f_type != DTYPE_VNODE) {
625 			error = EBADF;
626 			fdrop(fp, td);
627 			break;
628 		}
629 
630 		if (flp->l_whence == SEEK_CUR) {
631 			foffset = foffset_get(fp);
632 			if (foffset < 0 ||
633 			    (flp->l_start > 0 &&
634 			     foffset > OFF_MAX - flp->l_start)) {
635 				error = EOVERFLOW;
636 				fdrop(fp, td);
637 				break;
638 			}
639 			flp->l_start += foffset;
640 		}
641 
642 		vp = fp->f_vnode;
643 		switch (flp->l_type) {
644 		case F_RDLCK:
645 			if ((fp->f_flag & FREAD) == 0) {
646 				error = EBADF;
647 				break;
648 			}
649 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
650 				PROC_LOCK(p->p_leader);
651 				p->p_leader->p_flag |= P_ADVLOCK;
652 				PROC_UNLOCK(p->p_leader);
653 			}
654 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
655 			    flp, flg);
656 			break;
657 		case F_WRLCK:
658 			if ((fp->f_flag & FWRITE) == 0) {
659 				error = EBADF;
660 				break;
661 			}
662 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
663 				PROC_LOCK(p->p_leader);
664 				p->p_leader->p_flag |= P_ADVLOCK;
665 				PROC_UNLOCK(p->p_leader);
666 			}
667 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
668 			    flp, flg);
669 			break;
670 		case F_UNLCK:
671 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
672 			    flp, flg);
673 			break;
674 		case F_UNLCKSYS:
675 			if (flg != F_REMOTE) {
676 				error = EINVAL;
677 				break;
678 			}
679 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
680 			    F_UNLCKSYS, flp, flg);
681 			break;
682 		default:
683 			error = EINVAL;
684 			break;
685 		}
686 		if (error != 0 || flp->l_type == F_UNLCK ||
687 		    flp->l_type == F_UNLCKSYS) {
688 			fdrop(fp, td);
689 			break;
690 		}
691 
692 		/*
693 		 * Check for a race with close.
694 		 *
695 		 * The vnode is now advisory locked (or unlocked, but this case
696 		 * is not really important) as the caller requested.
697 		 * We had to drop the filedesc lock, so we need to recheck if
698 		 * the descriptor is still valid, because if it was closed
699 		 * in the meantime we need to remove advisory lock from the
700 		 * vnode - close on any descriptor leading to an advisory
701 		 * locked vnode, removes that lock.
702 		 * We will return 0 on purpose in that case, as the result of
703 		 * successful advisory lock might have been externally visible
704 		 * already. This is fine - effectively we pretend to the caller
705 		 * that the closing thread was a bit slower and that the
706 		 * advisory lock succeeded before the close.
707 		 */
708 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
709 		if (error != 0) {
710 			fdrop(fp, td);
711 			break;
712 		}
713 		if (fp != fp2) {
714 			flp->l_whence = SEEK_SET;
715 			flp->l_start = 0;
716 			flp->l_len = 0;
717 			flp->l_type = F_UNLCK;
718 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
719 			    F_UNLCK, flp, F_POSIX);
720 		}
721 		fdrop(fp, td);
722 		fdrop(fp2, td);
723 		break;
724 
725 	case F_GETLK:
726 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
727 		if (error != 0)
728 			break;
729 		if (fp->f_type != DTYPE_VNODE) {
730 			error = EBADF;
731 			fdrop(fp, td);
732 			break;
733 		}
734 		flp = (struct flock *)arg;
735 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
736 		    flp->l_type != F_UNLCK) {
737 			error = EINVAL;
738 			fdrop(fp, td);
739 			break;
740 		}
741 		if (flp->l_whence == SEEK_CUR) {
742 			foffset = foffset_get(fp);
743 			if ((flp->l_start > 0 &&
744 			    foffset > OFF_MAX - flp->l_start) ||
745 			    (flp->l_start < 0 &&
746 			    foffset < OFF_MIN - flp->l_start)) {
747 				error = EOVERFLOW;
748 				fdrop(fp, td);
749 				break;
750 			}
751 			flp->l_start += foffset;
752 		}
753 		vp = fp->f_vnode;
754 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
755 		    F_POSIX);
756 		fdrop(fp, td);
757 		break;
758 
759 	case F_RDAHEAD:
760 		arg = arg ? 128 * 1024: 0;
761 		/* FALLTHROUGH */
762 	case F_READAHEAD:
763 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
764 		if (error != 0)
765 			break;
766 		if (fp->f_type != DTYPE_VNODE) {
767 			fdrop(fp, td);
768 			error = EBADF;
769 			break;
770 		}
771 		vp = fp->f_vnode;
772 		/*
773 		 * Exclusive lock synchronizes against f_seqcount reads and
774 		 * writes in sequential_heuristic().
775 		 */
776 		error = vn_lock(vp, LK_EXCLUSIVE);
777 		if (error != 0) {
778 			fdrop(fp, td);
779 			break;
780 		}
781 		if (arg >= 0) {
782 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
783 			arg = MIN(arg, INT_MAX - bsize + 1);
784 			fp->f_seqcount = MIN(IO_SEQMAX,
785 			    (arg + bsize - 1) / bsize);
786 			atomic_set_int(&fp->f_flag, FRDAHEAD);
787 		} else {
788 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
789 		}
790 		VOP_UNLOCK(vp, 0);
791 		fdrop(fp, td);
792 		break;
793 
794 	default:
795 		error = EINVAL;
796 		break;
797 	}
798 	return (error);
799 }
800 
801 static int
802 getmaxfd(struct thread *td)
803 {
804 
805 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
806 }
807 
808 /*
809  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
810  */
811 int
812 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
813 {
814 	struct filedesc *fdp;
815 	struct filedescent *oldfde, *newfde;
816 	struct proc *p;
817 	struct file *delfp;
818 	u_long *oioctls, *nioctls;
819 	int error, maxfd;
820 
821 	p = td->td_proc;
822 	fdp = p->p_fd;
823 	oioctls = NULL;
824 
825 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
826 	MPASS(mode < FDDUP_LASTMODE);
827 
828 	AUDIT_ARG_FD(old);
829 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
830 
831 	/*
832 	 * Verify we have a valid descriptor to dup from and possibly to
833 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
834 	 * return EINVAL when the new descriptor is out of bounds.
835 	 */
836 	if (old < 0)
837 		return (EBADF);
838 	if (new < 0)
839 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
840 	maxfd = getmaxfd(td);
841 	if (new >= maxfd)
842 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
843 
844 	error = EBADF;
845 	FILEDESC_XLOCK(fdp);
846 	if (fget_locked(fdp, old) == NULL)
847 		goto unlock;
848 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
849 		td->td_retval[0] = new;
850 		if (flags & FDDUP_FLAG_CLOEXEC)
851 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
852 		error = 0;
853 		goto unlock;
854 	}
855 
856 	oldfde = &fdp->fd_ofiles[old];
857 	if (!fhold(oldfde->fde_file))
858 		goto unlock;
859 
860 	/*
861 	 * If the caller specified a file descriptor, make sure the file
862 	 * table is large enough to hold it, and grab it.  Otherwise, just
863 	 * allocate a new descriptor the usual way.
864 	 */
865 	switch (mode) {
866 	case FDDUP_NORMAL:
867 	case FDDUP_FCNTL:
868 		if ((error = fdalloc(td, new, &new)) != 0) {
869 			fdrop(oldfde->fde_file, td);
870 			goto unlock;
871 		}
872 		break;
873 	case FDDUP_MUSTREPLACE:
874 		/* Target file descriptor must exist. */
875 		if (fget_locked(fdp, new) == NULL) {
876 			fdrop(oldfde->fde_file, td);
877 			goto unlock;
878 		}
879 		break;
880 	case FDDUP_FIXED:
881 		if (new >= fdp->fd_nfiles) {
882 			/*
883 			 * The resource limits are here instead of e.g.
884 			 * fdalloc(), because the file descriptor table may be
885 			 * shared between processes, so we can't really use
886 			 * racct_add()/racct_sub().  Instead of counting the
887 			 * number of actually allocated descriptors, just put
888 			 * the limit on the size of the file descriptor table.
889 			 */
890 #ifdef RACCT
891 			if (RACCT_ENABLED()) {
892 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
893 				if (error != 0) {
894 					error = EMFILE;
895 					fdrop(oldfde->fde_file, td);
896 					goto unlock;
897 				}
898 			}
899 #endif
900 			fdgrowtable_exp(fdp, new + 1);
901 		}
902 		if (!fdisused(fdp, new))
903 			fdused(fdp, new);
904 		break;
905 	default:
906 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
907 	}
908 
909 	KASSERT(old != new, ("new fd is same as old"));
910 
911 	newfde = &fdp->fd_ofiles[new];
912 	delfp = newfde->fde_file;
913 
914 	oioctls = filecaps_free_prep(&newfde->fde_caps);
915 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
916 
917 	/*
918 	 * Duplicate the source descriptor.
919 	 */
920 #ifdef CAPABILITIES
921 	seqc_write_begin(&newfde->fde_seqc);
922 #endif
923 	memcpy(newfde, oldfde, fde_change_size);
924 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
925 	    nioctls);
926 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
927 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
928 	else
929 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
930 #ifdef CAPABILITIES
931 	seqc_write_end(&newfde->fde_seqc);
932 #endif
933 	td->td_retval[0] = new;
934 
935 	error = 0;
936 
937 	if (delfp != NULL) {
938 		(void) closefp(fdp, new, delfp, td, 1);
939 		FILEDESC_UNLOCK_ASSERT(fdp);
940 	} else {
941 unlock:
942 		FILEDESC_XUNLOCK(fdp);
943 	}
944 
945 	filecaps_free_finish(oioctls);
946 	return (error);
947 }
948 
949 /*
950  * If sigio is on the list associated with a process or process group,
951  * disable signalling from the device, remove sigio from the list and
952  * free sigio.
953  */
954 void
955 funsetown(struct sigio **sigiop)
956 {
957 	struct sigio *sigio;
958 
959 	if (*sigiop == NULL)
960 		return;
961 	SIGIO_LOCK();
962 	sigio = *sigiop;
963 	if (sigio == NULL) {
964 		SIGIO_UNLOCK();
965 		return;
966 	}
967 	*(sigio->sio_myref) = NULL;
968 	if ((sigio)->sio_pgid < 0) {
969 		struct pgrp *pg = (sigio)->sio_pgrp;
970 		PGRP_LOCK(pg);
971 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
972 			    sigio, sio_pgsigio);
973 		PGRP_UNLOCK(pg);
974 	} else {
975 		struct proc *p = (sigio)->sio_proc;
976 		PROC_LOCK(p);
977 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
978 			    sigio, sio_pgsigio);
979 		PROC_UNLOCK(p);
980 	}
981 	SIGIO_UNLOCK();
982 	crfree(sigio->sio_ucred);
983 	free(sigio, M_SIGIO);
984 }
985 
986 /*
987  * Free a list of sigio structures.
988  * We only need to lock the SIGIO_LOCK because we have made ourselves
989  * inaccessible to callers of fsetown and therefore do not need to lock
990  * the proc or pgrp struct for the list manipulation.
991  */
992 void
993 funsetownlst(struct sigiolst *sigiolst)
994 {
995 	struct proc *p;
996 	struct pgrp *pg;
997 	struct sigio *sigio;
998 
999 	sigio = SLIST_FIRST(sigiolst);
1000 	if (sigio == NULL)
1001 		return;
1002 	p = NULL;
1003 	pg = NULL;
1004 
1005 	/*
1006 	 * Every entry of the list should belong
1007 	 * to a single proc or pgrp.
1008 	 */
1009 	if (sigio->sio_pgid < 0) {
1010 		pg = sigio->sio_pgrp;
1011 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1012 	} else /* if (sigio->sio_pgid > 0) */ {
1013 		p = sigio->sio_proc;
1014 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1015 	}
1016 
1017 	SIGIO_LOCK();
1018 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1019 		*(sigio->sio_myref) = NULL;
1020 		if (pg != NULL) {
1021 			KASSERT(sigio->sio_pgid < 0,
1022 			    ("Proc sigio in pgrp sigio list"));
1023 			KASSERT(sigio->sio_pgrp == pg,
1024 			    ("Bogus pgrp in sigio list"));
1025 			PGRP_LOCK(pg);
1026 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1027 			    sio_pgsigio);
1028 			PGRP_UNLOCK(pg);
1029 		} else /* if (p != NULL) */ {
1030 			KASSERT(sigio->sio_pgid > 0,
1031 			    ("Pgrp sigio in proc sigio list"));
1032 			KASSERT(sigio->sio_proc == p,
1033 			    ("Bogus proc in sigio list"));
1034 			PROC_LOCK(p);
1035 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1036 			    sio_pgsigio);
1037 			PROC_UNLOCK(p);
1038 		}
1039 		SIGIO_UNLOCK();
1040 		crfree(sigio->sio_ucred);
1041 		free(sigio, M_SIGIO);
1042 		SIGIO_LOCK();
1043 	}
1044 	SIGIO_UNLOCK();
1045 }
1046 
1047 /*
1048  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1049  *
1050  * After permission checking, add a sigio structure to the sigio list for
1051  * the process or process group.
1052  */
1053 int
1054 fsetown(pid_t pgid, struct sigio **sigiop)
1055 {
1056 	struct proc *proc;
1057 	struct pgrp *pgrp;
1058 	struct sigio *sigio;
1059 	int ret;
1060 
1061 	if (pgid == 0) {
1062 		funsetown(sigiop);
1063 		return (0);
1064 	}
1065 
1066 	ret = 0;
1067 
1068 	/* Allocate and fill in the new sigio out of locks. */
1069 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1070 	sigio->sio_pgid = pgid;
1071 	sigio->sio_ucred = crhold(curthread->td_ucred);
1072 	sigio->sio_myref = sigiop;
1073 
1074 	sx_slock(&proctree_lock);
1075 	if (pgid > 0) {
1076 		proc = pfind(pgid);
1077 		if (proc == NULL) {
1078 			ret = ESRCH;
1079 			goto fail;
1080 		}
1081 
1082 		/*
1083 		 * Policy - Don't allow a process to FSETOWN a process
1084 		 * in another session.
1085 		 *
1086 		 * Remove this test to allow maximum flexibility or
1087 		 * restrict FSETOWN to the current process or process
1088 		 * group for maximum safety.
1089 		 */
1090 		PROC_UNLOCK(proc);
1091 		if (proc->p_session != curthread->td_proc->p_session) {
1092 			ret = EPERM;
1093 			goto fail;
1094 		}
1095 
1096 		pgrp = NULL;
1097 	} else /* if (pgid < 0) */ {
1098 		pgrp = pgfind(-pgid);
1099 		if (pgrp == NULL) {
1100 			ret = ESRCH;
1101 			goto fail;
1102 		}
1103 		PGRP_UNLOCK(pgrp);
1104 
1105 		/*
1106 		 * Policy - Don't allow a process to FSETOWN a process
1107 		 * in another session.
1108 		 *
1109 		 * Remove this test to allow maximum flexibility or
1110 		 * restrict FSETOWN to the current process or process
1111 		 * group for maximum safety.
1112 		 */
1113 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1114 			ret = EPERM;
1115 			goto fail;
1116 		}
1117 
1118 		proc = NULL;
1119 	}
1120 	funsetown(sigiop);
1121 	if (pgid > 0) {
1122 		PROC_LOCK(proc);
1123 		/*
1124 		 * Since funsetownlst() is called without the proctree
1125 		 * locked, we need to check for P_WEXIT.
1126 		 * XXX: is ESRCH correct?
1127 		 */
1128 		if ((proc->p_flag & P_WEXIT) != 0) {
1129 			PROC_UNLOCK(proc);
1130 			ret = ESRCH;
1131 			goto fail;
1132 		}
1133 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1134 		sigio->sio_proc = proc;
1135 		PROC_UNLOCK(proc);
1136 	} else {
1137 		PGRP_LOCK(pgrp);
1138 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1139 		sigio->sio_pgrp = pgrp;
1140 		PGRP_UNLOCK(pgrp);
1141 	}
1142 	sx_sunlock(&proctree_lock);
1143 	SIGIO_LOCK();
1144 	*sigiop = sigio;
1145 	SIGIO_UNLOCK();
1146 	return (0);
1147 
1148 fail:
1149 	sx_sunlock(&proctree_lock);
1150 	crfree(sigio->sio_ucred);
1151 	free(sigio, M_SIGIO);
1152 	return (ret);
1153 }
1154 
1155 /*
1156  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1157  */
1158 pid_t
1159 fgetown(struct sigio **sigiop)
1160 {
1161 	pid_t pgid;
1162 
1163 	SIGIO_LOCK();
1164 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1165 	SIGIO_UNLOCK();
1166 	return (pgid);
1167 }
1168 
1169 /*
1170  * Function drops the filedesc lock on return.
1171  */
1172 static int
1173 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1174     int holdleaders)
1175 {
1176 	int error;
1177 
1178 	FILEDESC_XLOCK_ASSERT(fdp);
1179 
1180 	if (holdleaders) {
1181 		if (td->td_proc->p_fdtol != NULL) {
1182 			/*
1183 			 * Ask fdfree() to sleep to ensure that all relevant
1184 			 * process leaders can be traversed in closef().
1185 			 */
1186 			fdp->fd_holdleaderscount++;
1187 		} else {
1188 			holdleaders = 0;
1189 		}
1190 	}
1191 
1192 	/*
1193 	 * We now hold the fp reference that used to be owned by the
1194 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1195 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1196 	 * added, and deleteing a knote for the new fd.
1197 	 */
1198 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1199 		knote_fdclose(td, fd);
1200 
1201 	/*
1202 	 * We need to notify mqueue if the object is of type mqueue.
1203 	 */
1204 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1205 		mq_fdclose(td, fd, fp);
1206 	FILEDESC_XUNLOCK(fdp);
1207 
1208 	error = closef(fp, td);
1209 	if (holdleaders) {
1210 		FILEDESC_XLOCK(fdp);
1211 		fdp->fd_holdleaderscount--;
1212 		if (fdp->fd_holdleaderscount == 0 &&
1213 		    fdp->fd_holdleaderswakeup != 0) {
1214 			fdp->fd_holdleaderswakeup = 0;
1215 			wakeup(&fdp->fd_holdleaderscount);
1216 		}
1217 		FILEDESC_XUNLOCK(fdp);
1218 	}
1219 	return (error);
1220 }
1221 
1222 /*
1223  * Close a file descriptor.
1224  */
1225 #ifndef _SYS_SYSPROTO_H_
1226 struct close_args {
1227 	int     fd;
1228 };
1229 #endif
1230 /* ARGSUSED */
1231 int
1232 sys_close(struct thread *td, struct close_args *uap)
1233 {
1234 
1235 	return (kern_close(td, uap->fd));
1236 }
1237 
1238 int
1239 kern_close(struct thread *td, int fd)
1240 {
1241 	struct filedesc *fdp;
1242 	struct file *fp;
1243 
1244 	fdp = td->td_proc->p_fd;
1245 
1246 	AUDIT_SYSCLOSE(td, fd);
1247 
1248 	FILEDESC_XLOCK(fdp);
1249 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1250 		FILEDESC_XUNLOCK(fdp);
1251 		return (EBADF);
1252 	}
1253 	fdfree(fdp, fd);
1254 
1255 	/* closefp() drops the FILEDESC lock for us. */
1256 	return (closefp(fdp, fd, fp, td, 1));
1257 }
1258 
1259 /*
1260  * Close open file descriptors.
1261  */
1262 #ifndef _SYS_SYSPROTO_H_
1263 struct closefrom_args {
1264 	int	lowfd;
1265 };
1266 #endif
1267 /* ARGSUSED */
1268 int
1269 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1270 {
1271 	struct filedesc *fdp;
1272 	int fd;
1273 
1274 	fdp = td->td_proc->p_fd;
1275 	AUDIT_ARG_FD(uap->lowfd);
1276 
1277 	/*
1278 	 * Treat negative starting file descriptor values identical to
1279 	 * closefrom(0) which closes all files.
1280 	 */
1281 	if (uap->lowfd < 0)
1282 		uap->lowfd = 0;
1283 	FILEDESC_SLOCK(fdp);
1284 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1285 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1286 			FILEDESC_SUNLOCK(fdp);
1287 			(void)kern_close(td, fd);
1288 			FILEDESC_SLOCK(fdp);
1289 		}
1290 	}
1291 	FILEDESC_SUNLOCK(fdp);
1292 	return (0);
1293 }
1294 
1295 #if defined(COMPAT_43)
1296 /*
1297  * Return status information about a file descriptor.
1298  */
1299 #ifndef _SYS_SYSPROTO_H_
1300 struct ofstat_args {
1301 	int	fd;
1302 	struct	ostat *sb;
1303 };
1304 #endif
1305 /* ARGSUSED */
1306 int
1307 ofstat(struct thread *td, struct ofstat_args *uap)
1308 {
1309 	struct ostat oub;
1310 	struct stat ub;
1311 	int error;
1312 
1313 	error = kern_fstat(td, uap->fd, &ub);
1314 	if (error == 0) {
1315 		cvtstat(&ub, &oub);
1316 		error = copyout(&oub, uap->sb, sizeof(oub));
1317 	}
1318 	return (error);
1319 }
1320 #endif /* COMPAT_43 */
1321 
1322 #if defined(COMPAT_FREEBSD11)
1323 int
1324 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1325 {
1326 	struct stat sb;
1327 	struct freebsd11_stat osb;
1328 	int error;
1329 
1330 	error = kern_fstat(td, uap->fd, &sb);
1331 	if (error != 0)
1332 		return (error);
1333 	error = freebsd11_cvtstat(&sb, &osb);
1334 	if (error == 0)
1335 		error = copyout(&osb, uap->sb, sizeof(osb));
1336 	return (error);
1337 }
1338 #endif	/* COMPAT_FREEBSD11 */
1339 
1340 /*
1341  * Return status information about a file descriptor.
1342  */
1343 #ifndef _SYS_SYSPROTO_H_
1344 struct fstat_args {
1345 	int	fd;
1346 	struct	stat *sb;
1347 };
1348 #endif
1349 /* ARGSUSED */
1350 int
1351 sys_fstat(struct thread *td, struct fstat_args *uap)
1352 {
1353 	struct stat ub;
1354 	int error;
1355 
1356 	error = kern_fstat(td, uap->fd, &ub);
1357 	if (error == 0)
1358 		error = copyout(&ub, uap->sb, sizeof(ub));
1359 	return (error);
1360 }
1361 
1362 int
1363 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1364 {
1365 	struct file *fp;
1366 	int error;
1367 
1368 	AUDIT_ARG_FD(fd);
1369 
1370 	error = fget(td, fd, &cap_fstat_rights, &fp);
1371 	if (error != 0)
1372 		return (error);
1373 
1374 	AUDIT_ARG_FILE(td->td_proc, fp);
1375 
1376 	error = fo_stat(fp, sbp, td->td_ucred, td);
1377 	fdrop(fp, td);
1378 #ifdef __STAT_TIME_T_EXT
1379 	if (error == 0) {
1380 		sbp->st_atim_ext = 0;
1381 		sbp->st_mtim_ext = 0;
1382 		sbp->st_ctim_ext = 0;
1383 		sbp->st_btim_ext = 0;
1384 	}
1385 #endif
1386 #ifdef KTRACE
1387 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1388 		ktrstat(sbp);
1389 #endif
1390 	return (error);
1391 }
1392 
1393 #if defined(COMPAT_FREEBSD11)
1394 /*
1395  * Return status information about a file descriptor.
1396  */
1397 #ifndef _SYS_SYSPROTO_H_
1398 struct freebsd11_nfstat_args {
1399 	int	fd;
1400 	struct	nstat *sb;
1401 };
1402 #endif
1403 /* ARGSUSED */
1404 int
1405 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1406 {
1407 	struct nstat nub;
1408 	struct stat ub;
1409 	int error;
1410 
1411 	error = kern_fstat(td, uap->fd, &ub);
1412 	if (error == 0) {
1413 		freebsd11_cvtnstat(&ub, &nub);
1414 		error = copyout(&nub, uap->sb, sizeof(nub));
1415 	}
1416 	return (error);
1417 }
1418 #endif /* COMPAT_FREEBSD11 */
1419 
1420 /*
1421  * Return pathconf information about a file descriptor.
1422  */
1423 #ifndef _SYS_SYSPROTO_H_
1424 struct fpathconf_args {
1425 	int	fd;
1426 	int	name;
1427 };
1428 #endif
1429 /* ARGSUSED */
1430 int
1431 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1432 {
1433 	long value;
1434 	int error;
1435 
1436 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1437 	if (error == 0)
1438 		td->td_retval[0] = value;
1439 	return (error);
1440 }
1441 
1442 int
1443 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1444 {
1445 	struct file *fp;
1446 	struct vnode *vp;
1447 	int error;
1448 
1449 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1450 	if (error != 0)
1451 		return (error);
1452 
1453 	if (name == _PC_ASYNC_IO) {
1454 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1455 		goto out;
1456 	}
1457 	vp = fp->f_vnode;
1458 	if (vp != NULL) {
1459 		vn_lock(vp, LK_SHARED | LK_RETRY);
1460 		error = VOP_PATHCONF(vp, name, valuep);
1461 		VOP_UNLOCK(vp, 0);
1462 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1463 		if (name != _PC_PIPE_BUF) {
1464 			error = EINVAL;
1465 		} else {
1466 			*valuep = PIPE_BUF;
1467 			error = 0;
1468 		}
1469 	} else {
1470 		error = EOPNOTSUPP;
1471 	}
1472 out:
1473 	fdrop(fp, td);
1474 	return (error);
1475 }
1476 
1477 /*
1478  * Copy filecaps structure allocating memory for ioctls array if needed.
1479  *
1480  * The last parameter indicates whether the fdtable is locked. If it is not and
1481  * ioctls are encountered, copying fails and the caller must lock the table.
1482  *
1483  * Note that if the table was not locked, the caller has to check the relevant
1484  * sequence counter to determine whether the operation was successful.
1485  */
1486 bool
1487 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1488 {
1489 	size_t size;
1490 
1491 	if (src->fc_ioctls != NULL && !locked)
1492 		return (false);
1493 	memcpy(dst, src, sizeof(*src));
1494 	if (src->fc_ioctls == NULL)
1495 		return (true);
1496 
1497 	KASSERT(src->fc_nioctls > 0,
1498 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1499 
1500 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1501 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1502 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1503 	return (true);
1504 }
1505 
1506 static u_long *
1507 filecaps_copy_prep(const struct filecaps *src)
1508 {
1509 	u_long *ioctls;
1510 	size_t size;
1511 
1512 	if (__predict_true(src->fc_ioctls == NULL))
1513 		return (NULL);
1514 
1515 	KASSERT(src->fc_nioctls > 0,
1516 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1517 
1518 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1519 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1520 	return (ioctls);
1521 }
1522 
1523 static void
1524 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1525     u_long *ioctls)
1526 {
1527 	size_t size;
1528 
1529 	*dst = *src;
1530 	if (__predict_true(src->fc_ioctls == NULL)) {
1531 		MPASS(ioctls == NULL);
1532 		return;
1533 	}
1534 
1535 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1536 	dst->fc_ioctls = ioctls;
1537 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1538 }
1539 
1540 /*
1541  * Move filecaps structure to the new place and clear the old place.
1542  */
1543 void
1544 filecaps_move(struct filecaps *src, struct filecaps *dst)
1545 {
1546 
1547 	*dst = *src;
1548 	bzero(src, sizeof(*src));
1549 }
1550 
1551 /*
1552  * Fill the given filecaps structure with full rights.
1553  */
1554 static void
1555 filecaps_fill(struct filecaps *fcaps)
1556 {
1557 
1558 	CAP_ALL(&fcaps->fc_rights);
1559 	fcaps->fc_ioctls = NULL;
1560 	fcaps->fc_nioctls = -1;
1561 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1562 }
1563 
1564 /*
1565  * Free memory allocated within filecaps structure.
1566  */
1567 void
1568 filecaps_free(struct filecaps *fcaps)
1569 {
1570 
1571 	free(fcaps->fc_ioctls, M_FILECAPS);
1572 	bzero(fcaps, sizeof(*fcaps));
1573 }
1574 
1575 static u_long *
1576 filecaps_free_prep(struct filecaps *fcaps)
1577 {
1578 	u_long *ioctls;
1579 
1580 	ioctls = fcaps->fc_ioctls;
1581 	bzero(fcaps, sizeof(*fcaps));
1582 	return (ioctls);
1583 }
1584 
1585 static void
1586 filecaps_free_finish(u_long *ioctls)
1587 {
1588 
1589 	free(ioctls, M_FILECAPS);
1590 }
1591 
1592 /*
1593  * Validate the given filecaps structure.
1594  */
1595 static void
1596 filecaps_validate(const struct filecaps *fcaps, const char *func)
1597 {
1598 
1599 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1600 	    ("%s: invalid rights", func));
1601 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1602 	    ("%s: invalid fcntls", func));
1603 	KASSERT(fcaps->fc_fcntls == 0 ||
1604 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1605 	    ("%s: fcntls without CAP_FCNTL", func));
1606 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1607 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1608 	    ("%s: invalid ioctls", func));
1609 	KASSERT(fcaps->fc_nioctls == 0 ||
1610 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1611 	    ("%s: ioctls without CAP_IOCTL", func));
1612 }
1613 
1614 static void
1615 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1616 {
1617 	int nfd1;
1618 
1619 	FILEDESC_XLOCK_ASSERT(fdp);
1620 
1621 	nfd1 = fdp->fd_nfiles * 2;
1622 	if (nfd1 < nfd)
1623 		nfd1 = nfd;
1624 	fdgrowtable(fdp, nfd1);
1625 }
1626 
1627 /*
1628  * Grow the file table to accommodate (at least) nfd descriptors.
1629  */
1630 static void
1631 fdgrowtable(struct filedesc *fdp, int nfd)
1632 {
1633 	struct filedesc0 *fdp0;
1634 	struct freetable *ft;
1635 	struct fdescenttbl *ntable;
1636 	struct fdescenttbl *otable;
1637 	int nnfiles, onfiles;
1638 	NDSLOTTYPE *nmap, *omap;
1639 
1640 	/*
1641 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1642 	 * growing it to accommodate for the one we are going to copy from. There
1643 	 * is no need to have a lock on this one as it's not visible to anyone.
1644 	 */
1645 	if (fdp->fd_lastfile != -1)
1646 		FILEDESC_XLOCK_ASSERT(fdp);
1647 
1648 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1649 
1650 	/* save old values */
1651 	onfiles = fdp->fd_nfiles;
1652 	otable = fdp->fd_files;
1653 	omap = fdp->fd_map;
1654 
1655 	/* compute the size of the new table */
1656 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1657 	if (nnfiles <= onfiles)
1658 		/* the table is already large enough */
1659 		return;
1660 
1661 	/*
1662 	 * Allocate a new table.  We need enough space for the number of
1663 	 * entries, file entries themselves and the struct freetable we will use
1664 	 * when we decommission the table and place it on the freelist.
1665 	 * We place the struct freetable in the middle so we don't have
1666 	 * to worry about padding.
1667 	 */
1668 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1669 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1670 	    sizeof(struct freetable),
1671 	    M_FILEDESC, M_ZERO | M_WAITOK);
1672 	/* copy the old data */
1673 	ntable->fdt_nfiles = nnfiles;
1674 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1675 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1676 
1677 	/*
1678 	 * Allocate a new map only if the old is not large enough.  It will
1679 	 * grow at a slower rate than the table as it can map more
1680 	 * entries than the table can hold.
1681 	 */
1682 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1683 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1684 		    M_ZERO | M_WAITOK);
1685 		/* copy over the old data and update the pointer */
1686 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1687 		fdp->fd_map = nmap;
1688 	}
1689 
1690 	/*
1691 	 * Make sure that ntable is correctly initialized before we replace
1692 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1693 	 * data.
1694 	 */
1695 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1696 
1697 	/*
1698 	 * Do not free the old file table, as some threads may still
1699 	 * reference entries within it.  Instead, place it on a freelist
1700 	 * which will be processed when the struct filedesc is released.
1701 	 *
1702 	 * Note that if onfiles == NDFILE, we're dealing with the original
1703 	 * static allocation contained within (struct filedesc0 *)fdp,
1704 	 * which must not be freed.
1705 	 */
1706 	if (onfiles > NDFILE) {
1707 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1708 		fdp0 = (struct filedesc0 *)fdp;
1709 		ft->ft_table = otable;
1710 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1711 	}
1712 	/*
1713 	 * The map does not have the same possibility of threads still
1714 	 * holding references to it.  So always free it as long as it
1715 	 * does not reference the original static allocation.
1716 	 */
1717 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1718 		free(omap, M_FILEDESC);
1719 }
1720 
1721 /*
1722  * Allocate a file descriptor for the process.
1723  */
1724 int
1725 fdalloc(struct thread *td, int minfd, int *result)
1726 {
1727 	struct proc *p = td->td_proc;
1728 	struct filedesc *fdp = p->p_fd;
1729 	int fd, maxfd, allocfd;
1730 #ifdef RACCT
1731 	int error;
1732 #endif
1733 
1734 	FILEDESC_XLOCK_ASSERT(fdp);
1735 
1736 	if (fdp->fd_freefile > minfd)
1737 		minfd = fdp->fd_freefile;
1738 
1739 	maxfd = getmaxfd(td);
1740 
1741 	/*
1742 	 * Search the bitmap for a free descriptor starting at minfd.
1743 	 * If none is found, grow the file table.
1744 	 */
1745 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1746 	if (fd >= maxfd)
1747 		return (EMFILE);
1748 	if (fd >= fdp->fd_nfiles) {
1749 		allocfd = min(fd * 2, maxfd);
1750 #ifdef RACCT
1751 		if (RACCT_ENABLED()) {
1752 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1753 			if (error != 0)
1754 				return (EMFILE);
1755 		}
1756 #endif
1757 		/*
1758 		 * fd is already equal to first free descriptor >= minfd, so
1759 		 * we only need to grow the table and we are done.
1760 		 */
1761 		fdgrowtable_exp(fdp, allocfd);
1762 	}
1763 
1764 	/*
1765 	 * Perform some sanity checks, then mark the file descriptor as
1766 	 * used and return it to the caller.
1767 	 */
1768 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1769 	    ("invalid descriptor %d", fd));
1770 	KASSERT(!fdisused(fdp, fd),
1771 	    ("fd_first_free() returned non-free descriptor"));
1772 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1773 	    ("file descriptor isn't free"));
1774 	fdused(fdp, fd);
1775 	*result = fd;
1776 	return (0);
1777 }
1778 
1779 /*
1780  * Allocate n file descriptors for the process.
1781  */
1782 int
1783 fdallocn(struct thread *td, int minfd, int *fds, int n)
1784 {
1785 	struct proc *p = td->td_proc;
1786 	struct filedesc *fdp = p->p_fd;
1787 	int i;
1788 
1789 	FILEDESC_XLOCK_ASSERT(fdp);
1790 
1791 	for (i = 0; i < n; i++)
1792 		if (fdalloc(td, 0, &fds[i]) != 0)
1793 			break;
1794 
1795 	if (i < n) {
1796 		for (i--; i >= 0; i--)
1797 			fdunused(fdp, fds[i]);
1798 		return (EMFILE);
1799 	}
1800 
1801 	return (0);
1802 }
1803 
1804 /*
1805  * Create a new open file structure and allocate a file descriptor for the
1806  * process that refers to it.  We add one reference to the file for the
1807  * descriptor table and one reference for resultfp. This is to prevent us
1808  * being preempted and the entry in the descriptor table closed after we
1809  * release the FILEDESC lock.
1810  */
1811 int
1812 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1813     struct filecaps *fcaps)
1814 {
1815 	struct file *fp;
1816 	int error, fd;
1817 
1818 	error = falloc_noinstall(td, &fp);
1819 	if (error)
1820 		return (error);		/* no reference held on error */
1821 
1822 	error = finstall(td, fp, &fd, flags, fcaps);
1823 	if (error) {
1824 		fdrop(fp, td);		/* one reference (fp only) */
1825 		return (error);
1826 	}
1827 
1828 	if (resultfp != NULL)
1829 		*resultfp = fp;		/* copy out result */
1830 	else
1831 		fdrop(fp, td);		/* release local reference */
1832 
1833 	if (resultfd != NULL)
1834 		*resultfd = fd;
1835 
1836 	return (0);
1837 }
1838 
1839 /*
1840  * Create a new open file structure without allocating a file descriptor.
1841  */
1842 int
1843 falloc_noinstall(struct thread *td, struct file **resultfp)
1844 {
1845 	struct file *fp;
1846 	int maxuserfiles = maxfiles - (maxfiles / 20);
1847 	int openfiles_new;
1848 	static struct timeval lastfail;
1849 	static int curfail;
1850 
1851 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1852 
1853 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1854 	if ((openfiles_new >= maxuserfiles &&
1855 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1856 	    openfiles_new >= maxfiles) {
1857 		atomic_subtract_int(&openfiles, 1);
1858 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1859 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1860 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1861 		}
1862 		return (ENFILE);
1863 	}
1864 	fp = uma_zalloc(file_zone, M_WAITOK);
1865 	bzero(fp, sizeof(*fp));
1866 	refcount_init(&fp->f_count, 1);
1867 	fp->f_cred = crhold(td->td_ucred);
1868 	fp->f_ops = &badfileops;
1869 	*resultfp = fp;
1870 	return (0);
1871 }
1872 
1873 /*
1874  * Install a file in a file descriptor table.
1875  */
1876 void
1877 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1878     struct filecaps *fcaps)
1879 {
1880 	struct filedescent *fde;
1881 
1882 	MPASS(fp != NULL);
1883 	if (fcaps != NULL)
1884 		filecaps_validate(fcaps, __func__);
1885 	FILEDESC_XLOCK_ASSERT(fdp);
1886 
1887 	fde = &fdp->fd_ofiles[fd];
1888 #ifdef CAPABILITIES
1889 	seqc_write_begin(&fde->fde_seqc);
1890 #endif
1891 	fde->fde_file = fp;
1892 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1893 	if (fcaps != NULL)
1894 		filecaps_move(fcaps, &fde->fde_caps);
1895 	else
1896 		filecaps_fill(&fde->fde_caps);
1897 #ifdef CAPABILITIES
1898 	seqc_write_end(&fde->fde_seqc);
1899 #endif
1900 }
1901 
1902 int
1903 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1904     struct filecaps *fcaps)
1905 {
1906 	struct filedesc *fdp = td->td_proc->p_fd;
1907 	int error;
1908 
1909 	MPASS(fd != NULL);
1910 
1911 	if (!fhold(fp))
1912 		return (EBADF);
1913 	FILEDESC_XLOCK(fdp);
1914 	if ((error = fdalloc(td, 0, fd))) {
1915 		FILEDESC_XUNLOCK(fdp);
1916 		fdrop(fp, td);
1917 		return (error);
1918 	}
1919 	_finstall(fdp, fp, *fd, flags, fcaps);
1920 	FILEDESC_XUNLOCK(fdp);
1921 	return (0);
1922 }
1923 
1924 /*
1925  * Build a new filedesc structure from another.
1926  * Copy the current, root, and jail root vnode references.
1927  *
1928  * If fdp is not NULL, return with it shared locked.
1929  */
1930 struct filedesc *
1931 fdinit(struct filedesc *fdp, bool prepfiles)
1932 {
1933 	struct filedesc0 *newfdp0;
1934 	struct filedesc *newfdp;
1935 
1936 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1937 	newfdp = &newfdp0->fd_fd;
1938 
1939 	/* Create the file descriptor table. */
1940 	FILEDESC_LOCK_INIT(newfdp);
1941 	refcount_init(&newfdp->fd_refcnt, 1);
1942 	refcount_init(&newfdp->fd_holdcnt, 1);
1943 	newfdp->fd_cmask = CMASK;
1944 	newfdp->fd_map = newfdp0->fd_dmap;
1945 	newfdp->fd_lastfile = -1;
1946 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1947 	newfdp->fd_files->fdt_nfiles = NDFILE;
1948 
1949 	if (fdp == NULL)
1950 		return (newfdp);
1951 
1952 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1953 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1954 
1955 	FILEDESC_SLOCK(fdp);
1956 	newfdp->fd_cdir = fdp->fd_cdir;
1957 	if (newfdp->fd_cdir)
1958 		vrefact(newfdp->fd_cdir);
1959 	newfdp->fd_rdir = fdp->fd_rdir;
1960 	if (newfdp->fd_rdir)
1961 		vrefact(newfdp->fd_rdir);
1962 	newfdp->fd_jdir = fdp->fd_jdir;
1963 	if (newfdp->fd_jdir)
1964 		vrefact(newfdp->fd_jdir);
1965 
1966 	if (!prepfiles) {
1967 		FILEDESC_SUNLOCK(fdp);
1968 	} else {
1969 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1970 			FILEDESC_SUNLOCK(fdp);
1971 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1972 			FILEDESC_SLOCK(fdp);
1973 		}
1974 	}
1975 
1976 	return (newfdp);
1977 }
1978 
1979 static struct filedesc *
1980 fdhold(struct proc *p)
1981 {
1982 	struct filedesc *fdp;
1983 
1984 	PROC_LOCK_ASSERT(p, MA_OWNED);
1985 	fdp = p->p_fd;
1986 	if (fdp != NULL)
1987 		refcount_acquire(&fdp->fd_holdcnt);
1988 	return (fdp);
1989 }
1990 
1991 static void
1992 fddrop(struct filedesc *fdp)
1993 {
1994 
1995 	if (fdp->fd_holdcnt > 1) {
1996 		if (refcount_release(&fdp->fd_holdcnt) == 0)
1997 			return;
1998 	}
1999 
2000 	FILEDESC_LOCK_DESTROY(fdp);
2001 	uma_zfree(filedesc0_zone, fdp);
2002 }
2003 
2004 /*
2005  * Share a filedesc structure.
2006  */
2007 struct filedesc *
2008 fdshare(struct filedesc *fdp)
2009 {
2010 
2011 	refcount_acquire(&fdp->fd_refcnt);
2012 	return (fdp);
2013 }
2014 
2015 /*
2016  * Unshare a filedesc structure, if necessary by making a copy
2017  */
2018 void
2019 fdunshare(struct thread *td)
2020 {
2021 	struct filedesc *tmp;
2022 	struct proc *p = td->td_proc;
2023 
2024 	if (p->p_fd->fd_refcnt == 1)
2025 		return;
2026 
2027 	tmp = fdcopy(p->p_fd);
2028 	fdescfree(td);
2029 	p->p_fd = tmp;
2030 }
2031 
2032 void
2033 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2034 {
2035 
2036 	fdescfree(td);
2037 	td->td_proc->p_fd = fdp;
2038 }
2039 
2040 /*
2041  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2042  * this is to ease callers, not catch errors.
2043  */
2044 struct filedesc *
2045 fdcopy(struct filedesc *fdp)
2046 {
2047 	struct filedesc *newfdp;
2048 	struct filedescent *nfde, *ofde;
2049 	int i;
2050 
2051 	MPASS(fdp != NULL);
2052 
2053 	newfdp = fdinit(fdp, true);
2054 	/* copy all passable descriptors (i.e. not kqueue) */
2055 	newfdp->fd_freefile = -1;
2056 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
2057 		ofde = &fdp->fd_ofiles[i];
2058 		if (ofde->fde_file == NULL ||
2059 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2060 		    !fhold(ofde->fde_file)) {
2061 			if (newfdp->fd_freefile == -1)
2062 				newfdp->fd_freefile = i;
2063 			continue;
2064 		}
2065 		nfde = &newfdp->fd_ofiles[i];
2066 		*nfde = *ofde;
2067 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2068 		fdused_init(newfdp, i);
2069 		newfdp->fd_lastfile = i;
2070 	}
2071 	if (newfdp->fd_freefile == -1)
2072 		newfdp->fd_freefile = i;
2073 	newfdp->fd_cmask = fdp->fd_cmask;
2074 	FILEDESC_SUNLOCK(fdp);
2075 	return (newfdp);
2076 }
2077 
2078 /*
2079  * Copies a filedesc structure, while remapping all file descriptors
2080  * stored inside using a translation table.
2081  *
2082  * File descriptors are copied over to the new file descriptor table,
2083  * regardless of whether the close-on-exec flag is set.
2084  */
2085 int
2086 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2087     struct filedesc **ret)
2088 {
2089 	struct filedesc *newfdp;
2090 	struct filedescent *nfde, *ofde;
2091 	int error, i;
2092 
2093 	MPASS(fdp != NULL);
2094 
2095 	newfdp = fdinit(fdp, true);
2096 	if (nfds > fdp->fd_lastfile + 1) {
2097 		/* New table cannot be larger than the old one. */
2098 		error = E2BIG;
2099 		goto bad;
2100 	}
2101 	/* Copy all passable descriptors (i.e. not kqueue). */
2102 	newfdp->fd_freefile = nfds;
2103 	for (i = 0; i < nfds; ++i) {
2104 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2105 			/* File descriptor out of bounds. */
2106 			error = EBADF;
2107 			goto bad;
2108 		}
2109 		ofde = &fdp->fd_ofiles[fds[i]];
2110 		if (ofde->fde_file == NULL) {
2111 			/* Unused file descriptor. */
2112 			error = EBADF;
2113 			goto bad;
2114 		}
2115 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2116 			/* File descriptor cannot be passed. */
2117 			error = EINVAL;
2118 			goto bad;
2119 		}
2120 		if (!fhold(nfde->fde_file)) {
2121 			error = EBADF;
2122 			goto bad;
2123 		}
2124 		nfde = &newfdp->fd_ofiles[i];
2125 		*nfde = *ofde;
2126 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2127 		fdused_init(newfdp, i);
2128 		newfdp->fd_lastfile = i;
2129 	}
2130 	newfdp->fd_cmask = fdp->fd_cmask;
2131 	FILEDESC_SUNLOCK(fdp);
2132 	*ret = newfdp;
2133 	return (0);
2134 bad:
2135 	FILEDESC_SUNLOCK(fdp);
2136 	fdescfree_remapped(newfdp);
2137 	return (error);
2138 }
2139 
2140 /*
2141  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2142  * one of processes using it exits) and the table used to be shared.
2143  */
2144 static void
2145 fdclearlocks(struct thread *td)
2146 {
2147 	struct filedesc *fdp;
2148 	struct filedesc_to_leader *fdtol;
2149 	struct flock lf;
2150 	struct file *fp;
2151 	struct proc *p;
2152 	struct vnode *vp;
2153 	int i;
2154 
2155 	p = td->td_proc;
2156 	fdp = p->p_fd;
2157 	fdtol = p->p_fdtol;
2158 	MPASS(fdtol != NULL);
2159 
2160 	FILEDESC_XLOCK(fdp);
2161 	KASSERT(fdtol->fdl_refcount > 0,
2162 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2163 	    fdtol->fdl_refcount));
2164 	if (fdtol->fdl_refcount == 1 &&
2165 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2166 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2167 			fp = fdp->fd_ofiles[i].fde_file;
2168 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2169 			    !fhold(fp))
2170 				continue;
2171 			FILEDESC_XUNLOCK(fdp);
2172 			lf.l_whence = SEEK_SET;
2173 			lf.l_start = 0;
2174 			lf.l_len = 0;
2175 			lf.l_type = F_UNLCK;
2176 			vp = fp->f_vnode;
2177 			(void) VOP_ADVLOCK(vp,
2178 			    (caddr_t)p->p_leader, F_UNLCK,
2179 			    &lf, F_POSIX);
2180 			FILEDESC_XLOCK(fdp);
2181 			fdrop(fp, td);
2182 		}
2183 	}
2184 retry:
2185 	if (fdtol->fdl_refcount == 1) {
2186 		if (fdp->fd_holdleaderscount > 0 &&
2187 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2188 			/*
2189 			 * close() or kern_dup() has cleared a reference
2190 			 * in a shared file descriptor table.
2191 			 */
2192 			fdp->fd_holdleaderswakeup = 1;
2193 			sx_sleep(&fdp->fd_holdleaderscount,
2194 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2195 			goto retry;
2196 		}
2197 		if (fdtol->fdl_holdcount > 0) {
2198 			/*
2199 			 * Ensure that fdtol->fdl_leader remains
2200 			 * valid in closef().
2201 			 */
2202 			fdtol->fdl_wakeup = 1;
2203 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2204 			    "fdlhold", 0);
2205 			goto retry;
2206 		}
2207 	}
2208 	fdtol->fdl_refcount--;
2209 	if (fdtol->fdl_refcount == 0 &&
2210 	    fdtol->fdl_holdcount == 0) {
2211 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2212 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2213 	} else
2214 		fdtol = NULL;
2215 	p->p_fdtol = NULL;
2216 	FILEDESC_XUNLOCK(fdp);
2217 	if (fdtol != NULL)
2218 		free(fdtol, M_FILEDESC_TO_LEADER);
2219 }
2220 
2221 /*
2222  * Release a filedesc structure.
2223  */
2224 static void
2225 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2226 {
2227 	struct filedesc0 *fdp0;
2228 	struct freetable *ft, *tft;
2229 	struct filedescent *fde;
2230 	struct file *fp;
2231 	int i;
2232 
2233 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2234 		fde = &fdp->fd_ofiles[i];
2235 		fp = fde->fde_file;
2236 		if (fp != NULL) {
2237 			fdefree_last(fde);
2238 			if (needclose)
2239 				(void) closef(fp, td);
2240 			else
2241 				fdrop(fp, td);
2242 		}
2243 	}
2244 
2245 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2246 		free(fdp->fd_map, M_FILEDESC);
2247 	if (fdp->fd_nfiles > NDFILE)
2248 		free(fdp->fd_files, M_FILEDESC);
2249 
2250 	fdp0 = (struct filedesc0 *)fdp;
2251 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2252 		free(ft->ft_table, M_FILEDESC);
2253 
2254 	fddrop(fdp);
2255 }
2256 
2257 void
2258 fdescfree(struct thread *td)
2259 {
2260 	struct proc *p;
2261 	struct filedesc *fdp;
2262 	struct vnode *cdir, *jdir, *rdir;
2263 
2264 	p = td->td_proc;
2265 	fdp = p->p_fd;
2266 	MPASS(fdp != NULL);
2267 
2268 #ifdef RACCT
2269 	if (RACCT_ENABLED())
2270 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2271 #endif
2272 
2273 	if (p->p_fdtol != NULL)
2274 		fdclearlocks(td);
2275 
2276 	PROC_LOCK(p);
2277 	p->p_fd = NULL;
2278 	PROC_UNLOCK(p);
2279 
2280 	if (refcount_release(&fdp->fd_refcnt) == 0)
2281 		return;
2282 
2283 	FILEDESC_XLOCK(fdp);
2284 	cdir = fdp->fd_cdir;
2285 	fdp->fd_cdir = NULL;
2286 	rdir = fdp->fd_rdir;
2287 	fdp->fd_rdir = NULL;
2288 	jdir = fdp->fd_jdir;
2289 	fdp->fd_jdir = NULL;
2290 	FILEDESC_XUNLOCK(fdp);
2291 
2292 	if (cdir != NULL)
2293 		vrele(cdir);
2294 	if (rdir != NULL)
2295 		vrele(rdir);
2296 	if (jdir != NULL)
2297 		vrele(jdir);
2298 
2299 	fdescfree_fds(td, fdp, 1);
2300 }
2301 
2302 void
2303 fdescfree_remapped(struct filedesc *fdp)
2304 {
2305 
2306 	if (fdp->fd_cdir != NULL)
2307 		vrele(fdp->fd_cdir);
2308 	if (fdp->fd_rdir != NULL)
2309 		vrele(fdp->fd_rdir);
2310 	if (fdp->fd_jdir != NULL)
2311 		vrele(fdp->fd_jdir);
2312 
2313 	fdescfree_fds(curthread, fdp, 0);
2314 }
2315 
2316 /*
2317  * For setugid programs, we don't want to people to use that setugidness
2318  * to generate error messages which write to a file which otherwise would
2319  * otherwise be off-limits to the process.  We check for filesystems where
2320  * the vnode can change out from under us after execve (like [lin]procfs).
2321  *
2322  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2323  * sufficient.  We also don't check for setugidness since we know we are.
2324  */
2325 static bool
2326 is_unsafe(struct file *fp)
2327 {
2328 	struct vnode *vp;
2329 
2330 	if (fp->f_type != DTYPE_VNODE)
2331 		return (false);
2332 
2333 	vp = fp->f_vnode;
2334 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2335 }
2336 
2337 /*
2338  * Make this setguid thing safe, if at all possible.
2339  */
2340 void
2341 fdsetugidsafety(struct thread *td)
2342 {
2343 	struct filedesc *fdp;
2344 	struct file *fp;
2345 	int i;
2346 
2347 	fdp = td->td_proc->p_fd;
2348 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2349 	MPASS(fdp->fd_nfiles >= 3);
2350 	for (i = 0; i <= 2; i++) {
2351 		fp = fdp->fd_ofiles[i].fde_file;
2352 		if (fp != NULL && is_unsafe(fp)) {
2353 			FILEDESC_XLOCK(fdp);
2354 			knote_fdclose(td, i);
2355 			/*
2356 			 * NULL-out descriptor prior to close to avoid
2357 			 * a race while close blocks.
2358 			 */
2359 			fdfree(fdp, i);
2360 			FILEDESC_XUNLOCK(fdp);
2361 			(void) closef(fp, td);
2362 		}
2363 	}
2364 }
2365 
2366 /*
2367  * If a specific file object occupies a specific file descriptor, close the
2368  * file descriptor entry and drop a reference on the file object.  This is a
2369  * convenience function to handle a subsequent error in a function that calls
2370  * falloc() that handles the race that another thread might have closed the
2371  * file descriptor out from under the thread creating the file object.
2372  */
2373 void
2374 fdclose(struct thread *td, struct file *fp, int idx)
2375 {
2376 	struct filedesc *fdp = td->td_proc->p_fd;
2377 
2378 	FILEDESC_XLOCK(fdp);
2379 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2380 		fdfree(fdp, idx);
2381 		FILEDESC_XUNLOCK(fdp);
2382 		fdrop(fp, td);
2383 	} else
2384 		FILEDESC_XUNLOCK(fdp);
2385 }
2386 
2387 /*
2388  * Close any files on exec?
2389  */
2390 void
2391 fdcloseexec(struct thread *td)
2392 {
2393 	struct filedesc *fdp;
2394 	struct filedescent *fde;
2395 	struct file *fp;
2396 	int i;
2397 
2398 	fdp = td->td_proc->p_fd;
2399 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2400 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2401 		fde = &fdp->fd_ofiles[i];
2402 		fp = fde->fde_file;
2403 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2404 		    (fde->fde_flags & UF_EXCLOSE))) {
2405 			FILEDESC_XLOCK(fdp);
2406 			fdfree(fdp, i);
2407 			(void) closefp(fdp, i, fp, td, 0);
2408 			FILEDESC_UNLOCK_ASSERT(fdp);
2409 		}
2410 	}
2411 }
2412 
2413 /*
2414  * It is unsafe for set[ug]id processes to be started with file
2415  * descriptors 0..2 closed, as these descriptors are given implicit
2416  * significance in the Standard C library.  fdcheckstd() will create a
2417  * descriptor referencing /dev/null for each of stdin, stdout, and
2418  * stderr that is not already open.
2419  */
2420 int
2421 fdcheckstd(struct thread *td)
2422 {
2423 	struct filedesc *fdp;
2424 	register_t save;
2425 	int i, error, devnull;
2426 
2427 	fdp = td->td_proc->p_fd;
2428 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2429 	MPASS(fdp->fd_nfiles >= 3);
2430 	devnull = -1;
2431 	for (i = 0; i <= 2; i++) {
2432 		if (fdp->fd_ofiles[i].fde_file != NULL)
2433 			continue;
2434 
2435 		save = td->td_retval[0];
2436 		if (devnull != -1) {
2437 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2438 		} else {
2439 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2440 			    UIO_SYSSPACE, O_RDWR, 0);
2441 			if (error == 0) {
2442 				devnull = td->td_retval[0];
2443 				KASSERT(devnull == i, ("we didn't get our fd"));
2444 			}
2445 		}
2446 		td->td_retval[0] = save;
2447 		if (error != 0)
2448 			return (error);
2449 	}
2450 	return (0);
2451 }
2452 
2453 /*
2454  * Internal form of close.  Decrement reference count on file structure.
2455  * Note: td may be NULL when closing a file that was being passed in a
2456  * message.
2457  */
2458 int
2459 closef(struct file *fp, struct thread *td)
2460 {
2461 	struct vnode *vp;
2462 	struct flock lf;
2463 	struct filedesc_to_leader *fdtol;
2464 	struct filedesc *fdp;
2465 
2466 	/*
2467 	 * POSIX record locking dictates that any close releases ALL
2468 	 * locks owned by this process.  This is handled by setting
2469 	 * a flag in the unlock to free ONLY locks obeying POSIX
2470 	 * semantics, and not to free BSD-style file locks.
2471 	 * If the descriptor was in a message, POSIX-style locks
2472 	 * aren't passed with the descriptor, and the thread pointer
2473 	 * will be NULL.  Callers should be careful only to pass a
2474 	 * NULL thread pointer when there really is no owning
2475 	 * context that might have locks, or the locks will be
2476 	 * leaked.
2477 	 */
2478 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2479 		vp = fp->f_vnode;
2480 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2481 			lf.l_whence = SEEK_SET;
2482 			lf.l_start = 0;
2483 			lf.l_len = 0;
2484 			lf.l_type = F_UNLCK;
2485 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2486 			    F_UNLCK, &lf, F_POSIX);
2487 		}
2488 		fdtol = td->td_proc->p_fdtol;
2489 		if (fdtol != NULL) {
2490 			/*
2491 			 * Handle special case where file descriptor table is
2492 			 * shared between multiple process leaders.
2493 			 */
2494 			fdp = td->td_proc->p_fd;
2495 			FILEDESC_XLOCK(fdp);
2496 			for (fdtol = fdtol->fdl_next;
2497 			    fdtol != td->td_proc->p_fdtol;
2498 			    fdtol = fdtol->fdl_next) {
2499 				if ((fdtol->fdl_leader->p_flag &
2500 				    P_ADVLOCK) == 0)
2501 					continue;
2502 				fdtol->fdl_holdcount++;
2503 				FILEDESC_XUNLOCK(fdp);
2504 				lf.l_whence = SEEK_SET;
2505 				lf.l_start = 0;
2506 				lf.l_len = 0;
2507 				lf.l_type = F_UNLCK;
2508 				vp = fp->f_vnode;
2509 				(void) VOP_ADVLOCK(vp,
2510 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2511 				    F_POSIX);
2512 				FILEDESC_XLOCK(fdp);
2513 				fdtol->fdl_holdcount--;
2514 				if (fdtol->fdl_holdcount == 0 &&
2515 				    fdtol->fdl_wakeup != 0) {
2516 					fdtol->fdl_wakeup = 0;
2517 					wakeup(fdtol);
2518 				}
2519 			}
2520 			FILEDESC_XUNLOCK(fdp);
2521 		}
2522 	}
2523 	return (fdrop(fp, td));
2524 }
2525 
2526 /*
2527  * Initialize the file pointer with the specified properties.
2528  *
2529  * The ops are set with release semantics to be certain that the flags, type,
2530  * and data are visible when ops is.  This is to prevent ops methods from being
2531  * called with bad data.
2532  */
2533 void
2534 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2535 {
2536 	fp->f_data = data;
2537 	fp->f_flag = flag;
2538 	fp->f_type = type;
2539 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2540 }
2541 
2542 int
2543 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2544     struct file **fpp, struct filecaps *havecapsp)
2545 {
2546 	struct filedescent *fde;
2547 	int error;
2548 
2549 	FILEDESC_LOCK_ASSERT(fdp);
2550 
2551 	fde = fdeget_locked(fdp, fd);
2552 	if (fde == NULL) {
2553 		error = EBADF;
2554 		goto out;
2555 	}
2556 
2557 #ifdef CAPABILITIES
2558 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2559 	if (error != 0)
2560 		goto out;
2561 #endif
2562 
2563 	if (havecapsp != NULL)
2564 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2565 
2566 	*fpp = fde->fde_file;
2567 
2568 	error = 0;
2569 out:
2570 	return (error);
2571 }
2572 
2573 int
2574 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2575     struct file **fpp, struct filecaps *havecapsp)
2576 {
2577 	struct filedesc *fdp = td->td_proc->p_fd;
2578 	int error;
2579 #ifndef CAPABILITIES
2580 	error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2581 	if (error == 0 && havecapsp != NULL)
2582 		filecaps_fill(havecapsp);
2583 #else
2584 	struct file *fp;
2585 	seqc_t seq;
2586 
2587 	for (;;) {
2588 		error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2589 		if (error != 0)
2590 			return (error);
2591 
2592 		if (havecapsp != NULL) {
2593 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2594 			    havecapsp, false)) {
2595 				fdrop(fp, td);
2596 				goto get_locked;
2597 			}
2598 		}
2599 
2600 		if (!fd_modified(fdp, fd, seq))
2601 			break;
2602 		fdrop(fp, td);
2603 	}
2604 
2605 	*fpp = fp;
2606 	return (0);
2607 
2608 get_locked:
2609 	FILEDESC_SLOCK(fdp);
2610 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2611 	if (error == 0 && !fhold(*fpp))
2612 		error = EBADF;
2613 	FILEDESC_SUNLOCK(fdp);
2614 #endif
2615 	return (error);
2616 }
2617 
2618 int
2619 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2620     struct file **fpp, seqc_t *seqp)
2621 {
2622 #ifdef CAPABILITIES
2623 	const struct filedescent *fde;
2624 #endif
2625 	const struct fdescenttbl *fdt;
2626 	struct file *fp;
2627 	u_int count;
2628 #ifdef CAPABILITIES
2629 	seqc_t seq;
2630 	cap_rights_t haverights;
2631 	int error;
2632 #endif
2633 
2634 	fdt = fdp->fd_files;
2635 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2636 		return (EBADF);
2637 	/*
2638 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2639 	 * never raising a refcount above 0.  To accomplish this we have
2640 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2641 	 * must be re-verified once we acquire a reference to be certain
2642 	 * that the identity is still correct and we did not lose a race
2643 	 * due to preemption.
2644 	 */
2645 	for (;;) {
2646 #ifdef CAPABILITIES
2647 		seq = seqc_read(fd_seqc(fdt, fd));
2648 		fde = &fdt->fdt_ofiles[fd];
2649 		haverights = *cap_rights_fde_inline(fde);
2650 		fp = fde->fde_file;
2651 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2652 			continue;
2653 #else
2654 		fp = fdt->fdt_ofiles[fd].fde_file;
2655 #endif
2656 		if (fp == NULL)
2657 			return (EBADF);
2658 #ifdef CAPABILITIES
2659 		error = cap_check(&haverights, needrightsp);
2660 		if (error != 0)
2661 			return (error);
2662 #endif
2663 		count = fp->f_count;
2664 	retry:
2665 		if (count == 0) {
2666 			/*
2667 			 * Force a reload. Other thread could reallocate the
2668 			 * table before this fd was closed, so it possible that
2669 			 * there is a stale fp pointer in cached version.
2670 			 */
2671 			fdt = *(const struct fdescenttbl * const volatile *)
2672 			    &(fdp->fd_files);
2673 			continue;
2674 		}
2675 		if (__predict_false(count + 1 < count))
2676 			return (EBADF);
2677 
2678 		/*
2679 		 * Use an acquire barrier to force re-reading of fdt so it is
2680 		 * refreshed for verification.
2681 		 */
2682 		if (__predict_false(atomic_fcmpset_acq_int(&fp->f_count,
2683 		    &count, count + 1) == 0))
2684 			goto retry;
2685 		fdt = fdp->fd_files;
2686 #ifdef	CAPABILITIES
2687 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2688 #else
2689 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2690 #endif
2691 			break;
2692 		fdrop(fp, curthread);
2693 	}
2694 	*fpp = fp;
2695 	if (seqp != NULL) {
2696 #ifdef CAPABILITIES
2697 		*seqp = seq;
2698 #endif
2699 	}
2700 	return (0);
2701 }
2702 
2703 /*
2704  * Extract the file pointer associated with the specified descriptor for the
2705  * current user process.
2706  *
2707  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2708  * returned.
2709  *
2710  * File's rights will be checked against the capability rights mask.
2711  *
2712  * If an error occurred the non-zero error is returned and *fpp is set to
2713  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2714  * responsible for fdrop().
2715  */
2716 static __inline int
2717 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2718     cap_rights_t *needrightsp, seqc_t *seqp)
2719 {
2720 	struct filedesc *fdp;
2721 	struct file *fp;
2722 	int error;
2723 
2724 	*fpp = NULL;
2725 	fdp = td->td_proc->p_fd;
2726 	error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2727 	if (error != 0)
2728 		return (error);
2729 	if (fp->f_ops == &badfileops) {
2730 		fdrop(fp, td);
2731 		return (EBADF);
2732 	}
2733 
2734 	/*
2735 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2736 	 */
2737 	error = 0;
2738 	switch (flags) {
2739 	case FREAD:
2740 	case FWRITE:
2741 		if ((fp->f_flag & flags) == 0)
2742 			error = EBADF;
2743 		break;
2744 	case FEXEC:
2745 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2746 		    ((fp->f_flag & FWRITE) != 0))
2747 			error = EBADF;
2748 		break;
2749 	case 0:
2750 		break;
2751 	default:
2752 		KASSERT(0, ("wrong flags"));
2753 	}
2754 
2755 	if (error != 0) {
2756 		fdrop(fp, td);
2757 		return (error);
2758 	}
2759 
2760 	*fpp = fp;
2761 	return (0);
2762 }
2763 
2764 int
2765 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2766 {
2767 
2768 	return (_fget(td, fd, fpp, 0, rightsp, NULL));
2769 }
2770 
2771 int
2772 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2773     struct file **fpp)
2774 {
2775 	int error;
2776 #ifndef CAPABILITIES
2777 	error = _fget(td, fd, fpp, 0, rightsp, NULL);
2778 	if (maxprotp != NULL)
2779 		*maxprotp = VM_PROT_ALL;
2780 #else
2781 	cap_rights_t fdrights;
2782 	struct filedesc *fdp = td->td_proc->p_fd;
2783 	seqc_t seq;
2784 
2785 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2786 	for (;;) {
2787 		error = _fget(td, fd, fpp, 0, rightsp, &seq);
2788 		if (error != 0)
2789 			return (error);
2790 		if (maxprotp != NULL)
2791 			fdrights = *cap_rights(fdp, fd);
2792 		if (!fd_modified(fdp, fd, seq))
2793 			break;
2794 		fdrop(*fpp, td);
2795 	}
2796 
2797 	/*
2798 	 * If requested, convert capability rights to access flags.
2799 	 */
2800 	if (maxprotp != NULL)
2801 		*maxprotp = cap_rights_to_vmprot(&fdrights);
2802 #endif
2803 	return (error);
2804 }
2805 
2806 int
2807 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2808 {
2809 
2810 	return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2811 }
2812 
2813 int
2814 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2815 {
2816 
2817 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2818 }
2819 
2820 int
2821 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2822     struct file **fpp)
2823 {
2824 	struct filedesc *fdp = td->td_proc->p_fd;
2825 #ifndef CAPABILITIES
2826 	return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2827 #else
2828 	int error;
2829 	seqc_t seq;
2830 
2831 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2832 	for (;;) {
2833 		error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2834 		if (error != 0)
2835 			return (error);
2836 		error = cap_fcntl_check(fdp, fd, needfcntl);
2837 		if (!fd_modified(fdp, fd, seq))
2838 			break;
2839 		fdrop(*fpp, td);
2840 	}
2841 	if (error != 0) {
2842 		fdrop(*fpp, td);
2843 		*fpp = NULL;
2844 	}
2845 	return (error);
2846 #endif
2847 }
2848 
2849 /*
2850  * Like fget() but loads the underlying vnode, or returns an error if the
2851  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2852  * never have VM objects.  The returned vnode will be vref()'d.
2853  *
2854  * XXX: what about the unused flags ?
2855  */
2856 static __inline int
2857 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2858     struct vnode **vpp)
2859 {
2860 	struct file *fp;
2861 	int error;
2862 
2863 	*vpp = NULL;
2864 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2865 	if (error != 0)
2866 		return (error);
2867 	if (fp->f_vnode == NULL) {
2868 		error = EINVAL;
2869 	} else {
2870 		*vpp = fp->f_vnode;
2871 		vrefact(*vpp);
2872 	}
2873 	fdrop(fp, td);
2874 
2875 	return (error);
2876 }
2877 
2878 int
2879 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2880 {
2881 
2882 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2883 }
2884 
2885 int
2886 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2887     struct filecaps *havecaps, struct vnode **vpp)
2888 {
2889 	struct filedesc *fdp;
2890 	struct filecaps caps;
2891 	struct file *fp;
2892 	int error;
2893 
2894 	fdp = td->td_proc->p_fd;
2895 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2896 	if (error != 0)
2897 		return (error);
2898 	if (fp->f_ops == &badfileops) {
2899 		error = EBADF;
2900 		goto out;
2901 	}
2902 	if (fp->f_vnode == NULL) {
2903 		error = EINVAL;
2904 		goto out;
2905 	}
2906 
2907 	*havecaps = caps;
2908 	*vpp = fp->f_vnode;
2909 	vrefact(*vpp);
2910 
2911 	return (0);
2912 out:
2913 	filecaps_free(&caps);
2914 	return (error);
2915 }
2916 
2917 int
2918 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2919 {
2920 
2921 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2922 }
2923 
2924 int
2925 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2926 {
2927 
2928 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2929 }
2930 
2931 #ifdef notyet
2932 int
2933 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2934     struct vnode **vpp)
2935 {
2936 
2937 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2938 }
2939 #endif
2940 
2941 /*
2942  * Handle the last reference to a file being closed.
2943  *
2944  * Without the noinline attribute clang keeps inlining the func thorough this
2945  * file when fdrop is used.
2946  */
2947 int __noinline
2948 _fdrop(struct file *fp, struct thread *td)
2949 {
2950 	int error;
2951 
2952 	if (fp->f_count != 0)
2953 		panic("fdrop: count %d", fp->f_count);
2954 	error = fo_close(fp, td);
2955 	atomic_subtract_int(&openfiles, 1);
2956 	crfree(fp->f_cred);
2957 	free(fp->f_advice, M_FADVISE);
2958 	uma_zfree(file_zone, fp);
2959 
2960 	return (error);
2961 }
2962 
2963 /*
2964  * Apply an advisory lock on a file descriptor.
2965  *
2966  * Just attempt to get a record lock of the requested type on the entire file
2967  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2968  */
2969 #ifndef _SYS_SYSPROTO_H_
2970 struct flock_args {
2971 	int	fd;
2972 	int	how;
2973 };
2974 #endif
2975 /* ARGSUSED */
2976 int
2977 sys_flock(struct thread *td, struct flock_args *uap)
2978 {
2979 	struct file *fp;
2980 	struct vnode *vp;
2981 	struct flock lf;
2982 	int error;
2983 
2984 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
2985 	if (error != 0)
2986 		return (error);
2987 	if (fp->f_type != DTYPE_VNODE) {
2988 		fdrop(fp, td);
2989 		return (EOPNOTSUPP);
2990 	}
2991 
2992 	vp = fp->f_vnode;
2993 	lf.l_whence = SEEK_SET;
2994 	lf.l_start = 0;
2995 	lf.l_len = 0;
2996 	if (uap->how & LOCK_UN) {
2997 		lf.l_type = F_UNLCK;
2998 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2999 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3000 		goto done2;
3001 	}
3002 	if (uap->how & LOCK_EX)
3003 		lf.l_type = F_WRLCK;
3004 	else if (uap->how & LOCK_SH)
3005 		lf.l_type = F_RDLCK;
3006 	else {
3007 		error = EBADF;
3008 		goto done2;
3009 	}
3010 	atomic_set_int(&fp->f_flag, FHASLOCK);
3011 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3012 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3013 done2:
3014 	fdrop(fp, td);
3015 	return (error);
3016 }
3017 /*
3018  * Duplicate the specified descriptor to a free descriptor.
3019  */
3020 int
3021 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3022     int openerror, int *indxp)
3023 {
3024 	struct filedescent *newfde, *oldfde;
3025 	struct file *fp;
3026 	u_long *ioctls;
3027 	int error, indx;
3028 
3029 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3030 	    ("unexpected error %d in %s", openerror, __func__));
3031 
3032 	/*
3033 	 * If the to-be-dup'd fd number is greater than the allowed number
3034 	 * of file descriptors, or the fd to be dup'd has already been
3035 	 * closed, then reject.
3036 	 */
3037 	FILEDESC_XLOCK(fdp);
3038 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3039 		FILEDESC_XUNLOCK(fdp);
3040 		return (EBADF);
3041 	}
3042 
3043 	error = fdalloc(td, 0, &indx);
3044 	if (error != 0) {
3045 		FILEDESC_XUNLOCK(fdp);
3046 		return (error);
3047 	}
3048 
3049 	/*
3050 	 * There are two cases of interest here.
3051 	 *
3052 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3053 	 *
3054 	 * For ENXIO steal away the file structure from (dfd) and store it in
3055 	 * (indx).  (dfd) is effectively closed by this operation.
3056 	 */
3057 	switch (openerror) {
3058 	case ENODEV:
3059 		/*
3060 		 * Check that the mode the file is being opened for is a
3061 		 * subset of the mode of the existing descriptor.
3062 		 */
3063 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3064 			fdunused(fdp, indx);
3065 			FILEDESC_XUNLOCK(fdp);
3066 			return (EACCES);
3067 		}
3068 		if (!fhold(fp)) {
3069 			fdunused(fdp, indx);
3070 			FILEDESC_XUNLOCK(fdp);
3071 			return (EBADF);
3072 		}
3073 		newfde = &fdp->fd_ofiles[indx];
3074 		oldfde = &fdp->fd_ofiles[dfd];
3075 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3076 #ifdef CAPABILITIES
3077 		seqc_write_begin(&newfde->fde_seqc);
3078 #endif
3079 		memcpy(newfde, oldfde, fde_change_size);
3080 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3081 		    ioctls);
3082 #ifdef CAPABILITIES
3083 		seqc_write_end(&newfde->fde_seqc);
3084 #endif
3085 		break;
3086 	case ENXIO:
3087 		/*
3088 		 * Steal away the file pointer from dfd and stuff it into indx.
3089 		 */
3090 		newfde = &fdp->fd_ofiles[indx];
3091 		oldfde = &fdp->fd_ofiles[dfd];
3092 #ifdef CAPABILITIES
3093 		seqc_write_begin(&newfde->fde_seqc);
3094 #endif
3095 		memcpy(newfde, oldfde, fde_change_size);
3096 		oldfde->fde_file = NULL;
3097 		fdunused(fdp, dfd);
3098 #ifdef CAPABILITIES
3099 		seqc_write_end(&newfde->fde_seqc);
3100 #endif
3101 		break;
3102 	}
3103 	FILEDESC_XUNLOCK(fdp);
3104 	*indxp = indx;
3105 	return (0);
3106 }
3107 
3108 /*
3109  * This sysctl determines if we will allow a process to chroot(2) if it
3110  * has a directory open:
3111  *	0: disallowed for all processes.
3112  *	1: allowed for processes that were not already chroot(2)'ed.
3113  *	2: allowed for all processes.
3114  */
3115 
3116 static int chroot_allow_open_directories = 1;
3117 
3118 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3119     &chroot_allow_open_directories, 0,
3120     "Allow a process to chroot(2) if it has a directory open");
3121 
3122 /*
3123  * Helper function for raised chroot(2) security function:  Refuse if
3124  * any filedescriptors are open directories.
3125  */
3126 static int
3127 chroot_refuse_vdir_fds(struct filedesc *fdp)
3128 {
3129 	struct vnode *vp;
3130 	struct file *fp;
3131 	int fd;
3132 
3133 	FILEDESC_LOCK_ASSERT(fdp);
3134 
3135 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3136 		fp = fget_locked(fdp, fd);
3137 		if (fp == NULL)
3138 			continue;
3139 		if (fp->f_type == DTYPE_VNODE) {
3140 			vp = fp->f_vnode;
3141 			if (vp->v_type == VDIR)
3142 				return (EPERM);
3143 		}
3144 	}
3145 	return (0);
3146 }
3147 
3148 /*
3149  * Common routine for kern_chroot() and jail_attach().  The caller is
3150  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3151  * authorize this operation.
3152  */
3153 int
3154 pwd_chroot(struct thread *td, struct vnode *vp)
3155 {
3156 	struct filedesc *fdp;
3157 	struct vnode *oldvp;
3158 	int error;
3159 
3160 	fdp = td->td_proc->p_fd;
3161 	FILEDESC_XLOCK(fdp);
3162 	if (chroot_allow_open_directories == 0 ||
3163 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3164 		error = chroot_refuse_vdir_fds(fdp);
3165 		if (error != 0) {
3166 			FILEDESC_XUNLOCK(fdp);
3167 			return (error);
3168 		}
3169 	}
3170 	oldvp = fdp->fd_rdir;
3171 	vrefact(vp);
3172 	fdp->fd_rdir = vp;
3173 	if (fdp->fd_jdir == NULL) {
3174 		vrefact(vp);
3175 		fdp->fd_jdir = vp;
3176 	}
3177 	FILEDESC_XUNLOCK(fdp);
3178 	vrele(oldvp);
3179 	return (0);
3180 }
3181 
3182 void
3183 pwd_chdir(struct thread *td, struct vnode *vp)
3184 {
3185 	struct filedesc *fdp;
3186 	struct vnode *oldvp;
3187 
3188 	fdp = td->td_proc->p_fd;
3189 	FILEDESC_XLOCK(fdp);
3190 	VNASSERT(vp->v_usecount > 0, vp,
3191 	    ("chdir to a vnode with zero usecount"));
3192 	oldvp = fdp->fd_cdir;
3193 	fdp->fd_cdir = vp;
3194 	FILEDESC_XUNLOCK(fdp);
3195 	vrele(oldvp);
3196 }
3197 
3198 /*
3199  * Scan all active processes and prisons to see if any of them have a current
3200  * or root directory of `olddp'. If so, replace them with the new mount point.
3201  */
3202 void
3203 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3204 {
3205 	struct filedesc *fdp;
3206 	struct prison *pr;
3207 	struct proc *p;
3208 	int nrele;
3209 
3210 	if (vrefcnt(olddp) == 1)
3211 		return;
3212 	nrele = 0;
3213 	sx_slock(&allproc_lock);
3214 	FOREACH_PROC_IN_SYSTEM(p) {
3215 		PROC_LOCK(p);
3216 		fdp = fdhold(p);
3217 		PROC_UNLOCK(p);
3218 		if (fdp == NULL)
3219 			continue;
3220 		FILEDESC_XLOCK(fdp);
3221 		if (fdp->fd_cdir == olddp) {
3222 			vrefact(newdp);
3223 			fdp->fd_cdir = newdp;
3224 			nrele++;
3225 		}
3226 		if (fdp->fd_rdir == olddp) {
3227 			vrefact(newdp);
3228 			fdp->fd_rdir = newdp;
3229 			nrele++;
3230 		}
3231 		if (fdp->fd_jdir == olddp) {
3232 			vrefact(newdp);
3233 			fdp->fd_jdir = newdp;
3234 			nrele++;
3235 		}
3236 		FILEDESC_XUNLOCK(fdp);
3237 		fddrop(fdp);
3238 	}
3239 	sx_sunlock(&allproc_lock);
3240 	if (rootvnode == olddp) {
3241 		vrefact(newdp);
3242 		rootvnode = newdp;
3243 		nrele++;
3244 	}
3245 	mtx_lock(&prison0.pr_mtx);
3246 	if (prison0.pr_root == olddp) {
3247 		vrefact(newdp);
3248 		prison0.pr_root = newdp;
3249 		nrele++;
3250 	}
3251 	mtx_unlock(&prison0.pr_mtx);
3252 	sx_slock(&allprison_lock);
3253 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3254 		mtx_lock(&pr->pr_mtx);
3255 		if (pr->pr_root == olddp) {
3256 			vrefact(newdp);
3257 			pr->pr_root = newdp;
3258 			nrele++;
3259 		}
3260 		mtx_unlock(&pr->pr_mtx);
3261 	}
3262 	sx_sunlock(&allprison_lock);
3263 	while (nrele--)
3264 		vrele(olddp);
3265 }
3266 
3267 struct filedesc_to_leader *
3268 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3269 {
3270 	struct filedesc_to_leader *fdtol;
3271 
3272 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3273 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3274 	fdtol->fdl_refcount = 1;
3275 	fdtol->fdl_holdcount = 0;
3276 	fdtol->fdl_wakeup = 0;
3277 	fdtol->fdl_leader = leader;
3278 	if (old != NULL) {
3279 		FILEDESC_XLOCK(fdp);
3280 		fdtol->fdl_next = old->fdl_next;
3281 		fdtol->fdl_prev = old;
3282 		old->fdl_next = fdtol;
3283 		fdtol->fdl_next->fdl_prev = fdtol;
3284 		FILEDESC_XUNLOCK(fdp);
3285 	} else {
3286 		fdtol->fdl_next = fdtol;
3287 		fdtol->fdl_prev = fdtol;
3288 	}
3289 	return (fdtol);
3290 }
3291 
3292 static int
3293 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3294 {
3295 	struct filedesc *fdp;
3296 	int i, count, slots;
3297 
3298 	if (*(int *)arg1 != 0)
3299 		return (EINVAL);
3300 
3301 	fdp = curproc->p_fd;
3302 	count = 0;
3303 	FILEDESC_SLOCK(fdp);
3304 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3305 	for (i = 0; i < slots; i++)
3306 		count += bitcountl(fdp->fd_map[i]);
3307 	FILEDESC_SUNLOCK(fdp);
3308 
3309 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3310 }
3311 
3312 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3313     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3314     "Number of open file descriptors");
3315 
3316 /*
3317  * Get file structures globally.
3318  */
3319 static int
3320 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3321 {
3322 	struct xfile xf;
3323 	struct filedesc *fdp;
3324 	struct file *fp;
3325 	struct proc *p;
3326 	int error, n;
3327 
3328 	error = sysctl_wire_old_buffer(req, 0);
3329 	if (error != 0)
3330 		return (error);
3331 	if (req->oldptr == NULL) {
3332 		n = 0;
3333 		sx_slock(&allproc_lock);
3334 		FOREACH_PROC_IN_SYSTEM(p) {
3335 			PROC_LOCK(p);
3336 			if (p->p_state == PRS_NEW) {
3337 				PROC_UNLOCK(p);
3338 				continue;
3339 			}
3340 			fdp = fdhold(p);
3341 			PROC_UNLOCK(p);
3342 			if (fdp == NULL)
3343 				continue;
3344 			/* overestimates sparse tables. */
3345 			if (fdp->fd_lastfile > 0)
3346 				n += fdp->fd_lastfile;
3347 			fddrop(fdp);
3348 		}
3349 		sx_sunlock(&allproc_lock);
3350 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3351 	}
3352 	error = 0;
3353 	bzero(&xf, sizeof(xf));
3354 	xf.xf_size = sizeof(xf);
3355 	sx_slock(&allproc_lock);
3356 	FOREACH_PROC_IN_SYSTEM(p) {
3357 		PROC_LOCK(p);
3358 		if (p->p_state == PRS_NEW) {
3359 			PROC_UNLOCK(p);
3360 			continue;
3361 		}
3362 		if (p_cansee(req->td, p) != 0) {
3363 			PROC_UNLOCK(p);
3364 			continue;
3365 		}
3366 		xf.xf_pid = p->p_pid;
3367 		xf.xf_uid = p->p_ucred->cr_uid;
3368 		fdp = fdhold(p);
3369 		PROC_UNLOCK(p);
3370 		if (fdp == NULL)
3371 			continue;
3372 		FILEDESC_SLOCK(fdp);
3373 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3374 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3375 				continue;
3376 			xf.xf_fd = n;
3377 			xf.xf_file = (uintptr_t)fp;
3378 			xf.xf_data = (uintptr_t)fp->f_data;
3379 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3380 			xf.xf_type = (uintptr_t)fp->f_type;
3381 			xf.xf_count = fp->f_count;
3382 			xf.xf_msgcount = 0;
3383 			xf.xf_offset = foffset_get(fp);
3384 			xf.xf_flag = fp->f_flag;
3385 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3386 			if (error)
3387 				break;
3388 		}
3389 		FILEDESC_SUNLOCK(fdp);
3390 		fddrop(fdp);
3391 		if (error)
3392 			break;
3393 	}
3394 	sx_sunlock(&allproc_lock);
3395 	return (error);
3396 }
3397 
3398 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3399     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3400 
3401 #ifdef KINFO_FILE_SIZE
3402 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3403 #endif
3404 
3405 static int
3406 xlate_fflags(int fflags)
3407 {
3408 	static const struct {
3409 		int	fflag;
3410 		int	kf_fflag;
3411 	} fflags_table[] = {
3412 		{ FAPPEND, KF_FLAG_APPEND },
3413 		{ FASYNC, KF_FLAG_ASYNC },
3414 		{ FFSYNC, KF_FLAG_FSYNC },
3415 		{ FHASLOCK, KF_FLAG_HASLOCK },
3416 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3417 		{ FREAD, KF_FLAG_READ },
3418 		{ FWRITE, KF_FLAG_WRITE },
3419 		{ O_CREAT, KF_FLAG_CREAT },
3420 		{ O_DIRECT, KF_FLAG_DIRECT },
3421 		{ O_EXCL, KF_FLAG_EXCL },
3422 		{ O_EXEC, KF_FLAG_EXEC },
3423 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3424 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3425 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3426 		{ O_TRUNC, KF_FLAG_TRUNC }
3427 	};
3428 	unsigned int i;
3429 	int kflags;
3430 
3431 	kflags = 0;
3432 	for (i = 0; i < nitems(fflags_table); i++)
3433 		if (fflags & fflags_table[i].fflag)
3434 			kflags |=  fflags_table[i].kf_fflag;
3435 	return (kflags);
3436 }
3437 
3438 /* Trim unused data from kf_path by truncating the structure size. */
3439 void
3440 pack_kinfo(struct kinfo_file *kif)
3441 {
3442 
3443 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3444 	    strlen(kif->kf_path) + 1;
3445 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3446 }
3447 
3448 static void
3449 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3450     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3451 {
3452 	int error;
3453 
3454 	bzero(kif, sizeof(*kif));
3455 
3456 	/* Set a default type to allow for empty fill_kinfo() methods. */
3457 	kif->kf_type = KF_TYPE_UNKNOWN;
3458 	kif->kf_flags = xlate_fflags(fp->f_flag);
3459 	if (rightsp != NULL)
3460 		kif->kf_cap_rights = *rightsp;
3461 	else
3462 		cap_rights_init(&kif->kf_cap_rights);
3463 	kif->kf_fd = fd;
3464 	kif->kf_ref_count = fp->f_count;
3465 	kif->kf_offset = foffset_get(fp);
3466 
3467 	/*
3468 	 * This may drop the filedesc lock, so the 'fp' cannot be
3469 	 * accessed after this call.
3470 	 */
3471 	error = fo_fill_kinfo(fp, kif, fdp);
3472 	if (error == 0)
3473 		kif->kf_status |= KF_ATTR_VALID;
3474 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3475 		pack_kinfo(kif);
3476 	else
3477 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3478 }
3479 
3480 static void
3481 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3482     struct kinfo_file *kif, int flags)
3483 {
3484 	int error;
3485 
3486 	bzero(kif, sizeof(*kif));
3487 
3488 	kif->kf_type = KF_TYPE_VNODE;
3489 	error = vn_fill_kinfo_vnode(vp, kif);
3490 	if (error == 0)
3491 		kif->kf_status |= KF_ATTR_VALID;
3492 	kif->kf_flags = xlate_fflags(fflags);
3493 	cap_rights_init(&kif->kf_cap_rights);
3494 	kif->kf_fd = fd;
3495 	kif->kf_ref_count = -1;
3496 	kif->kf_offset = -1;
3497 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3498 		pack_kinfo(kif);
3499 	else
3500 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3501 	vrele(vp);
3502 }
3503 
3504 struct export_fd_buf {
3505 	struct filedesc		*fdp;
3506 	struct sbuf 		*sb;
3507 	ssize_t			remainder;
3508 	struct kinfo_file	kif;
3509 	int			flags;
3510 };
3511 
3512 static int
3513 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3514 {
3515 	struct kinfo_file *kif;
3516 
3517 	kif = &efbuf->kif;
3518 	if (efbuf->remainder != -1) {
3519 		if (efbuf->remainder < kif->kf_structsize) {
3520 			/* Terminate export. */
3521 			efbuf->remainder = 0;
3522 			return (0);
3523 		}
3524 		efbuf->remainder -= kif->kf_structsize;
3525 	}
3526 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3527 }
3528 
3529 static int
3530 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3531     struct export_fd_buf *efbuf)
3532 {
3533 	int error;
3534 
3535 	if (efbuf->remainder == 0)
3536 		return (0);
3537 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3538 	    efbuf->flags);
3539 	FILEDESC_SUNLOCK(efbuf->fdp);
3540 	error = export_kinfo_to_sb(efbuf);
3541 	FILEDESC_SLOCK(efbuf->fdp);
3542 	return (error);
3543 }
3544 
3545 static int
3546 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3547     struct export_fd_buf *efbuf)
3548 {
3549 	int error;
3550 
3551 	if (efbuf->remainder == 0)
3552 		return (0);
3553 	if (efbuf->fdp != NULL)
3554 		FILEDESC_SUNLOCK(efbuf->fdp);
3555 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3556 	error = export_kinfo_to_sb(efbuf);
3557 	if (efbuf->fdp != NULL)
3558 		FILEDESC_SLOCK(efbuf->fdp);
3559 	return (error);
3560 }
3561 
3562 /*
3563  * Store a process file descriptor information to sbuf.
3564  *
3565  * Takes a locked proc as argument, and returns with the proc unlocked.
3566  */
3567 int
3568 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3569     int flags)
3570 {
3571 	struct file *fp;
3572 	struct filedesc *fdp;
3573 	struct export_fd_buf *efbuf;
3574 	struct vnode *cttyvp, *textvp, *tracevp;
3575 	int error, i;
3576 	cap_rights_t rights;
3577 
3578 	PROC_LOCK_ASSERT(p, MA_OWNED);
3579 
3580 	/* ktrace vnode */
3581 	tracevp = p->p_tracevp;
3582 	if (tracevp != NULL)
3583 		vrefact(tracevp);
3584 	/* text vnode */
3585 	textvp = p->p_textvp;
3586 	if (textvp != NULL)
3587 		vrefact(textvp);
3588 	/* Controlling tty. */
3589 	cttyvp = NULL;
3590 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3591 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3592 		if (cttyvp != NULL)
3593 			vrefact(cttyvp);
3594 	}
3595 	fdp = fdhold(p);
3596 	PROC_UNLOCK(p);
3597 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3598 	efbuf->fdp = NULL;
3599 	efbuf->sb = sb;
3600 	efbuf->remainder = maxlen;
3601 	efbuf->flags = flags;
3602 	if (tracevp != NULL)
3603 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3604 		    efbuf);
3605 	if (textvp != NULL)
3606 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3607 	if (cttyvp != NULL)
3608 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3609 		    efbuf);
3610 	error = 0;
3611 	if (fdp == NULL)
3612 		goto fail;
3613 	efbuf->fdp = fdp;
3614 	FILEDESC_SLOCK(fdp);
3615 	/* working directory */
3616 	if (fdp->fd_cdir != NULL) {
3617 		vrefact(fdp->fd_cdir);
3618 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3619 	}
3620 	/* root directory */
3621 	if (fdp->fd_rdir != NULL) {
3622 		vrefact(fdp->fd_rdir);
3623 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3624 	}
3625 	/* jail directory */
3626 	if (fdp->fd_jdir != NULL) {
3627 		vrefact(fdp->fd_jdir);
3628 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3629 	}
3630 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3631 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3632 			continue;
3633 #ifdef CAPABILITIES
3634 		rights = *cap_rights(fdp, i);
3635 #else /* !CAPABILITIES */
3636 		rights = cap_no_rights;
3637 #endif
3638 		/*
3639 		 * Create sysctl entry.  It is OK to drop the filedesc
3640 		 * lock inside of export_file_to_sb() as we will
3641 		 * re-validate and re-evaluate its properties when the
3642 		 * loop continues.
3643 		 */
3644 		error = export_file_to_sb(fp, i, &rights, efbuf);
3645 		if (error != 0 || efbuf->remainder == 0)
3646 			break;
3647 	}
3648 	FILEDESC_SUNLOCK(fdp);
3649 	fddrop(fdp);
3650 fail:
3651 	free(efbuf, M_TEMP);
3652 	return (error);
3653 }
3654 
3655 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3656 
3657 /*
3658  * Get per-process file descriptors for use by procstat(1), et al.
3659  */
3660 static int
3661 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3662 {
3663 	struct sbuf sb;
3664 	struct proc *p;
3665 	ssize_t maxlen;
3666 	int error, error2, *name;
3667 
3668 	name = (int *)arg1;
3669 
3670 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3671 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3672 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3673 	if (error != 0) {
3674 		sbuf_delete(&sb);
3675 		return (error);
3676 	}
3677 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3678 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3679 	    KERN_FILEDESC_PACK_KINFO);
3680 	error2 = sbuf_finish(&sb);
3681 	sbuf_delete(&sb);
3682 	return (error != 0 ? error : error2);
3683 }
3684 
3685 #ifdef COMPAT_FREEBSD7
3686 #ifdef KINFO_OFILE_SIZE
3687 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3688 #endif
3689 
3690 static void
3691 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3692 {
3693 
3694 	okif->kf_structsize = sizeof(*okif);
3695 	okif->kf_type = kif->kf_type;
3696 	okif->kf_fd = kif->kf_fd;
3697 	okif->kf_ref_count = kif->kf_ref_count;
3698 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3699 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3700 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3701 	okif->kf_offset = kif->kf_offset;
3702 	if (kif->kf_type == KF_TYPE_VNODE)
3703 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3704 	else
3705 		okif->kf_vnode_type = KF_VTYPE_VNON;
3706 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3707 	if (kif->kf_type == KF_TYPE_SOCKET) {
3708 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3709 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3710 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3711 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3712 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3713 	} else {
3714 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3715 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3716 	}
3717 }
3718 
3719 static int
3720 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3721     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3722 {
3723 	int error;
3724 
3725 	vrefact(vp);
3726 	FILEDESC_SUNLOCK(fdp);
3727 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3728 	kinfo_to_okinfo(kif, okif);
3729 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3730 	FILEDESC_SLOCK(fdp);
3731 	return (error);
3732 }
3733 
3734 /*
3735  * Get per-process file descriptors for use by procstat(1), et al.
3736  */
3737 static int
3738 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3739 {
3740 	struct kinfo_ofile *okif;
3741 	struct kinfo_file *kif;
3742 	struct filedesc *fdp;
3743 	int error, i, *name;
3744 	struct file *fp;
3745 	struct proc *p;
3746 
3747 	name = (int *)arg1;
3748 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3749 	if (error != 0)
3750 		return (error);
3751 	fdp = fdhold(p);
3752 	PROC_UNLOCK(p);
3753 	if (fdp == NULL)
3754 		return (ENOENT);
3755 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3756 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3757 	FILEDESC_SLOCK(fdp);
3758 	if (fdp->fd_cdir != NULL)
3759 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3760 		    okif, fdp, req);
3761 	if (fdp->fd_rdir != NULL)
3762 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3763 		    okif, fdp, req);
3764 	if (fdp->fd_jdir != NULL)
3765 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3766 		    okif, fdp, req);
3767 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3768 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3769 			continue;
3770 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3771 		    KERN_FILEDESC_PACK_KINFO);
3772 		FILEDESC_SUNLOCK(fdp);
3773 		kinfo_to_okinfo(kif, okif);
3774 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3775 		FILEDESC_SLOCK(fdp);
3776 		if (error)
3777 			break;
3778 	}
3779 	FILEDESC_SUNLOCK(fdp);
3780 	fddrop(fdp);
3781 	free(kif, M_TEMP);
3782 	free(okif, M_TEMP);
3783 	return (0);
3784 }
3785 
3786 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3787     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3788     "Process ofiledesc entries");
3789 #endif	/* COMPAT_FREEBSD7 */
3790 
3791 int
3792 vntype_to_kinfo(int vtype)
3793 {
3794 	struct {
3795 		int	vtype;
3796 		int	kf_vtype;
3797 	} vtypes_table[] = {
3798 		{ VBAD, KF_VTYPE_VBAD },
3799 		{ VBLK, KF_VTYPE_VBLK },
3800 		{ VCHR, KF_VTYPE_VCHR },
3801 		{ VDIR, KF_VTYPE_VDIR },
3802 		{ VFIFO, KF_VTYPE_VFIFO },
3803 		{ VLNK, KF_VTYPE_VLNK },
3804 		{ VNON, KF_VTYPE_VNON },
3805 		{ VREG, KF_VTYPE_VREG },
3806 		{ VSOCK, KF_VTYPE_VSOCK }
3807 	};
3808 	unsigned int i;
3809 
3810 	/*
3811 	 * Perform vtype translation.
3812 	 */
3813 	for (i = 0; i < nitems(vtypes_table); i++)
3814 		if (vtypes_table[i].vtype == vtype)
3815 			return (vtypes_table[i].kf_vtype);
3816 
3817 	return (KF_VTYPE_UNKNOWN);
3818 }
3819 
3820 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3821     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3822     "Process filedesc entries");
3823 
3824 /*
3825  * Store a process current working directory information to sbuf.
3826  *
3827  * Takes a locked proc as argument, and returns with the proc unlocked.
3828  */
3829 int
3830 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3831 {
3832 	struct filedesc *fdp;
3833 	struct export_fd_buf *efbuf;
3834 	int error;
3835 
3836 	PROC_LOCK_ASSERT(p, MA_OWNED);
3837 
3838 	fdp = fdhold(p);
3839 	PROC_UNLOCK(p);
3840 	if (fdp == NULL)
3841 		return (EINVAL);
3842 
3843 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3844 	efbuf->fdp = fdp;
3845 	efbuf->sb = sb;
3846 	efbuf->remainder = maxlen;
3847 
3848 	FILEDESC_SLOCK(fdp);
3849 	if (fdp->fd_cdir == NULL)
3850 		error = EINVAL;
3851 	else {
3852 		vrefact(fdp->fd_cdir);
3853 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3854 		    FREAD, efbuf);
3855 	}
3856 	FILEDESC_SUNLOCK(fdp);
3857 	fddrop(fdp);
3858 	free(efbuf, M_TEMP);
3859 	return (error);
3860 }
3861 
3862 /*
3863  * Get per-process current working directory.
3864  */
3865 static int
3866 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3867 {
3868 	struct sbuf sb;
3869 	struct proc *p;
3870 	ssize_t maxlen;
3871 	int error, error2, *name;
3872 
3873 	name = (int *)arg1;
3874 
3875 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3876 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3877 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3878 	if (error != 0) {
3879 		sbuf_delete(&sb);
3880 		return (error);
3881 	}
3882 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3883 	error = kern_proc_cwd_out(p, &sb, maxlen);
3884 	error2 = sbuf_finish(&sb);
3885 	sbuf_delete(&sb);
3886 	return (error != 0 ? error : error2);
3887 }
3888 
3889 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3890     sysctl_kern_proc_cwd, "Process current working directory");
3891 
3892 #ifdef DDB
3893 /*
3894  * For the purposes of debugging, generate a human-readable string for the
3895  * file type.
3896  */
3897 static const char *
3898 file_type_to_name(short type)
3899 {
3900 
3901 	switch (type) {
3902 	case 0:
3903 		return ("zero");
3904 	case DTYPE_VNODE:
3905 		return ("vnode");
3906 	case DTYPE_SOCKET:
3907 		return ("socket");
3908 	case DTYPE_PIPE:
3909 		return ("pipe");
3910 	case DTYPE_FIFO:
3911 		return ("fifo");
3912 	case DTYPE_KQUEUE:
3913 		return ("kqueue");
3914 	case DTYPE_CRYPTO:
3915 		return ("crypto");
3916 	case DTYPE_MQUEUE:
3917 		return ("mqueue");
3918 	case DTYPE_SHM:
3919 		return ("shm");
3920 	case DTYPE_SEM:
3921 		return ("ksem");
3922 	case DTYPE_PTS:
3923 		return ("pts");
3924 	case DTYPE_DEV:
3925 		return ("dev");
3926 	case DTYPE_PROCDESC:
3927 		return ("proc");
3928 	case DTYPE_LINUXEFD:
3929 		return ("levent");
3930 	case DTYPE_LINUXTFD:
3931 		return ("ltimer");
3932 	default:
3933 		return ("unkn");
3934 	}
3935 }
3936 
3937 /*
3938  * For the purposes of debugging, identify a process (if any, perhaps one of
3939  * many) that references the passed file in its file descriptor array. Return
3940  * NULL if none.
3941  */
3942 static struct proc *
3943 file_to_first_proc(struct file *fp)
3944 {
3945 	struct filedesc *fdp;
3946 	struct proc *p;
3947 	int n;
3948 
3949 	FOREACH_PROC_IN_SYSTEM(p) {
3950 		if (p->p_state == PRS_NEW)
3951 			continue;
3952 		fdp = p->p_fd;
3953 		if (fdp == NULL)
3954 			continue;
3955 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3956 			if (fp == fdp->fd_ofiles[n].fde_file)
3957 				return (p);
3958 		}
3959 	}
3960 	return (NULL);
3961 }
3962 
3963 static void
3964 db_print_file(struct file *fp, int header)
3965 {
3966 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3967 	struct proc *p;
3968 
3969 	if (header)
3970 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3971 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3972 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3973 		    "FCmd");
3974 	p = file_to_first_proc(fp);
3975 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3976 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3977 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3978 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3979 
3980 #undef XPTRWIDTH
3981 }
3982 
3983 DB_SHOW_COMMAND(file, db_show_file)
3984 {
3985 	struct file *fp;
3986 
3987 	if (!have_addr) {
3988 		db_printf("usage: show file <addr>\n");
3989 		return;
3990 	}
3991 	fp = (struct file *)addr;
3992 	db_print_file(fp, 1);
3993 }
3994 
3995 DB_SHOW_COMMAND(files, db_show_files)
3996 {
3997 	struct filedesc *fdp;
3998 	struct file *fp;
3999 	struct proc *p;
4000 	int header;
4001 	int n;
4002 
4003 	header = 1;
4004 	FOREACH_PROC_IN_SYSTEM(p) {
4005 		if (p->p_state == PRS_NEW)
4006 			continue;
4007 		if ((fdp = p->p_fd) == NULL)
4008 			continue;
4009 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
4010 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4011 				continue;
4012 			db_print_file(fp, header);
4013 			header = 0;
4014 		}
4015 	}
4016 }
4017 #endif
4018 
4019 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4020     &maxfilesperproc, 0, "Maximum files allowed open per process");
4021 
4022 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4023     &maxfiles, 0, "Maximum number of files");
4024 
4025 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4026     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4027 
4028 /* ARGSUSED*/
4029 static void
4030 filelistinit(void *dummy)
4031 {
4032 
4033 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4034 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4035 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4036 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4037 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4038 }
4039 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4040 
4041 /*-------------------------------------------------------------------*/
4042 
4043 static int
4044 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4045     int flags, struct thread *td)
4046 {
4047 
4048 	return (EBADF);
4049 }
4050 
4051 static int
4052 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4053     struct thread *td)
4054 {
4055 
4056 	return (EINVAL);
4057 }
4058 
4059 static int
4060 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4061     struct thread *td)
4062 {
4063 
4064 	return (EBADF);
4065 }
4066 
4067 static int
4068 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4069     struct thread *td)
4070 {
4071 
4072 	return (0);
4073 }
4074 
4075 static int
4076 badfo_kqfilter(struct file *fp, struct knote *kn)
4077 {
4078 
4079 	return (EBADF);
4080 }
4081 
4082 static int
4083 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4084     struct thread *td)
4085 {
4086 
4087 	return (EBADF);
4088 }
4089 
4090 static int
4091 badfo_close(struct file *fp, struct thread *td)
4092 {
4093 
4094 	return (0);
4095 }
4096 
4097 static int
4098 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4099     struct thread *td)
4100 {
4101 
4102 	return (EBADF);
4103 }
4104 
4105 static int
4106 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4107     struct thread *td)
4108 {
4109 
4110 	return (EBADF);
4111 }
4112 
4113 static int
4114 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4115     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4116     struct thread *td)
4117 {
4118 
4119 	return (EBADF);
4120 }
4121 
4122 static int
4123 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4124 {
4125 
4126 	return (0);
4127 }
4128 
4129 struct fileops badfileops = {
4130 	.fo_read = badfo_readwrite,
4131 	.fo_write = badfo_readwrite,
4132 	.fo_truncate = badfo_truncate,
4133 	.fo_ioctl = badfo_ioctl,
4134 	.fo_poll = badfo_poll,
4135 	.fo_kqfilter = badfo_kqfilter,
4136 	.fo_stat = badfo_stat,
4137 	.fo_close = badfo_close,
4138 	.fo_chmod = badfo_chmod,
4139 	.fo_chown = badfo_chown,
4140 	.fo_sendfile = badfo_sendfile,
4141 	.fo_fill_kinfo = badfo_fill_kinfo,
4142 };
4143 
4144 int
4145 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4146     int flags, struct thread *td)
4147 {
4148 
4149 	return (EOPNOTSUPP);
4150 }
4151 
4152 int
4153 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4154     struct thread *td)
4155 {
4156 
4157 	return (EINVAL);
4158 }
4159 
4160 int
4161 invfo_ioctl(struct file *fp, u_long com, void *data,
4162     struct ucred *active_cred, struct thread *td)
4163 {
4164 
4165 	return (ENOTTY);
4166 }
4167 
4168 int
4169 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4170     struct thread *td)
4171 {
4172 
4173 	return (poll_no_poll(events));
4174 }
4175 
4176 int
4177 invfo_kqfilter(struct file *fp, struct knote *kn)
4178 {
4179 
4180 	return (EINVAL);
4181 }
4182 
4183 int
4184 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4185     struct thread *td)
4186 {
4187 
4188 	return (EINVAL);
4189 }
4190 
4191 int
4192 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4193     struct thread *td)
4194 {
4195 
4196 	return (EINVAL);
4197 }
4198 
4199 int
4200 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4201     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4202     struct thread *td)
4203 {
4204 
4205 	return (EINVAL);
4206 }
4207 
4208 /*-------------------------------------------------------------------*/
4209 
4210 /*
4211  * File Descriptor pseudo-device driver (/dev/fd/).
4212  *
4213  * Opening minor device N dup()s the file (if any) connected to file
4214  * descriptor N belonging to the calling process.  Note that this driver
4215  * consists of only the ``open()'' routine, because all subsequent
4216  * references to this file will be direct to the other driver.
4217  *
4218  * XXX: we could give this one a cloning event handler if necessary.
4219  */
4220 
4221 /* ARGSUSED */
4222 static int
4223 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4224 {
4225 
4226 	/*
4227 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4228 	 * the file descriptor being sought for duplication. The error
4229 	 * return ensures that the vnode for this device will be released
4230 	 * by vn_open. Open will detect this special error and take the
4231 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4232 	 * will simply report the error.
4233 	 */
4234 	td->td_dupfd = dev2unit(dev);
4235 	return (ENODEV);
4236 }
4237 
4238 static struct cdevsw fildesc_cdevsw = {
4239 	.d_version =	D_VERSION,
4240 	.d_open =	fdopen,
4241 	.d_name =	"FD",
4242 };
4243 
4244 static void
4245 fildesc_drvinit(void *unused)
4246 {
4247 	struct cdev *dev;
4248 
4249 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4250 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4251 	make_dev_alias(dev, "stdin");
4252 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4253 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4254 	make_dev_alias(dev, "stdout");
4255 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4256 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4257 	make_dev_alias(dev, "stderr");
4258 }
4259 
4260 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4261