xref: /freebsd/sys/kern/kern_descrip.c (revision d485c77f203fb0f4cdc08dea5ff81631b51d8809)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98     "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 
102 MALLOC_DECLARE(M_FADVISE);
103 
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
107 VFS_SMR_DECLARE;
108 
109 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
110 		    struct thread *td, bool holdleaders, bool audit);
111 static int	fd_first_free(struct filedesc *fdp, int low, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the last used fd.
217  *
218  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219  * Otherwise use fdlastfile.
220  */
221 int
222 fdlastfile_single(struct filedesc *fdp)
223 {
224 	NDSLOTTYPE *map = fdp->fd_map;
225 	int off, minoff;
226 
227 	off = NDSLOT(fdp->fd_nfiles - 1);
228 	for (minoff = NDSLOT(0); off >= minoff; --off)
229 		if (map[off] != 0)
230 			return (off * NDENTRIES + flsl(map[off]) - 1);
231 	return (-1);
232 }
233 
234 int
235 fdlastfile(struct filedesc *fdp)
236 {
237 
238 	FILEDESC_LOCK_ASSERT(fdp);
239 	return (fdlastfile_single(fdp));
240 }
241 
242 static int
243 fdisused(struct filedesc *fdp, int fd)
244 {
245 
246 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 
249 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
250 }
251 
252 /*
253  * Mark a file descriptor as used.
254  */
255 static void
256 fdused_init(struct filedesc *fdp, int fd)
257 {
258 
259 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 
261 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
262 }
263 
264 static void
265 fdused(struct filedesc *fdp, int fd)
266 {
267 
268 	FILEDESC_XLOCK_ASSERT(fdp);
269 
270 	fdused_init(fdp, fd);
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile++;
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	FILEDESC_XLOCK_ASSERT(fdp);
311 	fde = &fdp->fd_ofiles[fd];
312 #ifdef CAPABILITIES
313 	seqc_write_begin(&fde->fde_seqc);
314 #endif
315 	fde->fde_file = NULL;
316 #ifdef CAPABILITIES
317 	seqc_write_end(&fde->fde_seqc);
318 #endif
319 	fdefree_last(fde);
320 	fdunused(fdp, fd);
321 }
322 
323 /*
324  * System calls on descriptors.
325  */
326 #ifndef _SYS_SYSPROTO_H_
327 struct getdtablesize_args {
328 	int	dummy;
329 };
330 #endif
331 /* ARGSUSED */
332 int
333 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
334 {
335 #ifdef	RACCT
336 	uint64_t lim;
337 #endif
338 
339 	td->td_retval[0] = getmaxfd(td);
340 #ifdef	RACCT
341 	PROC_LOCK(td->td_proc);
342 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
343 	PROC_UNLOCK(td->td_proc);
344 	if (lim < td->td_retval[0])
345 		td->td_retval[0] = lim;
346 #endif
347 	return (0);
348 }
349 
350 /*
351  * Duplicate a file descriptor to a particular value.
352  *
353  * Note: keep in mind that a potential race condition exists when closing
354  * descriptors from a shared descriptor table (via rfork).
355  */
356 #ifndef _SYS_SYSPROTO_H_
357 struct dup2_args {
358 	u_int	from;
359 	u_int	to;
360 };
361 #endif
362 /* ARGSUSED */
363 int
364 sys_dup2(struct thread *td, struct dup2_args *uap)
365 {
366 
367 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
368 }
369 
370 /*
371  * Duplicate a file descriptor.
372  */
373 #ifndef _SYS_SYSPROTO_H_
374 struct dup_args {
375 	u_int	fd;
376 };
377 #endif
378 /* ARGSUSED */
379 int
380 sys_dup(struct thread *td, struct dup_args *uap)
381 {
382 
383 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
384 }
385 
386 /*
387  * The file control system call.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct fcntl_args {
391 	int	fd;
392 	int	cmd;
393 	long	arg;
394 };
395 #endif
396 /* ARGSUSED */
397 int
398 sys_fcntl(struct thread *td, struct fcntl_args *uap)
399 {
400 
401 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
402 }
403 
404 int
405 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
406 {
407 	struct flock fl;
408 	struct __oflock ofl;
409 	intptr_t arg1;
410 	int error, newcmd;
411 
412 	error = 0;
413 	newcmd = cmd;
414 	switch (cmd) {
415 	case F_OGETLK:
416 	case F_OSETLK:
417 	case F_OSETLKW:
418 		/*
419 		 * Convert old flock structure to new.
420 		 */
421 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
422 		fl.l_start = ofl.l_start;
423 		fl.l_len = ofl.l_len;
424 		fl.l_pid = ofl.l_pid;
425 		fl.l_type = ofl.l_type;
426 		fl.l_whence = ofl.l_whence;
427 		fl.l_sysid = 0;
428 
429 		switch (cmd) {
430 		case F_OGETLK:
431 			newcmd = F_GETLK;
432 			break;
433 		case F_OSETLK:
434 			newcmd = F_SETLK;
435 			break;
436 		case F_OSETLKW:
437 			newcmd = F_SETLKW;
438 			break;
439 		}
440 		arg1 = (intptr_t)&fl;
441 		break;
442 	case F_GETLK:
443 	case F_SETLK:
444 	case F_SETLKW:
445 	case F_SETLK_REMOTE:
446 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
447 		arg1 = (intptr_t)&fl;
448 		break;
449 	default:
450 		arg1 = arg;
451 		break;
452 	}
453 	if (error)
454 		return (error);
455 	error = kern_fcntl(td, fd, newcmd, arg1);
456 	if (error)
457 		return (error);
458 	if (cmd == F_OGETLK) {
459 		ofl.l_start = fl.l_start;
460 		ofl.l_len = fl.l_len;
461 		ofl.l_pid = fl.l_pid;
462 		ofl.l_type = fl.l_type;
463 		ofl.l_whence = fl.l_whence;
464 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
465 	} else if (cmd == F_GETLK) {
466 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
467 	}
468 	return (error);
469 }
470 
471 int
472 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 {
474 	struct filedesc *fdp;
475 	struct flock *flp;
476 	struct file *fp, *fp2;
477 	struct filedescent *fde;
478 	struct proc *p;
479 	struct vnode *vp;
480 	struct mount *mp;
481 	int error, flg, seals, tmp;
482 	uint64_t bsize;
483 	off_t foffset;
484 
485 	error = 0;
486 	flg = F_POSIX;
487 	p = td->td_proc;
488 	fdp = p->p_fd;
489 
490 	AUDIT_ARG_FD(cmd);
491 	AUDIT_ARG_CMD(cmd);
492 	switch (cmd) {
493 	case F_DUPFD:
494 		tmp = arg;
495 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
496 		break;
497 
498 	case F_DUPFD_CLOEXEC:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
501 		break;
502 
503 	case F_DUP2FD:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD_CLOEXEC:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
511 		break;
512 
513 	case F_GETFD:
514 		error = EBADF;
515 		FILEDESC_SLOCK(fdp);
516 		fde = fdeget_locked(fdp, fd);
517 		if (fde != NULL) {
518 			td->td_retval[0] =
519 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
520 			error = 0;
521 		}
522 		FILEDESC_SUNLOCK(fdp);
523 		break;
524 
525 	case F_SETFD:
526 		error = EBADF;
527 		FILEDESC_XLOCK(fdp);
528 		fde = fdeget_locked(fdp, fd);
529 		if (fde != NULL) {
530 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
531 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
532 			error = 0;
533 		}
534 		FILEDESC_XUNLOCK(fdp);
535 		break;
536 
537 	case F_GETFL:
538 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
539 		if (error != 0)
540 			break;
541 		td->td_retval[0] = OFLAGS(fp->f_flag);
542 		fdrop(fp, td);
543 		break;
544 
545 	case F_SETFL:
546 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
547 		if (error != 0)
548 			break;
549 		do {
550 			tmp = flg = fp->f_flag;
551 			tmp &= ~FCNTLFLAGS;
552 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
553 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
554 		tmp = fp->f_flag & FNONBLOCK;
555 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
556 		if (error != 0) {
557 			fdrop(fp, td);
558 			break;
559 		}
560 		tmp = fp->f_flag & FASYNC;
561 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
562 		if (error == 0) {
563 			fdrop(fp, td);
564 			break;
565 		}
566 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
567 		tmp = 0;
568 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
569 		fdrop(fp, td);
570 		break;
571 
572 	case F_GETOWN:
573 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
574 		if (error != 0)
575 			break;
576 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 		if (error == 0)
578 			td->td_retval[0] = tmp;
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_SETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		tmp = arg;
587 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
588 		fdrop(fp, td);
589 		break;
590 
591 	case F_SETLK_REMOTE:
592 		error = priv_check(td, PRIV_NFS_LOCKD);
593 		if (error != 0)
594 			return (error);
595 		flg = F_REMOTE;
596 		goto do_setlk;
597 
598 	case F_SETLKW:
599 		flg |= F_WAIT;
600 		/* FALLTHROUGH F_SETLK */
601 
602 	case F_SETLK:
603 	do_setlk:
604 		flp = (struct flock *)arg;
605 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
606 			error = EINVAL;
607 			break;
608 		}
609 
610 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
611 		if (error != 0)
612 			break;
613 		if (fp->f_type != DTYPE_VNODE) {
614 			error = EBADF;
615 			fdrop(fp, td);
616 			break;
617 		}
618 
619 		if (flp->l_whence == SEEK_CUR) {
620 			foffset = foffset_get(fp);
621 			if (foffset < 0 ||
622 			    (flp->l_start > 0 &&
623 			     foffset > OFF_MAX - flp->l_start)) {
624 				error = EOVERFLOW;
625 				fdrop(fp, td);
626 				break;
627 			}
628 			flp->l_start += foffset;
629 		}
630 
631 		vp = fp->f_vnode;
632 		switch (flp->l_type) {
633 		case F_RDLCK:
634 			if ((fp->f_flag & FREAD) == 0) {
635 				error = EBADF;
636 				break;
637 			}
638 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
639 				PROC_LOCK(p->p_leader);
640 				p->p_leader->p_flag |= P_ADVLOCK;
641 				PROC_UNLOCK(p->p_leader);
642 			}
643 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
644 			    flp, flg);
645 			break;
646 		case F_WRLCK:
647 			if ((fp->f_flag & FWRITE) == 0) {
648 				error = EBADF;
649 				break;
650 			}
651 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
652 				PROC_LOCK(p->p_leader);
653 				p->p_leader->p_flag |= P_ADVLOCK;
654 				PROC_UNLOCK(p->p_leader);
655 			}
656 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
657 			    flp, flg);
658 			break;
659 		case F_UNLCK:
660 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
661 			    flp, flg);
662 			break;
663 		case F_UNLCKSYS:
664 			if (flg != F_REMOTE) {
665 				error = EINVAL;
666 				break;
667 			}
668 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
669 			    F_UNLCKSYS, flp, flg);
670 			break;
671 		default:
672 			error = EINVAL;
673 			break;
674 		}
675 		if (error != 0 || flp->l_type == F_UNLCK ||
676 		    flp->l_type == F_UNLCKSYS) {
677 			fdrop(fp, td);
678 			break;
679 		}
680 
681 		/*
682 		 * Check for a race with close.
683 		 *
684 		 * The vnode is now advisory locked (or unlocked, but this case
685 		 * is not really important) as the caller requested.
686 		 * We had to drop the filedesc lock, so we need to recheck if
687 		 * the descriptor is still valid, because if it was closed
688 		 * in the meantime we need to remove advisory lock from the
689 		 * vnode - close on any descriptor leading to an advisory
690 		 * locked vnode, removes that lock.
691 		 * We will return 0 on purpose in that case, as the result of
692 		 * successful advisory lock might have been externally visible
693 		 * already. This is fine - effectively we pretend to the caller
694 		 * that the closing thread was a bit slower and that the
695 		 * advisory lock succeeded before the close.
696 		 */
697 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
698 		if (error != 0) {
699 			fdrop(fp, td);
700 			break;
701 		}
702 		if (fp != fp2) {
703 			flp->l_whence = SEEK_SET;
704 			flp->l_start = 0;
705 			flp->l_len = 0;
706 			flp->l_type = F_UNLCK;
707 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
708 			    F_UNLCK, flp, F_POSIX);
709 		}
710 		fdrop(fp, td);
711 		fdrop(fp2, td);
712 		break;
713 
714 	case F_GETLK:
715 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
716 		if (error != 0)
717 			break;
718 		if (fp->f_type != DTYPE_VNODE) {
719 			error = EBADF;
720 			fdrop(fp, td);
721 			break;
722 		}
723 		flp = (struct flock *)arg;
724 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
725 		    flp->l_type != F_UNLCK) {
726 			error = EINVAL;
727 			fdrop(fp, td);
728 			break;
729 		}
730 		if (flp->l_whence == SEEK_CUR) {
731 			foffset = foffset_get(fp);
732 			if ((flp->l_start > 0 &&
733 			    foffset > OFF_MAX - flp->l_start) ||
734 			    (flp->l_start < 0 &&
735 			    foffset < OFF_MIN - flp->l_start)) {
736 				error = EOVERFLOW;
737 				fdrop(fp, td);
738 				break;
739 			}
740 			flp->l_start += foffset;
741 		}
742 		vp = fp->f_vnode;
743 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
744 		    F_POSIX);
745 		fdrop(fp, td);
746 		break;
747 
748 	case F_ADD_SEALS:
749 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
750 		if (error != 0)
751 			break;
752 		error = fo_add_seals(fp, arg);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_GET_SEALS:
757 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
758 		if (error != 0)
759 			break;
760 		if (fo_get_seals(fp, &seals) == 0)
761 			td->td_retval[0] = seals;
762 		else
763 			error = EINVAL;
764 		fdrop(fp, td);
765 		break;
766 
767 	case F_RDAHEAD:
768 		arg = arg ? 128 * 1024: 0;
769 		/* FALLTHROUGH */
770 	case F_READAHEAD:
771 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
772 		if (error != 0)
773 			break;
774 		if (fp->f_type != DTYPE_VNODE) {
775 			fdrop(fp, td);
776 			error = EBADF;
777 			break;
778 		}
779 		vp = fp->f_vnode;
780 		if (vp->v_type != VREG) {
781 			fdrop(fp, td);
782 			error = ENOTTY;
783 			break;
784 		}
785 
786 		/*
787 		 * Exclusive lock synchronizes against f_seqcount reads and
788 		 * writes in sequential_heuristic().
789 		 */
790 		error = vn_lock(vp, LK_EXCLUSIVE);
791 		if (error != 0) {
792 			fdrop(fp, td);
793 			break;
794 		}
795 		if (arg >= 0) {
796 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
797 			arg = MIN(arg, INT_MAX - bsize + 1);
798 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
799 			    (arg + bsize - 1) / bsize);
800 			atomic_set_int(&fp->f_flag, FRDAHEAD);
801 		} else {
802 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
803 		}
804 		VOP_UNLOCK(vp);
805 		fdrop(fp, td);
806 		break;
807 
808 	case F_ISUNIONSTACK:
809 		/*
810 		 * Check if the vnode is part of a union stack (either the
811 		 * "union" flag from mount(2) or unionfs).
812 		 *
813 		 * Prior to introduction of this op libc's readdir would call
814 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
815 		 * data just to check fs name and a mount flag.
816 		 *
817 		 * Fixing the code to handle everything in the kernel instead
818 		 * is a non-trivial endeavor and has low priority, thus this
819 		 * horrible kludge facilitates the current behavior in a much
820 		 * cheaper manner until someone(tm) sorts this out.
821 		 */
822 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
823 		if (error != 0)
824 			break;
825 		if (fp->f_type != DTYPE_VNODE) {
826 			fdrop(fp, td);
827 			error = EBADF;
828 			break;
829 		}
830 		vp = fp->f_vnode;
831 		/*
832 		 * Since we don't prevent dooming the vnode even non-null mp
833 		 * found can become immediately stale. This is tolerable since
834 		 * mount points are type-stable (providing safe memory access)
835 		 * and any vfs op on this vnode going forward will return an
836 		 * error (meaning return value in this case is meaningless).
837 		 */
838 		mp = atomic_load_ptr(&vp->v_mount);
839 		if (__predict_false(mp == NULL)) {
840 			fdrop(fp, td);
841 			error = EBADF;
842 			break;
843 		}
844 		td->td_retval[0] = 0;
845 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
846 		    mp->mnt_flag & MNT_UNION)
847 			td->td_retval[0] = 1;
848 		fdrop(fp, td);
849 		break;
850 
851 	default:
852 		error = EINVAL;
853 		break;
854 	}
855 	return (error);
856 }
857 
858 static int
859 getmaxfd(struct thread *td)
860 {
861 
862 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
863 }
864 
865 /*
866  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
867  */
868 int
869 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
870 {
871 	struct filedesc *fdp;
872 	struct filedescent *oldfde, *newfde;
873 	struct proc *p;
874 	struct file *delfp, *oldfp;
875 	u_long *oioctls, *nioctls;
876 	int error, maxfd;
877 
878 	p = td->td_proc;
879 	fdp = p->p_fd;
880 	oioctls = NULL;
881 
882 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
883 	MPASS(mode < FDDUP_LASTMODE);
884 
885 	AUDIT_ARG_FD(old);
886 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
887 
888 	/*
889 	 * Verify we have a valid descriptor to dup from and possibly to
890 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
891 	 * return EINVAL when the new descriptor is out of bounds.
892 	 */
893 	if (old < 0)
894 		return (EBADF);
895 	if (new < 0)
896 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
897 	maxfd = getmaxfd(td);
898 	if (new >= maxfd)
899 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
900 
901 	error = EBADF;
902 	FILEDESC_XLOCK(fdp);
903 	if (fget_locked(fdp, old) == NULL)
904 		goto unlock;
905 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
906 		td->td_retval[0] = new;
907 		if (flags & FDDUP_FLAG_CLOEXEC)
908 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
909 		error = 0;
910 		goto unlock;
911 	}
912 
913 	oldfde = &fdp->fd_ofiles[old];
914 	oldfp = oldfde->fde_file;
915 	if (!fhold(oldfp))
916 		goto unlock;
917 
918 	/*
919 	 * If the caller specified a file descriptor, make sure the file
920 	 * table is large enough to hold it, and grab it.  Otherwise, just
921 	 * allocate a new descriptor the usual way.
922 	 */
923 	switch (mode) {
924 	case FDDUP_NORMAL:
925 	case FDDUP_FCNTL:
926 		if ((error = fdalloc(td, new, &new)) != 0) {
927 			fdrop(oldfp, td);
928 			goto unlock;
929 		}
930 		break;
931 	case FDDUP_MUSTREPLACE:
932 		/* Target file descriptor must exist. */
933 		if (fget_locked(fdp, new) == NULL) {
934 			fdrop(oldfp, td);
935 			goto unlock;
936 		}
937 		break;
938 	case FDDUP_FIXED:
939 		if (new >= fdp->fd_nfiles) {
940 			/*
941 			 * The resource limits are here instead of e.g.
942 			 * fdalloc(), because the file descriptor table may be
943 			 * shared between processes, so we can't really use
944 			 * racct_add()/racct_sub().  Instead of counting the
945 			 * number of actually allocated descriptors, just put
946 			 * the limit on the size of the file descriptor table.
947 			 */
948 #ifdef RACCT
949 			if (RACCT_ENABLED()) {
950 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
951 				if (error != 0) {
952 					error = EMFILE;
953 					fdrop(oldfp, td);
954 					goto unlock;
955 				}
956 			}
957 #endif
958 			fdgrowtable_exp(fdp, new + 1);
959 		}
960 		if (!fdisused(fdp, new))
961 			fdused(fdp, new);
962 		break;
963 	default:
964 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
965 	}
966 
967 	KASSERT(old != new, ("new fd is same as old"));
968 
969 	/* Refetch oldfde because the table may have grown and old one freed. */
970 	oldfde = &fdp->fd_ofiles[old];
971 	KASSERT(oldfp == oldfde->fde_file,
972 	    ("fdt_ofiles shift from growth observed at fd %d",
973 	    old));
974 
975 	newfde = &fdp->fd_ofiles[new];
976 	delfp = newfde->fde_file;
977 
978 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
979 
980 	/*
981 	 * Duplicate the source descriptor.
982 	 */
983 #ifdef CAPABILITIES
984 	seqc_write_begin(&newfde->fde_seqc);
985 #endif
986 	oioctls = filecaps_free_prep(&newfde->fde_caps);
987 	memcpy(newfde, oldfde, fde_change_size);
988 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
989 	    nioctls);
990 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
991 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
992 	else
993 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
994 #ifdef CAPABILITIES
995 	seqc_write_end(&newfde->fde_seqc);
996 #endif
997 	td->td_retval[0] = new;
998 
999 	error = 0;
1000 
1001 	if (delfp != NULL) {
1002 		(void) closefp(fdp, new, delfp, td, true, false);
1003 		FILEDESC_UNLOCK_ASSERT(fdp);
1004 	} else {
1005 unlock:
1006 		FILEDESC_XUNLOCK(fdp);
1007 	}
1008 
1009 	filecaps_free_finish(oioctls);
1010 	return (error);
1011 }
1012 
1013 static void
1014 sigiofree(struct sigio *sigio)
1015 {
1016 	crfree(sigio->sio_ucred);
1017 	free(sigio, M_SIGIO);
1018 }
1019 
1020 static struct sigio *
1021 funsetown_locked(struct sigio *sigio)
1022 {
1023 	struct proc *p;
1024 	struct pgrp *pg;
1025 
1026 	SIGIO_ASSERT_LOCKED();
1027 
1028 	if (sigio == NULL)
1029 		return (NULL);
1030 	*(sigio->sio_myref) = NULL;
1031 	if (sigio->sio_pgid < 0) {
1032 		pg = sigio->sio_pgrp;
1033 		PGRP_LOCK(pg);
1034 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1035 		    sigio, sio_pgsigio);
1036 		PGRP_UNLOCK(pg);
1037 	} else {
1038 		p = sigio->sio_proc;
1039 		PROC_LOCK(p);
1040 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1041 		    sigio, sio_pgsigio);
1042 		PROC_UNLOCK(p);
1043 	}
1044 	return (sigio);
1045 }
1046 
1047 /*
1048  * If sigio is on the list associated with a process or process group,
1049  * disable signalling from the device, remove sigio from the list and
1050  * free sigio.
1051  */
1052 void
1053 funsetown(struct sigio **sigiop)
1054 {
1055 	struct sigio *sigio;
1056 
1057 	/* Racy check, consumers must provide synchronization. */
1058 	if (*sigiop == NULL)
1059 		return;
1060 
1061 	SIGIO_LOCK();
1062 	sigio = funsetown_locked(*sigiop);
1063 	SIGIO_UNLOCK();
1064 	if (sigio != NULL)
1065 		sigiofree(sigio);
1066 }
1067 
1068 /*
1069  * Free a list of sigio structures.  The caller must ensure that new sigio
1070  * structures cannot be added after this point.  For process groups this is
1071  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1072  * as an interlock.
1073  */
1074 void
1075 funsetownlst(struct sigiolst *sigiolst)
1076 {
1077 	struct proc *p;
1078 	struct pgrp *pg;
1079 	struct sigio *sigio, *tmp;
1080 
1081 	/* Racy check. */
1082 	sigio = SLIST_FIRST(sigiolst);
1083 	if (sigio == NULL)
1084 		return;
1085 
1086 	p = NULL;
1087 	pg = NULL;
1088 
1089 	SIGIO_LOCK();
1090 	sigio = SLIST_FIRST(sigiolst);
1091 	if (sigio == NULL) {
1092 		SIGIO_UNLOCK();
1093 		return;
1094 	}
1095 
1096 	/*
1097 	 * Every entry of the list should belong to a single proc or pgrp.
1098 	 */
1099 	if (sigio->sio_pgid < 0) {
1100 		pg = sigio->sio_pgrp;
1101 		sx_assert(&proctree_lock, SX_XLOCKED);
1102 		PGRP_LOCK(pg);
1103 	} else /* if (sigio->sio_pgid > 0) */ {
1104 		p = sigio->sio_proc;
1105 		PROC_LOCK(p);
1106 		KASSERT((p->p_flag & P_WEXIT) != 0,
1107 		    ("%s: process %p is not exiting", __func__, p));
1108 	}
1109 
1110 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1111 		*sigio->sio_myref = NULL;
1112 		if (pg != NULL) {
1113 			KASSERT(sigio->sio_pgid < 0,
1114 			    ("Proc sigio in pgrp sigio list"));
1115 			KASSERT(sigio->sio_pgrp == pg,
1116 			    ("Bogus pgrp in sigio list"));
1117 		} else /* if (p != NULL) */ {
1118 			KASSERT(sigio->sio_pgid > 0,
1119 			    ("Pgrp sigio in proc sigio list"));
1120 			KASSERT(sigio->sio_proc == p,
1121 			    ("Bogus proc in sigio list"));
1122 		}
1123 	}
1124 
1125 	if (pg != NULL)
1126 		PGRP_UNLOCK(pg);
1127 	else
1128 		PROC_UNLOCK(p);
1129 	SIGIO_UNLOCK();
1130 
1131 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1132 		sigiofree(sigio);
1133 }
1134 
1135 /*
1136  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1137  *
1138  * After permission checking, add a sigio structure to the sigio list for
1139  * the process or process group.
1140  */
1141 int
1142 fsetown(pid_t pgid, struct sigio **sigiop)
1143 {
1144 	struct proc *proc;
1145 	struct pgrp *pgrp;
1146 	struct sigio *osigio, *sigio;
1147 	int ret;
1148 
1149 	if (pgid == 0) {
1150 		funsetown(sigiop);
1151 		return (0);
1152 	}
1153 
1154 	ret = 0;
1155 
1156 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1157 	sigio->sio_pgid = pgid;
1158 	sigio->sio_ucred = crhold(curthread->td_ucred);
1159 	sigio->sio_myref = sigiop;
1160 
1161 	sx_slock(&proctree_lock);
1162 	SIGIO_LOCK();
1163 	osigio = funsetown_locked(*sigiop);
1164 	if (pgid > 0) {
1165 		proc = pfind(pgid);
1166 		if (proc == NULL) {
1167 			ret = ESRCH;
1168 			goto fail;
1169 		}
1170 
1171 		/*
1172 		 * Policy - Don't allow a process to FSETOWN a process
1173 		 * in another session.
1174 		 *
1175 		 * Remove this test to allow maximum flexibility or
1176 		 * restrict FSETOWN to the current process or process
1177 		 * group for maximum safety.
1178 		 */
1179 		if (proc->p_session != curthread->td_proc->p_session) {
1180 			PROC_UNLOCK(proc);
1181 			ret = EPERM;
1182 			goto fail;
1183 		}
1184 
1185 		sigio->sio_proc = proc;
1186 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1187 		PROC_UNLOCK(proc);
1188 	} else /* if (pgid < 0) */ {
1189 		pgrp = pgfind(-pgid);
1190 		if (pgrp == NULL) {
1191 			ret = ESRCH;
1192 			goto fail;
1193 		}
1194 
1195 		/*
1196 		 * Policy - Don't allow a process to FSETOWN a process
1197 		 * in another session.
1198 		 *
1199 		 * Remove this test to allow maximum flexibility or
1200 		 * restrict FSETOWN to the current process or process
1201 		 * group for maximum safety.
1202 		 */
1203 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1204 			PGRP_UNLOCK(pgrp);
1205 			ret = EPERM;
1206 			goto fail;
1207 		}
1208 
1209 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1210 		sigio->sio_pgrp = pgrp;
1211 		PGRP_UNLOCK(pgrp);
1212 	}
1213 	sx_sunlock(&proctree_lock);
1214 	*sigiop = sigio;
1215 	SIGIO_UNLOCK();
1216 	if (osigio != NULL)
1217 		sigiofree(osigio);
1218 	return (0);
1219 
1220 fail:
1221 	SIGIO_UNLOCK();
1222 	sx_sunlock(&proctree_lock);
1223 	sigiofree(sigio);
1224 	if (osigio != NULL)
1225 		sigiofree(osigio);
1226 	return (ret);
1227 }
1228 
1229 /*
1230  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1231  */
1232 pid_t
1233 fgetown(struct sigio **sigiop)
1234 {
1235 	pid_t pgid;
1236 
1237 	SIGIO_LOCK();
1238 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1239 	SIGIO_UNLOCK();
1240 	return (pgid);
1241 }
1242 
1243 static int
1244 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1245     bool audit)
1246 {
1247 	int error;
1248 
1249 	FILEDESC_XLOCK_ASSERT(fdp);
1250 
1251 	/*
1252 	 * We now hold the fp reference that used to be owned by the
1253 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1254 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1255 	 * added, and deleteing a knote for the new fd.
1256 	 */
1257 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1258 		knote_fdclose(td, fd);
1259 
1260 	/*
1261 	 * We need to notify mqueue if the object is of type mqueue.
1262 	 */
1263 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1264 		mq_fdclose(td, fd, fp);
1265 	FILEDESC_XUNLOCK(fdp);
1266 
1267 #ifdef AUDIT
1268 	if (AUDITING_TD(td) && audit)
1269 		audit_sysclose(td, fd, fp);
1270 #endif
1271 	error = closef(fp, td);
1272 
1273 	/*
1274 	 * All paths leading up to closefp() will have already removed or
1275 	 * replaced the fd in the filedesc table, so a restart would not
1276 	 * operate on the same file.
1277 	 */
1278 	if (error == ERESTART)
1279 		error = EINTR;
1280 
1281 	return (error);
1282 }
1283 
1284 static int
1285 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1286     bool holdleaders, bool audit)
1287 {
1288 	int error;
1289 
1290 	FILEDESC_XLOCK_ASSERT(fdp);
1291 
1292 	if (holdleaders) {
1293 		if (td->td_proc->p_fdtol != NULL) {
1294 			/*
1295 			 * Ask fdfree() to sleep to ensure that all relevant
1296 			 * process leaders can be traversed in closef().
1297 			 */
1298 			fdp->fd_holdleaderscount++;
1299 		} else {
1300 			holdleaders = false;
1301 		}
1302 	}
1303 
1304 	error = closefp_impl(fdp, fd, fp, td, audit);
1305 	if (holdleaders) {
1306 		FILEDESC_XLOCK(fdp);
1307 		fdp->fd_holdleaderscount--;
1308 		if (fdp->fd_holdleaderscount == 0 &&
1309 		    fdp->fd_holdleaderswakeup != 0) {
1310 			fdp->fd_holdleaderswakeup = 0;
1311 			wakeup(&fdp->fd_holdleaderscount);
1312 		}
1313 		FILEDESC_XUNLOCK(fdp);
1314 	}
1315 	return (error);
1316 }
1317 
1318 static int
1319 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1320     bool holdleaders, bool audit)
1321 {
1322 
1323 	FILEDESC_XLOCK_ASSERT(fdp);
1324 
1325 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1326 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1327 	} else {
1328 		return (closefp_impl(fdp, fd, fp, td, audit));
1329 	}
1330 }
1331 
1332 /*
1333  * Close a file descriptor.
1334  */
1335 #ifndef _SYS_SYSPROTO_H_
1336 struct close_args {
1337 	int     fd;
1338 };
1339 #endif
1340 /* ARGSUSED */
1341 int
1342 sys_close(struct thread *td, struct close_args *uap)
1343 {
1344 
1345 	return (kern_close(td, uap->fd));
1346 }
1347 
1348 int
1349 kern_close(struct thread *td, int fd)
1350 {
1351 	struct filedesc *fdp;
1352 	struct file *fp;
1353 
1354 	fdp = td->td_proc->p_fd;
1355 
1356 	FILEDESC_XLOCK(fdp);
1357 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1358 		FILEDESC_XUNLOCK(fdp);
1359 		return (EBADF);
1360 	}
1361 	fdfree(fdp, fd);
1362 
1363 	/* closefp() drops the FILEDESC lock for us. */
1364 	return (closefp(fdp, fd, fp, td, true, true));
1365 }
1366 
1367 int
1368 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1369 {
1370 	struct filedesc *fdp;
1371 	const struct fdescenttbl *fdt;
1372 	struct file *fp;
1373 	int fd;
1374 
1375 	/*
1376 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1377 	 * open should not be a usage error.  From a close_range() perspective,
1378 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1379 	 * be a usage error as all fd above 3 are in-fact already closed.
1380 	 */
1381 	if (highfd < lowfd) {
1382 		return (EINVAL);
1383 	}
1384 
1385 	fdp = td->td_proc->p_fd;
1386 	FILEDESC_XLOCK(fdp);
1387 	fdt = atomic_load_ptr(&fdp->fd_files);
1388 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1389 	fd = lowfd;
1390 	if (__predict_false(fd > highfd)) {
1391 		goto out_locked;
1392 	}
1393 	for (;;) {
1394 		fp = fdt->fdt_ofiles[fd].fde_file;
1395 		if (fp == NULL) {
1396 			if (fd == highfd)
1397 				goto out_locked;
1398 		} else {
1399 			fdfree(fdp, fd);
1400 			(void) closefp(fdp, fd, fp, td, true, true);
1401 			if (fd == highfd)
1402 				goto out_unlocked;
1403 			FILEDESC_XLOCK(fdp);
1404 			fdt = atomic_load_ptr(&fdp->fd_files);
1405 		}
1406 		fd++;
1407 	}
1408 out_locked:
1409 	FILEDESC_XUNLOCK(fdp);
1410 out_unlocked:
1411 	return (0);
1412 }
1413 
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct close_range_args {
1416 	u_int	lowfd;
1417 	u_int	highfd;
1418 	int	flags;
1419 };
1420 #endif
1421 int
1422 sys_close_range(struct thread *td, struct close_range_args *uap)
1423 {
1424 
1425 	/* No flags currently defined */
1426 	if (uap->flags != 0)
1427 		return (EINVAL);
1428 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1429 }
1430 
1431 #ifdef COMPAT_FREEBSD12
1432 /*
1433  * Close open file descriptors.
1434  */
1435 #ifndef _SYS_SYSPROTO_H_
1436 struct freebsd12_closefrom_args {
1437 	int	lowfd;
1438 };
1439 #endif
1440 /* ARGSUSED */
1441 int
1442 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1443 {
1444 	u_int lowfd;
1445 
1446 	AUDIT_ARG_FD(uap->lowfd);
1447 
1448 	/*
1449 	 * Treat negative starting file descriptor values identical to
1450 	 * closefrom(0) which closes all files.
1451 	 */
1452 	lowfd = MAX(0, uap->lowfd);
1453 	return (kern_close_range(td, lowfd, ~0U));
1454 }
1455 #endif	/* COMPAT_FREEBSD12 */
1456 
1457 #if defined(COMPAT_43)
1458 /*
1459  * Return status information about a file descriptor.
1460  */
1461 #ifndef _SYS_SYSPROTO_H_
1462 struct ofstat_args {
1463 	int	fd;
1464 	struct	ostat *sb;
1465 };
1466 #endif
1467 /* ARGSUSED */
1468 int
1469 ofstat(struct thread *td, struct ofstat_args *uap)
1470 {
1471 	struct ostat oub;
1472 	struct stat ub;
1473 	int error;
1474 
1475 	error = kern_fstat(td, uap->fd, &ub);
1476 	if (error == 0) {
1477 		cvtstat(&ub, &oub);
1478 		error = copyout(&oub, uap->sb, sizeof(oub));
1479 	}
1480 	return (error);
1481 }
1482 #endif /* COMPAT_43 */
1483 
1484 #if defined(COMPAT_FREEBSD11)
1485 int
1486 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1487 {
1488 	struct stat sb;
1489 	struct freebsd11_stat osb;
1490 	int error;
1491 
1492 	error = kern_fstat(td, uap->fd, &sb);
1493 	if (error != 0)
1494 		return (error);
1495 	error = freebsd11_cvtstat(&sb, &osb);
1496 	if (error == 0)
1497 		error = copyout(&osb, uap->sb, sizeof(osb));
1498 	return (error);
1499 }
1500 #endif	/* COMPAT_FREEBSD11 */
1501 
1502 /*
1503  * Return status information about a file descriptor.
1504  */
1505 #ifndef _SYS_SYSPROTO_H_
1506 struct fstat_args {
1507 	int	fd;
1508 	struct	stat *sb;
1509 };
1510 #endif
1511 /* ARGSUSED */
1512 int
1513 sys_fstat(struct thread *td, struct fstat_args *uap)
1514 {
1515 	struct stat ub;
1516 	int error;
1517 
1518 	error = kern_fstat(td, uap->fd, &ub);
1519 	if (error == 0)
1520 		error = copyout(&ub, uap->sb, sizeof(ub));
1521 	return (error);
1522 }
1523 
1524 int
1525 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1526 {
1527 	struct file *fp;
1528 	int error;
1529 
1530 	AUDIT_ARG_FD(fd);
1531 
1532 	error = fget(td, fd, &cap_fstat_rights, &fp);
1533 	if (__predict_false(error != 0))
1534 		return (error);
1535 
1536 	AUDIT_ARG_FILE(td->td_proc, fp);
1537 
1538 	error = fo_stat(fp, sbp, td->td_ucred, td);
1539 	fdrop(fp, td);
1540 #ifdef __STAT_TIME_T_EXT
1541 	sbp->st_atim_ext = 0;
1542 	sbp->st_mtim_ext = 0;
1543 	sbp->st_ctim_ext = 0;
1544 	sbp->st_btim_ext = 0;
1545 #endif
1546 #ifdef KTRACE
1547 	if (KTRPOINT(td, KTR_STRUCT))
1548 		ktrstat_error(sbp, error);
1549 #endif
1550 	return (error);
1551 }
1552 
1553 #if defined(COMPAT_FREEBSD11)
1554 /*
1555  * Return status information about a file descriptor.
1556  */
1557 #ifndef _SYS_SYSPROTO_H_
1558 struct freebsd11_nfstat_args {
1559 	int	fd;
1560 	struct	nstat *sb;
1561 };
1562 #endif
1563 /* ARGSUSED */
1564 int
1565 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1566 {
1567 	struct nstat nub;
1568 	struct stat ub;
1569 	int error;
1570 
1571 	error = kern_fstat(td, uap->fd, &ub);
1572 	if (error == 0) {
1573 		freebsd11_cvtnstat(&ub, &nub);
1574 		error = copyout(&nub, uap->sb, sizeof(nub));
1575 	}
1576 	return (error);
1577 }
1578 #endif /* COMPAT_FREEBSD11 */
1579 
1580 /*
1581  * Return pathconf information about a file descriptor.
1582  */
1583 #ifndef _SYS_SYSPROTO_H_
1584 struct fpathconf_args {
1585 	int	fd;
1586 	int	name;
1587 };
1588 #endif
1589 /* ARGSUSED */
1590 int
1591 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1592 {
1593 	long value;
1594 	int error;
1595 
1596 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1597 	if (error == 0)
1598 		td->td_retval[0] = value;
1599 	return (error);
1600 }
1601 
1602 int
1603 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1604 {
1605 	struct file *fp;
1606 	struct vnode *vp;
1607 	int error;
1608 
1609 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1610 	if (error != 0)
1611 		return (error);
1612 
1613 	if (name == _PC_ASYNC_IO) {
1614 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1615 		goto out;
1616 	}
1617 	vp = fp->f_vnode;
1618 	if (vp != NULL) {
1619 		vn_lock(vp, LK_SHARED | LK_RETRY);
1620 		error = VOP_PATHCONF(vp, name, valuep);
1621 		VOP_UNLOCK(vp);
1622 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1623 		if (name != _PC_PIPE_BUF) {
1624 			error = EINVAL;
1625 		} else {
1626 			*valuep = PIPE_BUF;
1627 			error = 0;
1628 		}
1629 	} else {
1630 		error = EOPNOTSUPP;
1631 	}
1632 out:
1633 	fdrop(fp, td);
1634 	return (error);
1635 }
1636 
1637 /*
1638  * Copy filecaps structure allocating memory for ioctls array if needed.
1639  *
1640  * The last parameter indicates whether the fdtable is locked. If it is not and
1641  * ioctls are encountered, copying fails and the caller must lock the table.
1642  *
1643  * Note that if the table was not locked, the caller has to check the relevant
1644  * sequence counter to determine whether the operation was successful.
1645  */
1646 bool
1647 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1648 {
1649 	size_t size;
1650 
1651 	if (src->fc_ioctls != NULL && !locked)
1652 		return (false);
1653 	memcpy(dst, src, sizeof(*src));
1654 	if (src->fc_ioctls == NULL)
1655 		return (true);
1656 
1657 	KASSERT(src->fc_nioctls > 0,
1658 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1659 
1660 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1661 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1662 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1663 	return (true);
1664 }
1665 
1666 static u_long *
1667 filecaps_copy_prep(const struct filecaps *src)
1668 {
1669 	u_long *ioctls;
1670 	size_t size;
1671 
1672 	if (__predict_true(src->fc_ioctls == NULL))
1673 		return (NULL);
1674 
1675 	KASSERT(src->fc_nioctls > 0,
1676 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1677 
1678 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1679 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1680 	return (ioctls);
1681 }
1682 
1683 static void
1684 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1685     u_long *ioctls)
1686 {
1687 	size_t size;
1688 
1689 	*dst = *src;
1690 	if (__predict_true(src->fc_ioctls == NULL)) {
1691 		MPASS(ioctls == NULL);
1692 		return;
1693 	}
1694 
1695 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1696 	dst->fc_ioctls = ioctls;
1697 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1698 }
1699 
1700 /*
1701  * Move filecaps structure to the new place and clear the old place.
1702  */
1703 void
1704 filecaps_move(struct filecaps *src, struct filecaps *dst)
1705 {
1706 
1707 	*dst = *src;
1708 	bzero(src, sizeof(*src));
1709 }
1710 
1711 /*
1712  * Fill the given filecaps structure with full rights.
1713  */
1714 static void
1715 filecaps_fill(struct filecaps *fcaps)
1716 {
1717 
1718 	CAP_ALL(&fcaps->fc_rights);
1719 	fcaps->fc_ioctls = NULL;
1720 	fcaps->fc_nioctls = -1;
1721 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1722 }
1723 
1724 /*
1725  * Free memory allocated within filecaps structure.
1726  */
1727 void
1728 filecaps_free(struct filecaps *fcaps)
1729 {
1730 
1731 	free(fcaps->fc_ioctls, M_FILECAPS);
1732 	bzero(fcaps, sizeof(*fcaps));
1733 }
1734 
1735 static u_long *
1736 filecaps_free_prep(struct filecaps *fcaps)
1737 {
1738 	u_long *ioctls;
1739 
1740 	ioctls = fcaps->fc_ioctls;
1741 	bzero(fcaps, sizeof(*fcaps));
1742 	return (ioctls);
1743 }
1744 
1745 static void
1746 filecaps_free_finish(u_long *ioctls)
1747 {
1748 
1749 	free(ioctls, M_FILECAPS);
1750 }
1751 
1752 /*
1753  * Validate the given filecaps structure.
1754  */
1755 static void
1756 filecaps_validate(const struct filecaps *fcaps, const char *func)
1757 {
1758 
1759 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1760 	    ("%s: invalid rights", func));
1761 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1762 	    ("%s: invalid fcntls", func));
1763 	KASSERT(fcaps->fc_fcntls == 0 ||
1764 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1765 	    ("%s: fcntls without CAP_FCNTL", func));
1766 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1767 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1768 	    ("%s: invalid ioctls", func));
1769 	KASSERT(fcaps->fc_nioctls == 0 ||
1770 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1771 	    ("%s: ioctls without CAP_IOCTL", func));
1772 }
1773 
1774 static void
1775 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1776 {
1777 	int nfd1;
1778 
1779 	FILEDESC_XLOCK_ASSERT(fdp);
1780 
1781 	nfd1 = fdp->fd_nfiles * 2;
1782 	if (nfd1 < nfd)
1783 		nfd1 = nfd;
1784 	fdgrowtable(fdp, nfd1);
1785 }
1786 
1787 /*
1788  * Grow the file table to accommodate (at least) nfd descriptors.
1789  */
1790 static void
1791 fdgrowtable(struct filedesc *fdp, int nfd)
1792 {
1793 	struct filedesc0 *fdp0;
1794 	struct freetable *ft;
1795 	struct fdescenttbl *ntable;
1796 	struct fdescenttbl *otable;
1797 	int nnfiles, onfiles;
1798 	NDSLOTTYPE *nmap, *omap;
1799 
1800 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1801 
1802 	/* save old values */
1803 	onfiles = fdp->fd_nfiles;
1804 	otable = fdp->fd_files;
1805 	omap = fdp->fd_map;
1806 
1807 	/* compute the size of the new table */
1808 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1809 	if (nnfiles <= onfiles)
1810 		/* the table is already large enough */
1811 		return;
1812 
1813 	/*
1814 	 * Allocate a new table.  We need enough space for the number of
1815 	 * entries, file entries themselves and the struct freetable we will use
1816 	 * when we decommission the table and place it on the freelist.
1817 	 * We place the struct freetable in the middle so we don't have
1818 	 * to worry about padding.
1819 	 */
1820 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1821 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1822 	    sizeof(struct freetable),
1823 	    M_FILEDESC, M_ZERO | M_WAITOK);
1824 	/* copy the old data */
1825 	ntable->fdt_nfiles = nnfiles;
1826 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1827 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1828 
1829 	/*
1830 	 * Allocate a new map only if the old is not large enough.  It will
1831 	 * grow at a slower rate than the table as it can map more
1832 	 * entries than the table can hold.
1833 	 */
1834 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1835 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1836 		    M_ZERO | M_WAITOK);
1837 		/* copy over the old data and update the pointer */
1838 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1839 		fdp->fd_map = nmap;
1840 	}
1841 
1842 	/*
1843 	 * Make sure that ntable is correctly initialized before we replace
1844 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1845 	 * data.
1846 	 */
1847 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1848 
1849 	/*
1850 	 * Free the old file table when not shared by other threads or processes.
1851 	 * The old file table is considered to be shared when either are true:
1852 	 * - The process has more than one thread.
1853 	 * - The file descriptor table has been shared via fdshare().
1854 	 *
1855 	 * When shared, the old file table will be placed on a freelist
1856 	 * which will be processed when the struct filedesc is released.
1857 	 *
1858 	 * Note that if onfiles == NDFILE, we're dealing with the original
1859 	 * static allocation contained within (struct filedesc0 *)fdp,
1860 	 * which must not be freed.
1861 	 */
1862 	if (onfiles > NDFILE) {
1863 		/*
1864 		 * Note we may be called here from fdinit while allocating a
1865 		 * table for a new process in which case ->p_fd points
1866 		 * elsewhere.
1867 		 */
1868 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1869 			free(otable, M_FILEDESC);
1870 		} else {
1871 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1872 			fdp0 = (struct filedesc0 *)fdp;
1873 			ft->ft_table = otable;
1874 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1875 		}
1876 	}
1877 	/*
1878 	 * The map does not have the same possibility of threads still
1879 	 * holding references to it.  So always free it as long as it
1880 	 * does not reference the original static allocation.
1881 	 */
1882 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1883 		free(omap, M_FILEDESC);
1884 }
1885 
1886 /*
1887  * Allocate a file descriptor for the process.
1888  */
1889 int
1890 fdalloc(struct thread *td, int minfd, int *result)
1891 {
1892 	struct proc *p = td->td_proc;
1893 	struct filedesc *fdp = p->p_fd;
1894 	int fd, maxfd, allocfd;
1895 #ifdef RACCT
1896 	int error;
1897 #endif
1898 
1899 	FILEDESC_XLOCK_ASSERT(fdp);
1900 
1901 	if (fdp->fd_freefile > minfd)
1902 		minfd = fdp->fd_freefile;
1903 
1904 	maxfd = getmaxfd(td);
1905 
1906 	/*
1907 	 * Search the bitmap for a free descriptor starting at minfd.
1908 	 * If none is found, grow the file table.
1909 	 */
1910 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1911 	if (__predict_false(fd >= maxfd))
1912 		return (EMFILE);
1913 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1914 		allocfd = min(fd * 2, maxfd);
1915 #ifdef RACCT
1916 		if (RACCT_ENABLED()) {
1917 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1918 			if (error != 0)
1919 				return (EMFILE);
1920 		}
1921 #endif
1922 		/*
1923 		 * fd is already equal to first free descriptor >= minfd, so
1924 		 * we only need to grow the table and we are done.
1925 		 */
1926 		fdgrowtable_exp(fdp, allocfd);
1927 	}
1928 
1929 	/*
1930 	 * Perform some sanity checks, then mark the file descriptor as
1931 	 * used and return it to the caller.
1932 	 */
1933 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1934 	    ("invalid descriptor %d", fd));
1935 	KASSERT(!fdisused(fdp, fd),
1936 	    ("fd_first_free() returned non-free descriptor"));
1937 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1938 	    ("file descriptor isn't free"));
1939 	fdused(fdp, fd);
1940 	*result = fd;
1941 	return (0);
1942 }
1943 
1944 /*
1945  * Allocate n file descriptors for the process.
1946  */
1947 int
1948 fdallocn(struct thread *td, int minfd, int *fds, int n)
1949 {
1950 	struct proc *p = td->td_proc;
1951 	struct filedesc *fdp = p->p_fd;
1952 	int i;
1953 
1954 	FILEDESC_XLOCK_ASSERT(fdp);
1955 
1956 	for (i = 0; i < n; i++)
1957 		if (fdalloc(td, 0, &fds[i]) != 0)
1958 			break;
1959 
1960 	if (i < n) {
1961 		for (i--; i >= 0; i--)
1962 			fdunused(fdp, fds[i]);
1963 		return (EMFILE);
1964 	}
1965 
1966 	return (0);
1967 }
1968 
1969 /*
1970  * Create a new open file structure and allocate a file descriptor for the
1971  * process that refers to it.  We add one reference to the file for the
1972  * descriptor table and one reference for resultfp. This is to prevent us
1973  * being preempted and the entry in the descriptor table closed after we
1974  * release the FILEDESC lock.
1975  */
1976 int
1977 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1978     struct filecaps *fcaps)
1979 {
1980 	struct file *fp;
1981 	int error, fd;
1982 
1983 	MPASS(resultfp != NULL);
1984 	MPASS(resultfd != NULL);
1985 
1986 	error = _falloc_noinstall(td, &fp, 2);
1987 	if (__predict_false(error != 0)) {
1988 		return (error);
1989 	}
1990 
1991 	error = finstall_refed(td, fp, &fd, flags, fcaps);
1992 	if (__predict_false(error != 0)) {
1993 		falloc_abort(td, fp);
1994 		return (error);
1995 	}
1996 
1997 	*resultfp = fp;
1998 	*resultfd = fd;
1999 
2000 	return (0);
2001 }
2002 
2003 /*
2004  * Create a new open file structure without allocating a file descriptor.
2005  */
2006 int
2007 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2008 {
2009 	struct file *fp;
2010 	int maxuserfiles = maxfiles - (maxfiles / 20);
2011 	int openfiles_new;
2012 	static struct timeval lastfail;
2013 	static int curfail;
2014 
2015 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2016 	MPASS(n > 0);
2017 
2018 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2019 	if ((openfiles_new >= maxuserfiles &&
2020 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2021 	    openfiles_new >= maxfiles) {
2022 		atomic_subtract_int(&openfiles, 1);
2023 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2024 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2025 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2026 		}
2027 		return (ENFILE);
2028 	}
2029 	fp = uma_zalloc(file_zone, M_WAITOK);
2030 	bzero(fp, sizeof(*fp));
2031 	refcount_init(&fp->f_count, n);
2032 	fp->f_cred = crhold(td->td_ucred);
2033 	fp->f_ops = &badfileops;
2034 	*resultfp = fp;
2035 	return (0);
2036 }
2037 
2038 void
2039 falloc_abort(struct thread *td, struct file *fp)
2040 {
2041 
2042 	/*
2043 	 * For assertion purposes.
2044 	 */
2045 	refcount_init(&fp->f_count, 0);
2046 	_fdrop(fp, td);
2047 }
2048 
2049 /*
2050  * Install a file in a file descriptor table.
2051  */
2052 void
2053 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2054     struct filecaps *fcaps)
2055 {
2056 	struct filedescent *fde;
2057 
2058 	MPASS(fp != NULL);
2059 	if (fcaps != NULL)
2060 		filecaps_validate(fcaps, __func__);
2061 	FILEDESC_XLOCK_ASSERT(fdp);
2062 
2063 	fde = &fdp->fd_ofiles[fd];
2064 #ifdef CAPABILITIES
2065 	seqc_write_begin(&fde->fde_seqc);
2066 #endif
2067 	fde->fde_file = fp;
2068 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2069 	if (fcaps != NULL)
2070 		filecaps_move(fcaps, &fde->fde_caps);
2071 	else
2072 		filecaps_fill(&fde->fde_caps);
2073 #ifdef CAPABILITIES
2074 	seqc_write_end(&fde->fde_seqc);
2075 #endif
2076 }
2077 
2078 int
2079 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2080     struct filecaps *fcaps)
2081 {
2082 	struct filedesc *fdp = td->td_proc->p_fd;
2083 	int error;
2084 
2085 	MPASS(fd != NULL);
2086 
2087 	FILEDESC_XLOCK(fdp);
2088 	error = fdalloc(td, 0, fd);
2089 	if (__predict_true(error == 0)) {
2090 		_finstall(fdp, fp, *fd, flags, fcaps);
2091 	}
2092 	FILEDESC_XUNLOCK(fdp);
2093 	return (error);
2094 }
2095 
2096 int
2097 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2098     struct filecaps *fcaps)
2099 {
2100 	int error;
2101 
2102 	MPASS(fd != NULL);
2103 
2104 	if (!fhold(fp))
2105 		return (EBADF);
2106 	error = finstall_refed(td, fp, fd, flags, fcaps);
2107 	if (__predict_false(error != 0)) {
2108 		fdrop(fp, td);
2109 	}
2110 	return (error);
2111 }
2112 
2113 /*
2114  * Build a new filedesc structure from another.
2115  *
2116  * If fdp is not NULL, return with it shared locked.
2117  */
2118 struct filedesc *
2119 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2120 {
2121 	struct filedesc0 *newfdp0;
2122 	struct filedesc *newfdp;
2123 
2124 	if (prepfiles)
2125 		MPASS(lastfile != NULL);
2126 	else
2127 		MPASS(lastfile == NULL);
2128 
2129 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2130 	newfdp = &newfdp0->fd_fd;
2131 
2132 	/* Create the file descriptor table. */
2133 	FILEDESC_LOCK_INIT(newfdp);
2134 	refcount_init(&newfdp->fd_refcnt, 1);
2135 	refcount_init(&newfdp->fd_holdcnt, 1);
2136 	newfdp->fd_map = newfdp0->fd_dmap;
2137 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2138 	newfdp->fd_files->fdt_nfiles = NDFILE;
2139 
2140 	if (fdp == NULL)
2141 		return (newfdp);
2142 
2143 	FILEDESC_SLOCK(fdp);
2144 	if (!prepfiles) {
2145 		FILEDESC_SUNLOCK(fdp);
2146 		return (newfdp);
2147 	}
2148 
2149 	for (;;) {
2150 		*lastfile = fdlastfile(fdp);
2151 		if (*lastfile < newfdp->fd_nfiles)
2152 			break;
2153 		FILEDESC_SUNLOCK(fdp);
2154 		fdgrowtable(newfdp, *lastfile + 1);
2155 		FILEDESC_SLOCK(fdp);
2156 	}
2157 
2158 	return (newfdp);
2159 }
2160 
2161 /*
2162  * Build a pwddesc structure from another.
2163  * Copy the current, root, and jail root vnode references.
2164  *
2165  * If pdp is not NULL, return with it shared locked.
2166  */
2167 struct pwddesc *
2168 pdinit(struct pwddesc *pdp, bool keeplock)
2169 {
2170 	struct pwddesc *newpdp;
2171 	struct pwd *newpwd;
2172 
2173 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2174 
2175 	PWDDESC_LOCK_INIT(newpdp);
2176 	refcount_init(&newpdp->pd_refcount, 1);
2177 	newpdp->pd_cmask = CMASK;
2178 
2179 	if (pdp == NULL) {
2180 		newpwd = pwd_alloc();
2181 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2182 		return (newpdp);
2183 	}
2184 
2185 	PWDDESC_XLOCK(pdp);
2186 	newpwd = pwd_hold_pwddesc(pdp);
2187 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2188 	if (!keeplock)
2189 		PWDDESC_XUNLOCK(pdp);
2190 	return (newpdp);
2191 }
2192 
2193 static struct filedesc *
2194 fdhold(struct proc *p)
2195 {
2196 	struct filedesc *fdp;
2197 
2198 	PROC_LOCK_ASSERT(p, MA_OWNED);
2199 	fdp = p->p_fd;
2200 	if (fdp != NULL)
2201 		refcount_acquire(&fdp->fd_holdcnt);
2202 	return (fdp);
2203 }
2204 
2205 static struct pwddesc *
2206 pdhold(struct proc *p)
2207 {
2208 	struct pwddesc *pdp;
2209 
2210 	PROC_LOCK_ASSERT(p, MA_OWNED);
2211 	pdp = p->p_pd;
2212 	if (pdp != NULL)
2213 		refcount_acquire(&pdp->pd_refcount);
2214 	return (pdp);
2215 }
2216 
2217 static void
2218 fddrop(struct filedesc *fdp)
2219 {
2220 
2221 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2222 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2223 			return;
2224 	}
2225 
2226 	FILEDESC_LOCK_DESTROY(fdp);
2227 	uma_zfree(filedesc0_zone, fdp);
2228 }
2229 
2230 static void
2231 pddrop(struct pwddesc *pdp)
2232 {
2233 	struct pwd *pwd;
2234 
2235 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2236 		return;
2237 
2238 	PWDDESC_XLOCK(pdp);
2239 	if (refcount_release(&pdp->pd_refcount) == 0) {
2240 		PWDDESC_XUNLOCK(pdp);
2241 		return;
2242 	}
2243 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2244 	pwd_set(pdp, NULL);
2245 	PWDDESC_XUNLOCK(pdp);
2246 	pwd_drop(pwd);
2247 
2248 	PWDDESC_LOCK_DESTROY(pdp);
2249 	free(pdp, M_PWDDESC);
2250 }
2251 
2252 /*
2253  * Share a filedesc structure.
2254  */
2255 struct filedesc *
2256 fdshare(struct filedesc *fdp)
2257 {
2258 
2259 	refcount_acquire(&fdp->fd_refcnt);
2260 	return (fdp);
2261 }
2262 
2263 /*
2264  * Share a pwddesc structure.
2265  */
2266 struct pwddesc *
2267 pdshare(struct pwddesc *pdp)
2268 {
2269 	refcount_acquire(&pdp->pd_refcount);
2270 	return (pdp);
2271 }
2272 
2273 /*
2274  * Unshare a filedesc structure, if necessary by making a copy
2275  */
2276 void
2277 fdunshare(struct thread *td)
2278 {
2279 	struct filedesc *tmp;
2280 	struct proc *p = td->td_proc;
2281 
2282 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2283 		return;
2284 
2285 	tmp = fdcopy(p->p_fd);
2286 	fdescfree(td);
2287 	p->p_fd = tmp;
2288 }
2289 
2290 /*
2291  * Unshare a pwddesc structure.
2292  */
2293 void
2294 pdunshare(struct thread *td)
2295 {
2296 	struct pwddesc *pdp;
2297 	struct proc *p;
2298 
2299 	p = td->td_proc;
2300 	/* Not shared. */
2301 	if (p->p_pd->pd_refcount == 1)
2302 		return;
2303 
2304 	pdp = pdcopy(p->p_pd);
2305 	pdescfree(td);
2306 	p->p_pd = pdp;
2307 }
2308 
2309 void
2310 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2311 {
2312 
2313 	fdescfree(td);
2314 	td->td_proc->p_fd = fdp;
2315 }
2316 
2317 /*
2318  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2319  * this is to ease callers, not catch errors.
2320  */
2321 struct filedesc *
2322 fdcopy(struct filedesc *fdp)
2323 {
2324 	struct filedesc *newfdp;
2325 	struct filedescent *nfde, *ofde;
2326 	int i, lastfile;
2327 
2328 	MPASS(fdp != NULL);
2329 
2330 	newfdp = fdinit(fdp, true, &lastfile);
2331 	/* copy all passable descriptors (i.e. not kqueue) */
2332 	newfdp->fd_freefile = -1;
2333 	for (i = 0; i <= lastfile; ++i) {
2334 		ofde = &fdp->fd_ofiles[i];
2335 		if (ofde->fde_file == NULL ||
2336 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2337 		    !fhold(ofde->fde_file)) {
2338 			if (newfdp->fd_freefile == -1)
2339 				newfdp->fd_freefile = i;
2340 			continue;
2341 		}
2342 		nfde = &newfdp->fd_ofiles[i];
2343 		*nfde = *ofde;
2344 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2345 		fdused_init(newfdp, i);
2346 	}
2347 	if (newfdp->fd_freefile == -1)
2348 		newfdp->fd_freefile = i;
2349 	FILEDESC_SUNLOCK(fdp);
2350 	return (newfdp);
2351 }
2352 
2353 /*
2354  * Copy a pwddesc structure.
2355  */
2356 struct pwddesc *
2357 pdcopy(struct pwddesc *pdp)
2358 {
2359 	struct pwddesc *newpdp;
2360 
2361 	MPASS(pdp != NULL);
2362 
2363 	newpdp = pdinit(pdp, true);
2364 	newpdp->pd_cmask = pdp->pd_cmask;
2365 	PWDDESC_XUNLOCK(pdp);
2366 	return (newpdp);
2367 }
2368 
2369 /*
2370  * Copies a filedesc structure, while remapping all file descriptors
2371  * stored inside using a translation table.
2372  *
2373  * File descriptors are copied over to the new file descriptor table,
2374  * regardless of whether the close-on-exec flag is set.
2375  */
2376 int
2377 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2378     struct filedesc **ret)
2379 {
2380 	struct filedesc *newfdp;
2381 	struct filedescent *nfde, *ofde;
2382 	int error, i, lastfile;
2383 
2384 	MPASS(fdp != NULL);
2385 
2386 	newfdp = fdinit(fdp, true, &lastfile);
2387 	if (nfds > lastfile + 1) {
2388 		/* New table cannot be larger than the old one. */
2389 		error = E2BIG;
2390 		goto bad;
2391 	}
2392 	/* Copy all passable descriptors (i.e. not kqueue). */
2393 	newfdp->fd_freefile = nfds;
2394 	for (i = 0; i < nfds; ++i) {
2395 		if (fds[i] < 0 || fds[i] > lastfile) {
2396 			/* File descriptor out of bounds. */
2397 			error = EBADF;
2398 			goto bad;
2399 		}
2400 		ofde = &fdp->fd_ofiles[fds[i]];
2401 		if (ofde->fde_file == NULL) {
2402 			/* Unused file descriptor. */
2403 			error = EBADF;
2404 			goto bad;
2405 		}
2406 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2407 			/* File descriptor cannot be passed. */
2408 			error = EINVAL;
2409 			goto bad;
2410 		}
2411 		if (!fhold(ofde->fde_file)) {
2412 			error = EBADF;
2413 			goto bad;
2414 		}
2415 		nfde = &newfdp->fd_ofiles[i];
2416 		*nfde = *ofde;
2417 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2418 		fdused_init(newfdp, i);
2419 	}
2420 	FILEDESC_SUNLOCK(fdp);
2421 	*ret = newfdp;
2422 	return (0);
2423 bad:
2424 	FILEDESC_SUNLOCK(fdp);
2425 	fdescfree_remapped(newfdp);
2426 	return (error);
2427 }
2428 
2429 /*
2430  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2431  * one of processes using it exits) and the table used to be shared.
2432  */
2433 static void
2434 fdclearlocks(struct thread *td)
2435 {
2436 	struct filedesc *fdp;
2437 	struct filedesc_to_leader *fdtol;
2438 	struct flock lf;
2439 	struct file *fp;
2440 	struct proc *p;
2441 	struct vnode *vp;
2442 	int i, lastfile;
2443 
2444 	p = td->td_proc;
2445 	fdp = p->p_fd;
2446 	fdtol = p->p_fdtol;
2447 	MPASS(fdtol != NULL);
2448 
2449 	FILEDESC_XLOCK(fdp);
2450 	KASSERT(fdtol->fdl_refcount > 0,
2451 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2452 	    fdtol->fdl_refcount));
2453 	if (fdtol->fdl_refcount == 1 &&
2454 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2455 		lastfile = fdlastfile(fdp);
2456 		for (i = 0; i <= lastfile; i++) {
2457 			fp = fdp->fd_ofiles[i].fde_file;
2458 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2459 			    !fhold(fp))
2460 				continue;
2461 			FILEDESC_XUNLOCK(fdp);
2462 			lf.l_whence = SEEK_SET;
2463 			lf.l_start = 0;
2464 			lf.l_len = 0;
2465 			lf.l_type = F_UNLCK;
2466 			vp = fp->f_vnode;
2467 			(void) VOP_ADVLOCK(vp,
2468 			    (caddr_t)p->p_leader, F_UNLCK,
2469 			    &lf, F_POSIX);
2470 			FILEDESC_XLOCK(fdp);
2471 			fdrop(fp, td);
2472 		}
2473 	}
2474 retry:
2475 	if (fdtol->fdl_refcount == 1) {
2476 		if (fdp->fd_holdleaderscount > 0 &&
2477 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2478 			/*
2479 			 * close() or kern_dup() has cleared a reference
2480 			 * in a shared file descriptor table.
2481 			 */
2482 			fdp->fd_holdleaderswakeup = 1;
2483 			sx_sleep(&fdp->fd_holdleaderscount,
2484 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2485 			goto retry;
2486 		}
2487 		if (fdtol->fdl_holdcount > 0) {
2488 			/*
2489 			 * Ensure that fdtol->fdl_leader remains
2490 			 * valid in closef().
2491 			 */
2492 			fdtol->fdl_wakeup = 1;
2493 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2494 			    "fdlhold", 0);
2495 			goto retry;
2496 		}
2497 	}
2498 	fdtol->fdl_refcount--;
2499 	if (fdtol->fdl_refcount == 0 &&
2500 	    fdtol->fdl_holdcount == 0) {
2501 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2502 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2503 	} else
2504 		fdtol = NULL;
2505 	p->p_fdtol = NULL;
2506 	FILEDESC_XUNLOCK(fdp);
2507 	if (fdtol != NULL)
2508 		free(fdtol, M_FILEDESC_TO_LEADER);
2509 }
2510 
2511 /*
2512  * Release a filedesc structure.
2513  */
2514 static void
2515 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2516 {
2517 	struct filedesc0 *fdp0;
2518 	struct freetable *ft, *tft;
2519 	struct filedescent *fde;
2520 	struct file *fp;
2521 	int i, lastfile;
2522 
2523 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2524 	    ("%s: fd table %p carries references", __func__, fdp));
2525 
2526 	/*
2527 	 * Serialize with threads iterating over the table, if any.
2528 	 */
2529 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2530 		FILEDESC_XLOCK(fdp);
2531 		FILEDESC_XUNLOCK(fdp);
2532 	}
2533 
2534 	lastfile = fdlastfile_single(fdp);
2535 	for (i = 0; i <= lastfile; i++) {
2536 		fde = &fdp->fd_ofiles[i];
2537 		fp = fde->fde_file;
2538 		if (fp != NULL) {
2539 			fdefree_last(fde);
2540 			if (needclose)
2541 				(void) closef(fp, td);
2542 			else
2543 				fdrop(fp, td);
2544 		}
2545 	}
2546 
2547 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2548 		free(fdp->fd_map, M_FILEDESC);
2549 	if (fdp->fd_nfiles > NDFILE)
2550 		free(fdp->fd_files, M_FILEDESC);
2551 
2552 	fdp0 = (struct filedesc0 *)fdp;
2553 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2554 		free(ft->ft_table, M_FILEDESC);
2555 
2556 	fddrop(fdp);
2557 }
2558 
2559 void
2560 fdescfree(struct thread *td)
2561 {
2562 	struct proc *p;
2563 	struct filedesc *fdp;
2564 
2565 	p = td->td_proc;
2566 	fdp = p->p_fd;
2567 	MPASS(fdp != NULL);
2568 
2569 #ifdef RACCT
2570 	if (RACCT_ENABLED())
2571 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2572 #endif
2573 
2574 	if (p->p_fdtol != NULL)
2575 		fdclearlocks(td);
2576 
2577 	PROC_LOCK(p);
2578 	p->p_fd = NULL;
2579 	PROC_UNLOCK(p);
2580 
2581 	if (refcount_release(&fdp->fd_refcnt) == 0)
2582 		return;
2583 
2584 	fdescfree_fds(td, fdp, 1);
2585 }
2586 
2587 void
2588 pdescfree(struct thread *td)
2589 {
2590 	struct proc *p;
2591 	struct pwddesc *pdp;
2592 
2593 	p = td->td_proc;
2594 	pdp = p->p_pd;
2595 	MPASS(pdp != NULL);
2596 
2597 	PROC_LOCK(p);
2598 	p->p_pd = NULL;
2599 	PROC_UNLOCK(p);
2600 
2601 	pddrop(pdp);
2602 }
2603 
2604 void
2605 fdescfree_remapped(struct filedesc *fdp)
2606 {
2607 #ifdef INVARIANTS
2608 	/* fdescfree_fds() asserts that fd_refcnt == 0. */
2609 	if (!refcount_release(&fdp->fd_refcnt))
2610 		panic("%s: fd table %p has extra references", __func__, fdp);
2611 #endif
2612 	fdescfree_fds(curthread, fdp, 0);
2613 }
2614 
2615 /*
2616  * For setugid programs, we don't want to people to use that setugidness
2617  * to generate error messages which write to a file which otherwise would
2618  * otherwise be off-limits to the process.  We check for filesystems where
2619  * the vnode can change out from under us after execve (like [lin]procfs).
2620  *
2621  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2622  * sufficient.  We also don't check for setugidness since we know we are.
2623  */
2624 static bool
2625 is_unsafe(struct file *fp)
2626 {
2627 	struct vnode *vp;
2628 
2629 	if (fp->f_type != DTYPE_VNODE)
2630 		return (false);
2631 
2632 	vp = fp->f_vnode;
2633 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2634 }
2635 
2636 /*
2637  * Make this setguid thing safe, if at all possible.
2638  */
2639 void
2640 fdsetugidsafety(struct thread *td)
2641 {
2642 	struct filedesc *fdp;
2643 	struct file *fp;
2644 	int i;
2645 
2646 	fdp = td->td_proc->p_fd;
2647 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2648 	    ("the fdtable should not be shared"));
2649 	MPASS(fdp->fd_nfiles >= 3);
2650 	for (i = 0; i <= 2; i++) {
2651 		fp = fdp->fd_ofiles[i].fde_file;
2652 		if (fp != NULL && is_unsafe(fp)) {
2653 			FILEDESC_XLOCK(fdp);
2654 			knote_fdclose(td, i);
2655 			/*
2656 			 * NULL-out descriptor prior to close to avoid
2657 			 * a race while close blocks.
2658 			 */
2659 			fdfree(fdp, i);
2660 			FILEDESC_XUNLOCK(fdp);
2661 			(void) closef(fp, td);
2662 		}
2663 	}
2664 }
2665 
2666 /*
2667  * If a specific file object occupies a specific file descriptor, close the
2668  * file descriptor entry and drop a reference on the file object.  This is a
2669  * convenience function to handle a subsequent error in a function that calls
2670  * falloc() that handles the race that another thread might have closed the
2671  * file descriptor out from under the thread creating the file object.
2672  */
2673 void
2674 fdclose(struct thread *td, struct file *fp, int idx)
2675 {
2676 	struct filedesc *fdp = td->td_proc->p_fd;
2677 
2678 	FILEDESC_XLOCK(fdp);
2679 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2680 		fdfree(fdp, idx);
2681 		FILEDESC_XUNLOCK(fdp);
2682 		fdrop(fp, td);
2683 	} else
2684 		FILEDESC_XUNLOCK(fdp);
2685 }
2686 
2687 /*
2688  * Close any files on exec?
2689  */
2690 void
2691 fdcloseexec(struct thread *td)
2692 {
2693 	struct filedesc *fdp;
2694 	struct filedescent *fde;
2695 	struct file *fp;
2696 	int i, lastfile;
2697 
2698 	fdp = td->td_proc->p_fd;
2699 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2700 	    ("the fdtable should not be shared"));
2701 	lastfile = fdlastfile_single(fdp);
2702 	for (i = 0; i <= lastfile; i++) {
2703 		fde = &fdp->fd_ofiles[i];
2704 		fp = fde->fde_file;
2705 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2706 		    (fde->fde_flags & UF_EXCLOSE))) {
2707 			FILEDESC_XLOCK(fdp);
2708 			fdfree(fdp, i);
2709 			(void) closefp(fdp, i, fp, td, false, false);
2710 			FILEDESC_UNLOCK_ASSERT(fdp);
2711 		}
2712 	}
2713 }
2714 
2715 /*
2716  * It is unsafe for set[ug]id processes to be started with file
2717  * descriptors 0..2 closed, as these descriptors are given implicit
2718  * significance in the Standard C library.  fdcheckstd() will create a
2719  * descriptor referencing /dev/null for each of stdin, stdout, and
2720  * stderr that is not already open.
2721  */
2722 int
2723 fdcheckstd(struct thread *td)
2724 {
2725 	struct filedesc *fdp;
2726 	register_t save;
2727 	int i, error, devnull;
2728 
2729 	fdp = td->td_proc->p_fd;
2730 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2731 	    ("the fdtable should not be shared"));
2732 	MPASS(fdp->fd_nfiles >= 3);
2733 	devnull = -1;
2734 	for (i = 0; i <= 2; i++) {
2735 		if (fdp->fd_ofiles[i].fde_file != NULL)
2736 			continue;
2737 
2738 		save = td->td_retval[0];
2739 		if (devnull != -1) {
2740 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2741 		} else {
2742 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2743 			    UIO_SYSSPACE, O_RDWR, 0);
2744 			if (error == 0) {
2745 				devnull = td->td_retval[0];
2746 				KASSERT(devnull == i, ("we didn't get our fd"));
2747 			}
2748 		}
2749 		td->td_retval[0] = save;
2750 		if (error != 0)
2751 			return (error);
2752 	}
2753 	return (0);
2754 }
2755 
2756 /*
2757  * Internal form of close.  Decrement reference count on file structure.
2758  * Note: td may be NULL when closing a file that was being passed in a
2759  * message.
2760  */
2761 int
2762 closef(struct file *fp, struct thread *td)
2763 {
2764 	struct vnode *vp;
2765 	struct flock lf;
2766 	struct filedesc_to_leader *fdtol;
2767 	struct filedesc *fdp;
2768 
2769 	MPASS(td != NULL);
2770 
2771 	/*
2772 	 * POSIX record locking dictates that any close releases ALL
2773 	 * locks owned by this process.  This is handled by setting
2774 	 * a flag in the unlock to free ONLY locks obeying POSIX
2775 	 * semantics, and not to free BSD-style file locks.
2776 	 * If the descriptor was in a message, POSIX-style locks
2777 	 * aren't passed with the descriptor, and the thread pointer
2778 	 * will be NULL.  Callers should be careful only to pass a
2779 	 * NULL thread pointer when there really is no owning
2780 	 * context that might have locks, or the locks will be
2781 	 * leaked.
2782 	 */
2783 	if (fp->f_type == DTYPE_VNODE) {
2784 		vp = fp->f_vnode;
2785 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2786 			lf.l_whence = SEEK_SET;
2787 			lf.l_start = 0;
2788 			lf.l_len = 0;
2789 			lf.l_type = F_UNLCK;
2790 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2791 			    F_UNLCK, &lf, F_POSIX);
2792 		}
2793 		fdtol = td->td_proc->p_fdtol;
2794 		if (fdtol != NULL) {
2795 			/*
2796 			 * Handle special case where file descriptor table is
2797 			 * shared between multiple process leaders.
2798 			 */
2799 			fdp = td->td_proc->p_fd;
2800 			FILEDESC_XLOCK(fdp);
2801 			for (fdtol = fdtol->fdl_next;
2802 			    fdtol != td->td_proc->p_fdtol;
2803 			    fdtol = fdtol->fdl_next) {
2804 				if ((fdtol->fdl_leader->p_flag &
2805 				    P_ADVLOCK) == 0)
2806 					continue;
2807 				fdtol->fdl_holdcount++;
2808 				FILEDESC_XUNLOCK(fdp);
2809 				lf.l_whence = SEEK_SET;
2810 				lf.l_start = 0;
2811 				lf.l_len = 0;
2812 				lf.l_type = F_UNLCK;
2813 				vp = fp->f_vnode;
2814 				(void) VOP_ADVLOCK(vp,
2815 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2816 				    F_POSIX);
2817 				FILEDESC_XLOCK(fdp);
2818 				fdtol->fdl_holdcount--;
2819 				if (fdtol->fdl_holdcount == 0 &&
2820 				    fdtol->fdl_wakeup != 0) {
2821 					fdtol->fdl_wakeup = 0;
2822 					wakeup(fdtol);
2823 				}
2824 			}
2825 			FILEDESC_XUNLOCK(fdp);
2826 		}
2827 	}
2828 	return (fdrop_close(fp, td));
2829 }
2830 
2831 /*
2832  * Hack for file descriptor passing code.
2833  */
2834 void
2835 closef_nothread(struct file *fp)
2836 {
2837 
2838 	fdrop(fp, NULL);
2839 }
2840 
2841 /*
2842  * Initialize the file pointer with the specified properties.
2843  *
2844  * The ops are set with release semantics to be certain that the flags, type,
2845  * and data are visible when ops is.  This is to prevent ops methods from being
2846  * called with bad data.
2847  */
2848 void
2849 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2850 {
2851 	fp->f_data = data;
2852 	fp->f_flag = flag;
2853 	fp->f_type = type;
2854 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2855 }
2856 
2857 void
2858 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2859 {
2860 	fp->f_seqcount[UIO_READ] = 1;
2861 	fp->f_seqcount[UIO_WRITE] = 1;
2862 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2863 	    data, ops);
2864 }
2865 
2866 int
2867 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2868     struct file **fpp, struct filecaps *havecapsp)
2869 {
2870 	struct filedescent *fde;
2871 	int error;
2872 
2873 	FILEDESC_LOCK_ASSERT(fdp);
2874 
2875 	fde = fdeget_locked(fdp, fd);
2876 	if (fde == NULL) {
2877 		error = EBADF;
2878 		goto out;
2879 	}
2880 
2881 #ifdef CAPABILITIES
2882 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2883 	if (error != 0)
2884 		goto out;
2885 #endif
2886 
2887 	if (havecapsp != NULL)
2888 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2889 
2890 	*fpp = fde->fde_file;
2891 
2892 	error = 0;
2893 out:
2894 	return (error);
2895 }
2896 
2897 int
2898 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2899     struct file **fpp, struct filecaps *havecapsp)
2900 {
2901 	struct filedesc *fdp = td->td_proc->p_fd;
2902 	int error;
2903 #ifndef CAPABILITIES
2904 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2905 	if (havecapsp != NULL && error == 0)
2906 		filecaps_fill(havecapsp);
2907 #else
2908 	struct file *fp;
2909 	seqc_t seq;
2910 
2911 	*fpp = NULL;
2912 	for (;;) {
2913 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2914 		if (error != 0)
2915 			return (error);
2916 
2917 		if (havecapsp != NULL) {
2918 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2919 			    havecapsp, false)) {
2920 				fdrop(fp, td);
2921 				goto get_locked;
2922 			}
2923 		}
2924 
2925 		if (!fd_modified(fdp, fd, seq))
2926 			break;
2927 		fdrop(fp, td);
2928 	}
2929 
2930 	*fpp = fp;
2931 	return (0);
2932 
2933 get_locked:
2934 	FILEDESC_SLOCK(fdp);
2935 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2936 	if (error == 0 && !fhold(*fpp))
2937 		error = EBADF;
2938 	FILEDESC_SUNLOCK(fdp);
2939 #endif
2940 	return (error);
2941 }
2942 
2943 #ifdef CAPABILITIES
2944 int
2945 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2946 {
2947 	const struct filedescent *fde;
2948 	const struct fdescenttbl *fdt;
2949 	struct filedesc *fdp;
2950 	struct file *fp;
2951 	struct vnode *vp;
2952 	const cap_rights_t *haverights;
2953 	cap_rights_t rights;
2954 	seqc_t seq;
2955 
2956 	VFS_SMR_ASSERT_ENTERED();
2957 
2958 	rights = *ndp->ni_rightsneeded;
2959 	cap_rights_set_one(&rights, CAP_LOOKUP);
2960 
2961 	fdp = curproc->p_fd;
2962 	fdt = fdp->fd_files;
2963 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2964 		return (EBADF);
2965 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2966 	fde = &fdt->fdt_ofiles[fd];
2967 	haverights = cap_rights_fde_inline(fde);
2968 	fp = fde->fde_file;
2969 	if (__predict_false(fp == NULL))
2970 		return (EAGAIN);
2971 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2972 		return (EAGAIN);
2973 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2974 	vp = fp->f_vnode;
2975 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2976 		return (EAGAIN);
2977 	}
2978 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2979 		return (EAGAIN);
2980 	}
2981 	/*
2982 	 * Use an acquire barrier to force re-reading of fdt so it is
2983 	 * refreshed for verification.
2984 	 */
2985 	atomic_thread_fence_acq();
2986 	fdt = fdp->fd_files;
2987 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2988 		return (EAGAIN);
2989 	/*
2990 	 * If file descriptor doesn't have all rights,
2991 	 * all lookups relative to it must also be
2992 	 * strictly relative.
2993 	 *
2994 	 * Not yet supported by fast path.
2995 	 */
2996 	CAP_ALL(&rights);
2997 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2998 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2999 	    ndp->ni_filecaps.fc_nioctls != -1) {
3000 #ifdef notyet
3001 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3002 #else
3003 		return (EAGAIN);
3004 #endif
3005 	}
3006 	*vpp = vp;
3007 	return (0);
3008 }
3009 #else
3010 int
3011 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3012 {
3013 	const struct fdescenttbl *fdt;
3014 	struct filedesc *fdp;
3015 	struct file *fp;
3016 	struct vnode *vp;
3017 
3018 	VFS_SMR_ASSERT_ENTERED();
3019 
3020 	fdp = curproc->p_fd;
3021 	fdt = fdp->fd_files;
3022 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3023 		return (EBADF);
3024 	fp = fdt->fdt_ofiles[fd].fde_file;
3025 	if (__predict_false(fp == NULL))
3026 		return (EAGAIN);
3027 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
3028 	vp = fp->f_vnode;
3029 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3030 		return (EAGAIN);
3031 	}
3032 	/*
3033 	 * Use an acquire barrier to force re-reading of fdt so it is
3034 	 * refreshed for verification.
3035 	 */
3036 	atomic_thread_fence_acq();
3037 	fdt = fdp->fd_files;
3038 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3039 		return (EAGAIN);
3040 	filecaps_fill(&ndp->ni_filecaps);
3041 	*vpp = vp;
3042 	return (0);
3043 }
3044 #endif
3045 
3046 int
3047 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3048     struct file **fpp, seqc_t *seqp)
3049 {
3050 #ifdef CAPABILITIES
3051 	const struct filedescent *fde;
3052 #endif
3053 	const struct fdescenttbl *fdt;
3054 	struct file *fp;
3055 #ifdef CAPABILITIES
3056 	seqc_t seq;
3057 	cap_rights_t haverights;
3058 	int error;
3059 #endif
3060 
3061 	fdt = fdp->fd_files;
3062 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3063 		return (EBADF);
3064 	/*
3065 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
3066 	 * never raising a refcount above 0.  To accomplish this we have
3067 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3068 	 * must be re-verified once we acquire a reference to be certain
3069 	 * that the identity is still correct and we did not lose a race
3070 	 * due to preemption.
3071 	 */
3072 	for (;;) {
3073 #ifdef CAPABILITIES
3074 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3075 		fde = &fdt->fdt_ofiles[fd];
3076 		haverights = *cap_rights_fde_inline(fde);
3077 		fp = fde->fde_file;
3078 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3079 			continue;
3080 #else
3081 		fp = fdt->fdt_ofiles[fd].fde_file;
3082 #endif
3083 		if (fp == NULL)
3084 			return (EBADF);
3085 #ifdef CAPABILITIES
3086 		error = cap_check_inline(&haverights, needrightsp);
3087 		if (error != 0)
3088 			return (error);
3089 #endif
3090 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3091 			/*
3092 			 * Force a reload. Other thread could reallocate the
3093 			 * table before this fd was closed, so it is possible
3094 			 * that there is a stale fp pointer in cached version.
3095 			 */
3096 			fdt = atomic_load_ptr(&fdp->fd_files);
3097 			continue;
3098 		}
3099 		/*
3100 		 * Use an acquire barrier to force re-reading of fdt so it is
3101 		 * refreshed for verification.
3102 		 */
3103 		atomic_thread_fence_acq();
3104 		fdt = fdp->fd_files;
3105 #ifdef	CAPABILITIES
3106 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3107 #else
3108 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3109 #endif
3110 			break;
3111 		fdrop(fp, curthread);
3112 	}
3113 	*fpp = fp;
3114 	if (seqp != NULL) {
3115 #ifdef CAPABILITIES
3116 		*seqp = seq;
3117 #endif
3118 	}
3119 	return (0);
3120 }
3121 
3122 /*
3123  * See the comments in fget_unlocked_seq for an explanation of how this works.
3124  *
3125  * This is a simplified variant which bails out to the aforementioned routine
3126  * if anything goes wrong. In practice this only happens when userspace is
3127  * racing with itself.
3128  */
3129 int
3130 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3131     struct file **fpp)
3132 {
3133 #ifdef CAPABILITIES
3134 	const struct filedescent *fde;
3135 #endif
3136 	const struct fdescenttbl *fdt;
3137 	struct file *fp;
3138 #ifdef CAPABILITIES
3139 	seqc_t seq;
3140 	const cap_rights_t *haverights;
3141 #endif
3142 
3143 	fdt = fdp->fd_files;
3144 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3145 		return (EBADF);
3146 #ifdef CAPABILITIES
3147 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3148 	fde = &fdt->fdt_ofiles[fd];
3149 	haverights = cap_rights_fde_inline(fde);
3150 	fp = fde->fde_file;
3151 #else
3152 	fp = fdt->fdt_ofiles[fd].fde_file;
3153 #endif
3154 	if (__predict_false(fp == NULL))
3155 		goto out_fallback;
3156 #ifdef CAPABILITIES
3157 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3158 		goto out_fallback;
3159 #endif
3160 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3161 		goto out_fallback;
3162 
3163 	/*
3164 	 * Use an acquire barrier to force re-reading of fdt so it is
3165 	 * refreshed for verification.
3166 	 */
3167 	atomic_thread_fence_acq();
3168 	fdt = fdp->fd_files;
3169 #ifdef	CAPABILITIES
3170 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3171 #else
3172 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3173 #endif
3174 		goto out_fdrop;
3175 	*fpp = fp;
3176 	return (0);
3177 out_fdrop:
3178 	fdrop(fp, curthread);
3179 out_fallback:
3180 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3181 }
3182 
3183 /*
3184  * Translate fd -> file when the caller guarantees the file descriptor table
3185  * can't be changed by others.
3186  *
3187  * Note this does not mean the file object itself is only visible to the caller,
3188  * merely that it wont disappear without having to be referenced.
3189  *
3190  * Must be paired with fput_only_user.
3191  */
3192 #ifdef	CAPABILITIES
3193 int
3194 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3195     struct file **fpp)
3196 {
3197 	const struct filedescent *fde;
3198 	const struct fdescenttbl *fdt;
3199 	const cap_rights_t *haverights;
3200 	struct file *fp;
3201 	int error;
3202 
3203 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3204 
3205 	if (__predict_false(fd >= fdp->fd_nfiles))
3206 		return (EBADF);
3207 
3208 	fdt = fdp->fd_files;
3209 	fde = &fdt->fdt_ofiles[fd];
3210 	fp = fde->fde_file;
3211 	if (__predict_false(fp == NULL))
3212 		return (EBADF);
3213 	MPASS(refcount_load(&fp->f_count) > 0);
3214 	haverights = cap_rights_fde_inline(fde);
3215 	error = cap_check_inline(haverights, needrightsp);
3216 	if (__predict_false(error != 0))
3217 		return (error);
3218 	*fpp = fp;
3219 	return (0);
3220 }
3221 #else
3222 int
3223 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3224     struct file **fpp)
3225 {
3226 	struct file *fp;
3227 
3228 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3229 
3230 	if (__predict_false(fd >= fdp->fd_nfiles))
3231 		return (EBADF);
3232 
3233 	fp = fdp->fd_ofiles[fd].fde_file;
3234 	if (__predict_false(fp == NULL))
3235 		return (EBADF);
3236 
3237 	MPASS(refcount_load(&fp->f_count) > 0);
3238 	*fpp = fp;
3239 	return (0);
3240 }
3241 #endif
3242 
3243 /*
3244  * Extract the file pointer associated with the specified descriptor for the
3245  * current user process.
3246  *
3247  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3248  * returned.
3249  *
3250  * File's rights will be checked against the capability rights mask.
3251  *
3252  * If an error occurred the non-zero error is returned and *fpp is set to
3253  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3254  * responsible for fdrop().
3255  */
3256 static __inline int
3257 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3258     cap_rights_t *needrightsp)
3259 {
3260 	struct filedesc *fdp;
3261 	struct file *fp;
3262 	int error;
3263 
3264 	*fpp = NULL;
3265 	fdp = td->td_proc->p_fd;
3266 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3267 	if (__predict_false(error != 0))
3268 		return (error);
3269 	if (__predict_false(fp->f_ops == &badfileops)) {
3270 		fdrop(fp, td);
3271 		return (EBADF);
3272 	}
3273 
3274 	/*
3275 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3276 	 */
3277 	error = 0;
3278 	switch (flags) {
3279 	case FREAD:
3280 	case FWRITE:
3281 		if ((fp->f_flag & flags) == 0)
3282 			error = EBADF;
3283 		break;
3284 	case FEXEC:
3285 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3286 		    ((fp->f_flag & FWRITE) != 0))
3287 			error = EBADF;
3288 		break;
3289 	case 0:
3290 		break;
3291 	default:
3292 		KASSERT(0, ("wrong flags"));
3293 	}
3294 
3295 	if (error != 0) {
3296 		fdrop(fp, td);
3297 		return (error);
3298 	}
3299 
3300 	*fpp = fp;
3301 	return (0);
3302 }
3303 
3304 int
3305 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3306 {
3307 
3308 	return (_fget(td, fd, fpp, 0, rightsp));
3309 }
3310 
3311 int
3312 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3313     struct file **fpp)
3314 {
3315 	int error;
3316 #ifndef CAPABILITIES
3317 	error = _fget(td, fd, fpp, 0, rightsp);
3318 	if (maxprotp != NULL)
3319 		*maxprotp = VM_PROT_ALL;
3320 	return (error);
3321 #else
3322 	cap_rights_t fdrights;
3323 	struct filedesc *fdp;
3324 	struct file *fp;
3325 	seqc_t seq;
3326 
3327 	*fpp = NULL;
3328 	fdp = td->td_proc->p_fd;
3329 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3330 	for (;;) {
3331 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3332 		if (__predict_false(error != 0))
3333 			return (error);
3334 		if (__predict_false(fp->f_ops == &badfileops)) {
3335 			fdrop(fp, td);
3336 			return (EBADF);
3337 		}
3338 		if (maxprotp != NULL)
3339 			fdrights = *cap_rights(fdp, fd);
3340 		if (!fd_modified(fdp, fd, seq))
3341 			break;
3342 		fdrop(fp, td);
3343 	}
3344 
3345 	/*
3346 	 * If requested, convert capability rights to access flags.
3347 	 */
3348 	if (maxprotp != NULL)
3349 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3350 	*fpp = fp;
3351 	return (0);
3352 #endif
3353 }
3354 
3355 int
3356 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3357 {
3358 
3359 	return (_fget(td, fd, fpp, FREAD, rightsp));
3360 }
3361 
3362 int
3363 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3364 {
3365 
3366 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3367 }
3368 
3369 int
3370 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3371     struct file **fpp)
3372 {
3373 	struct filedesc *fdp = td->td_proc->p_fd;
3374 #ifndef CAPABILITIES
3375 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3376 #else
3377 	struct file *fp;
3378 	int error;
3379 	seqc_t seq;
3380 
3381 	*fpp = NULL;
3382 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3383 	for (;;) {
3384 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3385 		if (error != 0)
3386 			return (error);
3387 		error = cap_fcntl_check(fdp, fd, needfcntl);
3388 		if (!fd_modified(fdp, fd, seq))
3389 			break;
3390 		fdrop(fp, td);
3391 	}
3392 	if (error != 0) {
3393 		fdrop(fp, td);
3394 		return (error);
3395 	}
3396 	*fpp = fp;
3397 	return (0);
3398 #endif
3399 }
3400 
3401 /*
3402  * Like fget() but loads the underlying vnode, or returns an error if the
3403  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3404  * never have VM objects.  The returned vnode will be vref()'d.
3405  *
3406  * XXX: what about the unused flags ?
3407  */
3408 static __inline int
3409 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3410     struct vnode **vpp)
3411 {
3412 	struct file *fp;
3413 	int error;
3414 
3415 	*vpp = NULL;
3416 	error = _fget(td, fd, &fp, flags, needrightsp);
3417 	if (error != 0)
3418 		return (error);
3419 	if (fp->f_vnode == NULL) {
3420 		error = EINVAL;
3421 	} else {
3422 		*vpp = fp->f_vnode;
3423 		vrefact(*vpp);
3424 	}
3425 	fdrop(fp, td);
3426 
3427 	return (error);
3428 }
3429 
3430 int
3431 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3432 {
3433 
3434 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3435 }
3436 
3437 int
3438 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3439     struct filecaps *havecaps, struct vnode **vpp)
3440 {
3441 	struct filecaps caps;
3442 	struct file *fp;
3443 	int error;
3444 
3445 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3446 	if (error != 0)
3447 		return (error);
3448 	if (fp->f_ops == &badfileops) {
3449 		error = EBADF;
3450 		goto out;
3451 	}
3452 	if (fp->f_vnode == NULL) {
3453 		error = EINVAL;
3454 		goto out;
3455 	}
3456 
3457 	*havecaps = caps;
3458 	*vpp = fp->f_vnode;
3459 	vrefact(*vpp);
3460 	fdrop(fp, td);
3461 
3462 	return (0);
3463 out:
3464 	filecaps_free(&caps);
3465 	fdrop(fp, td);
3466 	return (error);
3467 }
3468 
3469 int
3470 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3471 {
3472 
3473 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3474 }
3475 
3476 int
3477 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3478 {
3479 
3480 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3481 }
3482 
3483 #ifdef notyet
3484 int
3485 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3486     struct vnode **vpp)
3487 {
3488 
3489 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3490 }
3491 #endif
3492 
3493 /*
3494  * Handle the last reference to a file being closed.
3495  *
3496  * Without the noinline attribute clang keeps inlining the func thorough this
3497  * file when fdrop is used.
3498  */
3499 int __noinline
3500 _fdrop(struct file *fp, struct thread *td)
3501 {
3502 	int error;
3503 #ifdef INVARIANTS
3504 	int count;
3505 
3506 	count = refcount_load(&fp->f_count);
3507 	if (count != 0)
3508 		panic("fdrop: fp %p count %d", fp, count);
3509 #endif
3510 	error = fo_close(fp, td);
3511 	atomic_subtract_int(&openfiles, 1);
3512 	crfree(fp->f_cred);
3513 	free(fp->f_advice, M_FADVISE);
3514 	uma_zfree(file_zone, fp);
3515 
3516 	return (error);
3517 }
3518 
3519 /*
3520  * Apply an advisory lock on a file descriptor.
3521  *
3522  * Just attempt to get a record lock of the requested type on the entire file
3523  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3524  */
3525 #ifndef _SYS_SYSPROTO_H_
3526 struct flock_args {
3527 	int	fd;
3528 	int	how;
3529 };
3530 #endif
3531 /* ARGSUSED */
3532 int
3533 sys_flock(struct thread *td, struct flock_args *uap)
3534 {
3535 	struct file *fp;
3536 	struct vnode *vp;
3537 	struct flock lf;
3538 	int error;
3539 
3540 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3541 	if (error != 0)
3542 		return (error);
3543 	if (fp->f_type != DTYPE_VNODE) {
3544 		fdrop(fp, td);
3545 		return (EOPNOTSUPP);
3546 	}
3547 
3548 	vp = fp->f_vnode;
3549 	lf.l_whence = SEEK_SET;
3550 	lf.l_start = 0;
3551 	lf.l_len = 0;
3552 	if (uap->how & LOCK_UN) {
3553 		lf.l_type = F_UNLCK;
3554 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3555 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3556 		goto done2;
3557 	}
3558 	if (uap->how & LOCK_EX)
3559 		lf.l_type = F_WRLCK;
3560 	else if (uap->how & LOCK_SH)
3561 		lf.l_type = F_RDLCK;
3562 	else {
3563 		error = EBADF;
3564 		goto done2;
3565 	}
3566 	atomic_set_int(&fp->f_flag, FHASLOCK);
3567 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3568 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3569 done2:
3570 	fdrop(fp, td);
3571 	return (error);
3572 }
3573 /*
3574  * Duplicate the specified descriptor to a free descriptor.
3575  */
3576 int
3577 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3578     int openerror, int *indxp)
3579 {
3580 	struct filedescent *newfde, *oldfde;
3581 	struct file *fp;
3582 	u_long *ioctls;
3583 	int error, indx;
3584 
3585 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3586 	    ("unexpected error %d in %s", openerror, __func__));
3587 
3588 	/*
3589 	 * If the to-be-dup'd fd number is greater than the allowed number
3590 	 * of file descriptors, or the fd to be dup'd has already been
3591 	 * closed, then reject.
3592 	 */
3593 	FILEDESC_XLOCK(fdp);
3594 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3595 		FILEDESC_XUNLOCK(fdp);
3596 		return (EBADF);
3597 	}
3598 
3599 	error = fdalloc(td, 0, &indx);
3600 	if (error != 0) {
3601 		FILEDESC_XUNLOCK(fdp);
3602 		return (error);
3603 	}
3604 
3605 	/*
3606 	 * There are two cases of interest here.
3607 	 *
3608 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3609 	 *
3610 	 * For ENXIO steal away the file structure from (dfd) and store it in
3611 	 * (indx).  (dfd) is effectively closed by this operation.
3612 	 */
3613 	switch (openerror) {
3614 	case ENODEV:
3615 		/*
3616 		 * Check that the mode the file is being opened for is a
3617 		 * subset of the mode of the existing descriptor.
3618 		 */
3619 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3620 			fdunused(fdp, indx);
3621 			FILEDESC_XUNLOCK(fdp);
3622 			return (EACCES);
3623 		}
3624 		if (!fhold(fp)) {
3625 			fdunused(fdp, indx);
3626 			FILEDESC_XUNLOCK(fdp);
3627 			return (EBADF);
3628 		}
3629 		newfde = &fdp->fd_ofiles[indx];
3630 		oldfde = &fdp->fd_ofiles[dfd];
3631 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3632 #ifdef CAPABILITIES
3633 		seqc_write_begin(&newfde->fde_seqc);
3634 #endif
3635 		memcpy(newfde, oldfde, fde_change_size);
3636 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3637 		    ioctls);
3638 #ifdef CAPABILITIES
3639 		seqc_write_end(&newfde->fde_seqc);
3640 #endif
3641 		break;
3642 	case ENXIO:
3643 		/*
3644 		 * Steal away the file pointer from dfd and stuff it into indx.
3645 		 */
3646 		newfde = &fdp->fd_ofiles[indx];
3647 		oldfde = &fdp->fd_ofiles[dfd];
3648 #ifdef CAPABILITIES
3649 		seqc_write_begin(&newfde->fde_seqc);
3650 #endif
3651 		memcpy(newfde, oldfde, fde_change_size);
3652 		oldfde->fde_file = NULL;
3653 		fdunused(fdp, dfd);
3654 #ifdef CAPABILITIES
3655 		seqc_write_end(&newfde->fde_seqc);
3656 #endif
3657 		break;
3658 	}
3659 	FILEDESC_XUNLOCK(fdp);
3660 	*indxp = indx;
3661 	return (0);
3662 }
3663 
3664 /*
3665  * This sysctl determines if we will allow a process to chroot(2) if it
3666  * has a directory open:
3667  *	0: disallowed for all processes.
3668  *	1: allowed for processes that were not already chroot(2)'ed.
3669  *	2: allowed for all processes.
3670  */
3671 
3672 static int chroot_allow_open_directories = 1;
3673 
3674 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3675     &chroot_allow_open_directories, 0,
3676     "Allow a process to chroot(2) if it has a directory open");
3677 
3678 /*
3679  * Helper function for raised chroot(2) security function:  Refuse if
3680  * any filedescriptors are open directories.
3681  */
3682 static int
3683 chroot_refuse_vdir_fds(struct filedesc *fdp)
3684 {
3685 	struct vnode *vp;
3686 	struct file *fp;
3687 	int fd, lastfile;
3688 
3689 	FILEDESC_LOCK_ASSERT(fdp);
3690 
3691 	lastfile = fdlastfile(fdp);
3692 	for (fd = 0; fd <= lastfile; fd++) {
3693 		fp = fget_locked(fdp, fd);
3694 		if (fp == NULL)
3695 			continue;
3696 		if (fp->f_type == DTYPE_VNODE) {
3697 			vp = fp->f_vnode;
3698 			if (vp->v_type == VDIR)
3699 				return (EPERM);
3700 		}
3701 	}
3702 	return (0);
3703 }
3704 
3705 static void
3706 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3707 {
3708 
3709 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3710 		vrefact(oldpwd->pwd_cdir);
3711 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3712 	}
3713 
3714 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3715 		vrefact(oldpwd->pwd_rdir);
3716 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3717 	}
3718 
3719 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3720 		vrefact(oldpwd->pwd_jdir);
3721 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3722 	}
3723 }
3724 
3725 struct pwd *
3726 pwd_hold_pwddesc(struct pwddesc *pdp)
3727 {
3728 	struct pwd *pwd;
3729 
3730 	PWDDESC_ASSERT_XLOCKED(pdp);
3731 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3732 	if (pwd != NULL)
3733 		refcount_acquire(&pwd->pwd_refcount);
3734 	return (pwd);
3735 }
3736 
3737 bool
3738 pwd_hold_smr(struct pwd *pwd)
3739 {
3740 
3741 	MPASS(pwd != NULL);
3742 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3743 		return (true);
3744 	}
3745 	return (false);
3746 }
3747 
3748 struct pwd *
3749 pwd_hold(struct thread *td)
3750 {
3751 	struct pwddesc *pdp;
3752 	struct pwd *pwd;
3753 
3754 	pdp = td->td_proc->p_pd;
3755 
3756 	vfs_smr_enter();
3757 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3758 	if (pwd_hold_smr(pwd)) {
3759 		vfs_smr_exit();
3760 		return (pwd);
3761 	}
3762 	vfs_smr_exit();
3763 	PWDDESC_XLOCK(pdp);
3764 	pwd = pwd_hold_pwddesc(pdp);
3765 	MPASS(pwd != NULL);
3766 	PWDDESC_XUNLOCK(pdp);
3767 	return (pwd);
3768 }
3769 
3770 static struct pwd *
3771 pwd_alloc(void)
3772 {
3773 	struct pwd *pwd;
3774 
3775 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3776 	bzero(pwd, sizeof(*pwd));
3777 	refcount_init(&pwd->pwd_refcount, 1);
3778 	return (pwd);
3779 }
3780 
3781 void
3782 pwd_drop(struct pwd *pwd)
3783 {
3784 
3785 	if (!refcount_release(&pwd->pwd_refcount))
3786 		return;
3787 
3788 	if (pwd->pwd_cdir != NULL)
3789 		vrele(pwd->pwd_cdir);
3790 	if (pwd->pwd_rdir != NULL)
3791 		vrele(pwd->pwd_rdir);
3792 	if (pwd->pwd_jdir != NULL)
3793 		vrele(pwd->pwd_jdir);
3794 	uma_zfree_smr(pwd_zone, pwd);
3795 }
3796 
3797 /*
3798 * The caller is responsible for invoking priv_check() and
3799 * mac_vnode_check_chroot() to authorize this operation.
3800 */
3801 int
3802 pwd_chroot(struct thread *td, struct vnode *vp)
3803 {
3804 	struct pwddesc *pdp;
3805 	struct filedesc *fdp;
3806 	struct pwd *newpwd, *oldpwd;
3807 	int error;
3808 
3809 	fdp = td->td_proc->p_fd;
3810 	pdp = td->td_proc->p_pd;
3811 	newpwd = pwd_alloc();
3812 	FILEDESC_SLOCK(fdp);
3813 	PWDDESC_XLOCK(pdp);
3814 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3815 	if (chroot_allow_open_directories == 0 ||
3816 	    (chroot_allow_open_directories == 1 &&
3817 	    oldpwd->pwd_rdir != rootvnode)) {
3818 		error = chroot_refuse_vdir_fds(fdp);
3819 		FILEDESC_SUNLOCK(fdp);
3820 		if (error != 0) {
3821 			PWDDESC_XUNLOCK(pdp);
3822 			pwd_drop(newpwd);
3823 			return (error);
3824 		}
3825 	} else {
3826 		FILEDESC_SUNLOCK(fdp);
3827 	}
3828 
3829 	vrefact(vp);
3830 	newpwd->pwd_rdir = vp;
3831 	if (oldpwd->pwd_jdir == NULL) {
3832 		vrefact(vp);
3833 		newpwd->pwd_jdir = vp;
3834 	}
3835 	pwd_fill(oldpwd, newpwd);
3836 	pwd_set(pdp, newpwd);
3837 	PWDDESC_XUNLOCK(pdp);
3838 	pwd_drop(oldpwd);
3839 	return (0);
3840 }
3841 
3842 void
3843 pwd_chdir(struct thread *td, struct vnode *vp)
3844 {
3845 	struct pwddesc *pdp;
3846 	struct pwd *newpwd, *oldpwd;
3847 
3848 	VNPASS(vp->v_usecount > 0, vp);
3849 
3850 	newpwd = pwd_alloc();
3851 	pdp = td->td_proc->p_pd;
3852 	PWDDESC_XLOCK(pdp);
3853 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3854 	newpwd->pwd_cdir = vp;
3855 	pwd_fill(oldpwd, newpwd);
3856 	pwd_set(pdp, newpwd);
3857 	PWDDESC_XUNLOCK(pdp);
3858 	pwd_drop(oldpwd);
3859 }
3860 
3861 /*
3862  * jail_attach(2) changes both root and working directories.
3863  */
3864 int
3865 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3866 {
3867 	struct pwddesc *pdp;
3868 	struct filedesc *fdp;
3869 	struct pwd *newpwd, *oldpwd;
3870 	int error;
3871 
3872 	fdp = td->td_proc->p_fd;
3873 	pdp = td->td_proc->p_pd;
3874 	newpwd = pwd_alloc();
3875 	FILEDESC_SLOCK(fdp);
3876 	PWDDESC_XLOCK(pdp);
3877 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3878 	error = chroot_refuse_vdir_fds(fdp);
3879 	FILEDESC_SUNLOCK(fdp);
3880 	if (error != 0) {
3881 		PWDDESC_XUNLOCK(pdp);
3882 		pwd_drop(newpwd);
3883 		return (error);
3884 	}
3885 
3886 	vrefact(vp);
3887 	newpwd->pwd_rdir = vp;
3888 	vrefact(vp);
3889 	newpwd->pwd_cdir = vp;
3890 	if (oldpwd->pwd_jdir == NULL) {
3891 		vrefact(vp);
3892 		newpwd->pwd_jdir = vp;
3893 	}
3894 	pwd_fill(oldpwd, newpwd);
3895 	pwd_set(pdp, newpwd);
3896 	PWDDESC_XUNLOCK(pdp);
3897 	pwd_drop(oldpwd);
3898 	return (0);
3899 }
3900 
3901 void
3902 pwd_ensure_dirs(void)
3903 {
3904 	struct pwddesc *pdp;
3905 	struct pwd *oldpwd, *newpwd;
3906 
3907 	pdp = curproc->p_pd;
3908 	PWDDESC_XLOCK(pdp);
3909 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3910 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3911 		PWDDESC_XUNLOCK(pdp);
3912 		return;
3913 	}
3914 	PWDDESC_XUNLOCK(pdp);
3915 
3916 	newpwd = pwd_alloc();
3917 	PWDDESC_XLOCK(pdp);
3918 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3919 	pwd_fill(oldpwd, newpwd);
3920 	if (newpwd->pwd_cdir == NULL) {
3921 		vrefact(rootvnode);
3922 		newpwd->pwd_cdir = rootvnode;
3923 	}
3924 	if (newpwd->pwd_rdir == NULL) {
3925 		vrefact(rootvnode);
3926 		newpwd->pwd_rdir = rootvnode;
3927 	}
3928 	pwd_set(pdp, newpwd);
3929 	PWDDESC_XUNLOCK(pdp);
3930 	pwd_drop(oldpwd);
3931 }
3932 
3933 void
3934 pwd_set_rootvnode(void)
3935 {
3936 	struct pwddesc *pdp;
3937 	struct pwd *oldpwd, *newpwd;
3938 
3939 	pdp = curproc->p_pd;
3940 
3941 	newpwd = pwd_alloc();
3942 	PWDDESC_XLOCK(pdp);
3943 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3944 	vrefact(rootvnode);
3945 	newpwd->pwd_cdir = rootvnode;
3946 	vrefact(rootvnode);
3947 	newpwd->pwd_rdir = rootvnode;
3948 	pwd_fill(oldpwd, newpwd);
3949 	pwd_set(pdp, newpwd);
3950 	PWDDESC_XUNLOCK(pdp);
3951 	pwd_drop(oldpwd);
3952 }
3953 
3954 /*
3955  * Scan all active processes and prisons to see if any of them have a current
3956  * or root directory of `olddp'. If so, replace them with the new mount point.
3957  */
3958 void
3959 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3960 {
3961 	struct pwddesc *pdp;
3962 	struct pwd *newpwd, *oldpwd;
3963 	struct prison *pr;
3964 	struct proc *p;
3965 	int nrele;
3966 
3967 	if (vrefcnt(olddp) == 1)
3968 		return;
3969 	nrele = 0;
3970 	newpwd = pwd_alloc();
3971 	sx_slock(&allproc_lock);
3972 	FOREACH_PROC_IN_SYSTEM(p) {
3973 		PROC_LOCK(p);
3974 		pdp = pdhold(p);
3975 		PROC_UNLOCK(p);
3976 		if (pdp == NULL)
3977 			continue;
3978 		PWDDESC_XLOCK(pdp);
3979 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3980 		if (oldpwd == NULL ||
3981 		    (oldpwd->pwd_cdir != olddp &&
3982 		    oldpwd->pwd_rdir != olddp &&
3983 		    oldpwd->pwd_jdir != olddp)) {
3984 			PWDDESC_XUNLOCK(pdp);
3985 			pddrop(pdp);
3986 			continue;
3987 		}
3988 		if (oldpwd->pwd_cdir == olddp) {
3989 			vrefact(newdp);
3990 			newpwd->pwd_cdir = newdp;
3991 		}
3992 		if (oldpwd->pwd_rdir == olddp) {
3993 			vrefact(newdp);
3994 			newpwd->pwd_rdir = newdp;
3995 		}
3996 		if (oldpwd->pwd_jdir == olddp) {
3997 			vrefact(newdp);
3998 			newpwd->pwd_jdir = newdp;
3999 		}
4000 		pwd_fill(oldpwd, newpwd);
4001 		pwd_set(pdp, newpwd);
4002 		PWDDESC_XUNLOCK(pdp);
4003 		pwd_drop(oldpwd);
4004 		pddrop(pdp);
4005 		newpwd = pwd_alloc();
4006 	}
4007 	sx_sunlock(&allproc_lock);
4008 	pwd_drop(newpwd);
4009 	if (rootvnode == olddp) {
4010 		vrefact(newdp);
4011 		rootvnode = newdp;
4012 		nrele++;
4013 	}
4014 	mtx_lock(&prison0.pr_mtx);
4015 	if (prison0.pr_root == olddp) {
4016 		vrefact(newdp);
4017 		prison0.pr_root = newdp;
4018 		nrele++;
4019 	}
4020 	mtx_unlock(&prison0.pr_mtx);
4021 	sx_slock(&allprison_lock);
4022 	TAILQ_FOREACH(pr, &allprison, pr_list) {
4023 		mtx_lock(&pr->pr_mtx);
4024 		if (pr->pr_root == olddp) {
4025 			vrefact(newdp);
4026 			pr->pr_root = newdp;
4027 			nrele++;
4028 		}
4029 		mtx_unlock(&pr->pr_mtx);
4030 	}
4031 	sx_sunlock(&allprison_lock);
4032 	while (nrele--)
4033 		vrele(olddp);
4034 }
4035 
4036 struct filedesc_to_leader *
4037 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4038 {
4039 	struct filedesc_to_leader *fdtol;
4040 
4041 	fdtol = malloc(sizeof(struct filedesc_to_leader),
4042 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4043 	fdtol->fdl_refcount = 1;
4044 	fdtol->fdl_holdcount = 0;
4045 	fdtol->fdl_wakeup = 0;
4046 	fdtol->fdl_leader = leader;
4047 	if (old != NULL) {
4048 		FILEDESC_XLOCK(fdp);
4049 		fdtol->fdl_next = old->fdl_next;
4050 		fdtol->fdl_prev = old;
4051 		old->fdl_next = fdtol;
4052 		fdtol->fdl_next->fdl_prev = fdtol;
4053 		FILEDESC_XUNLOCK(fdp);
4054 	} else {
4055 		fdtol->fdl_next = fdtol;
4056 		fdtol->fdl_prev = fdtol;
4057 	}
4058 	return (fdtol);
4059 }
4060 
4061 static int
4062 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4063 {
4064 	NDSLOTTYPE *map;
4065 	struct filedesc *fdp;
4066 	int count, off, minoff;
4067 
4068 	if (*(int *)arg1 != 0)
4069 		return (EINVAL);
4070 
4071 	fdp = curproc->p_fd;
4072 	count = 0;
4073 	FILEDESC_SLOCK(fdp);
4074 	map = fdp->fd_map;
4075 	off = NDSLOT(fdp->fd_nfiles - 1);
4076 	for (minoff = NDSLOT(0); off >= minoff; --off)
4077 		count += bitcountl(map[off]);
4078 	FILEDESC_SUNLOCK(fdp);
4079 
4080 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4081 }
4082 
4083 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4084     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4085     "Number of open file descriptors");
4086 
4087 /*
4088  * Get file structures globally.
4089  */
4090 static int
4091 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4092 {
4093 	struct xfile xf;
4094 	struct filedesc *fdp;
4095 	struct file *fp;
4096 	struct proc *p;
4097 	int error, n, lastfile;
4098 
4099 	error = sysctl_wire_old_buffer(req, 0);
4100 	if (error != 0)
4101 		return (error);
4102 	if (req->oldptr == NULL) {
4103 		n = 0;
4104 		sx_slock(&allproc_lock);
4105 		FOREACH_PROC_IN_SYSTEM(p) {
4106 			PROC_LOCK(p);
4107 			if (p->p_state == PRS_NEW) {
4108 				PROC_UNLOCK(p);
4109 				continue;
4110 			}
4111 			fdp = fdhold(p);
4112 			PROC_UNLOCK(p);
4113 			if (fdp == NULL)
4114 				continue;
4115 			/* overestimates sparse tables. */
4116 			n += fdp->fd_nfiles;
4117 			fddrop(fdp);
4118 		}
4119 		sx_sunlock(&allproc_lock);
4120 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4121 	}
4122 	error = 0;
4123 	bzero(&xf, sizeof(xf));
4124 	xf.xf_size = sizeof(xf);
4125 	sx_slock(&allproc_lock);
4126 	FOREACH_PROC_IN_SYSTEM(p) {
4127 		PROC_LOCK(p);
4128 		if (p->p_state == PRS_NEW) {
4129 			PROC_UNLOCK(p);
4130 			continue;
4131 		}
4132 		if (p_cansee(req->td, p) != 0) {
4133 			PROC_UNLOCK(p);
4134 			continue;
4135 		}
4136 		xf.xf_pid = p->p_pid;
4137 		xf.xf_uid = p->p_ucred->cr_uid;
4138 		fdp = fdhold(p);
4139 		PROC_UNLOCK(p);
4140 		if (fdp == NULL)
4141 			continue;
4142 		FILEDESC_SLOCK(fdp);
4143 		lastfile = fdlastfile(fdp);
4144 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4145 		    n++) {
4146 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4147 				continue;
4148 			xf.xf_fd = n;
4149 			xf.xf_file = (uintptr_t)fp;
4150 			xf.xf_data = (uintptr_t)fp->f_data;
4151 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4152 			xf.xf_type = (uintptr_t)fp->f_type;
4153 			xf.xf_count = refcount_load(&fp->f_count);
4154 			xf.xf_msgcount = 0;
4155 			xf.xf_offset = foffset_get(fp);
4156 			xf.xf_flag = fp->f_flag;
4157 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4158 			if (error)
4159 				break;
4160 		}
4161 		FILEDESC_SUNLOCK(fdp);
4162 		fddrop(fdp);
4163 		if (error)
4164 			break;
4165 	}
4166 	sx_sunlock(&allproc_lock);
4167 	return (error);
4168 }
4169 
4170 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4171     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4172 
4173 #ifdef KINFO_FILE_SIZE
4174 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4175 #endif
4176 
4177 static int
4178 xlate_fflags(int fflags)
4179 {
4180 	static const struct {
4181 		int	fflag;
4182 		int	kf_fflag;
4183 	} fflags_table[] = {
4184 		{ FAPPEND, KF_FLAG_APPEND },
4185 		{ FASYNC, KF_FLAG_ASYNC },
4186 		{ FFSYNC, KF_FLAG_FSYNC },
4187 		{ FHASLOCK, KF_FLAG_HASLOCK },
4188 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4189 		{ FREAD, KF_FLAG_READ },
4190 		{ FWRITE, KF_FLAG_WRITE },
4191 		{ O_CREAT, KF_FLAG_CREAT },
4192 		{ O_DIRECT, KF_FLAG_DIRECT },
4193 		{ O_EXCL, KF_FLAG_EXCL },
4194 		{ O_EXEC, KF_FLAG_EXEC },
4195 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4196 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4197 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4198 		{ O_TRUNC, KF_FLAG_TRUNC }
4199 	};
4200 	unsigned int i;
4201 	int kflags;
4202 
4203 	kflags = 0;
4204 	for (i = 0; i < nitems(fflags_table); i++)
4205 		if (fflags & fflags_table[i].fflag)
4206 			kflags |=  fflags_table[i].kf_fflag;
4207 	return (kflags);
4208 }
4209 
4210 /* Trim unused data from kf_path by truncating the structure size. */
4211 void
4212 pack_kinfo(struct kinfo_file *kif)
4213 {
4214 
4215 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4216 	    strlen(kif->kf_path) + 1;
4217 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4218 }
4219 
4220 static void
4221 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4222     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4223 {
4224 	int error;
4225 
4226 	bzero(kif, sizeof(*kif));
4227 
4228 	/* Set a default type to allow for empty fill_kinfo() methods. */
4229 	kif->kf_type = KF_TYPE_UNKNOWN;
4230 	kif->kf_flags = xlate_fflags(fp->f_flag);
4231 	if (rightsp != NULL)
4232 		kif->kf_cap_rights = *rightsp;
4233 	else
4234 		cap_rights_init_zero(&kif->kf_cap_rights);
4235 	kif->kf_fd = fd;
4236 	kif->kf_ref_count = refcount_load(&fp->f_count);
4237 	kif->kf_offset = foffset_get(fp);
4238 
4239 	/*
4240 	 * This may drop the filedesc lock, so the 'fp' cannot be
4241 	 * accessed after this call.
4242 	 */
4243 	error = fo_fill_kinfo(fp, kif, fdp);
4244 	if (error == 0)
4245 		kif->kf_status |= KF_ATTR_VALID;
4246 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4247 		pack_kinfo(kif);
4248 	else
4249 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4250 }
4251 
4252 static void
4253 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4254     struct kinfo_file *kif, int flags)
4255 {
4256 	int error;
4257 
4258 	bzero(kif, sizeof(*kif));
4259 
4260 	kif->kf_type = KF_TYPE_VNODE;
4261 	error = vn_fill_kinfo_vnode(vp, kif);
4262 	if (error == 0)
4263 		kif->kf_status |= KF_ATTR_VALID;
4264 	kif->kf_flags = xlate_fflags(fflags);
4265 	cap_rights_init_zero(&kif->kf_cap_rights);
4266 	kif->kf_fd = fd;
4267 	kif->kf_ref_count = -1;
4268 	kif->kf_offset = -1;
4269 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4270 		pack_kinfo(kif);
4271 	else
4272 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4273 	vrele(vp);
4274 }
4275 
4276 struct export_fd_buf {
4277 	struct filedesc		*fdp;
4278 	struct pwddesc	*pdp;
4279 	struct sbuf 		*sb;
4280 	ssize_t			remainder;
4281 	struct kinfo_file	kif;
4282 	int			flags;
4283 };
4284 
4285 static int
4286 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4287 {
4288 	struct kinfo_file *kif;
4289 
4290 	kif = &efbuf->kif;
4291 	if (efbuf->remainder != -1) {
4292 		if (efbuf->remainder < kif->kf_structsize) {
4293 			/* Terminate export. */
4294 			efbuf->remainder = 0;
4295 			return (0);
4296 		}
4297 		efbuf->remainder -= kif->kf_structsize;
4298 	}
4299 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4300 }
4301 
4302 static int
4303 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4304     struct export_fd_buf *efbuf)
4305 {
4306 	int error;
4307 
4308 	if (efbuf->remainder == 0)
4309 		return (0);
4310 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4311 	    efbuf->flags);
4312 	FILEDESC_SUNLOCK(efbuf->fdp);
4313 	error = export_kinfo_to_sb(efbuf);
4314 	FILEDESC_SLOCK(efbuf->fdp);
4315 	return (error);
4316 }
4317 
4318 static int
4319 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4320     struct export_fd_buf *efbuf)
4321 {
4322 	int error;
4323 
4324 	if (efbuf->remainder == 0)
4325 		return (0);
4326 	if (efbuf->pdp != NULL)
4327 		PWDDESC_XUNLOCK(efbuf->pdp);
4328 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4329 	error = export_kinfo_to_sb(efbuf);
4330 	if (efbuf->pdp != NULL)
4331 		PWDDESC_XLOCK(efbuf->pdp);
4332 	return (error);
4333 }
4334 
4335 /*
4336  * Store a process file descriptor information to sbuf.
4337  *
4338  * Takes a locked proc as argument, and returns with the proc unlocked.
4339  */
4340 int
4341 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4342     int flags)
4343 {
4344 	struct file *fp;
4345 	struct filedesc *fdp;
4346 	struct pwddesc *pdp;
4347 	struct export_fd_buf *efbuf;
4348 	struct vnode *cttyvp, *textvp, *tracevp;
4349 	struct pwd *pwd;
4350 	int error, i, lastfile;
4351 	cap_rights_t rights;
4352 
4353 	PROC_LOCK_ASSERT(p, MA_OWNED);
4354 
4355 	/* ktrace vnode */
4356 	tracevp = p->p_tracevp;
4357 	if (tracevp != NULL)
4358 		vrefact(tracevp);
4359 	/* text vnode */
4360 	textvp = p->p_textvp;
4361 	if (textvp != NULL)
4362 		vrefact(textvp);
4363 	/* Controlling tty. */
4364 	cttyvp = NULL;
4365 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4366 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4367 		if (cttyvp != NULL)
4368 			vrefact(cttyvp);
4369 	}
4370 	fdp = fdhold(p);
4371 	pdp = pdhold(p);
4372 	PROC_UNLOCK(p);
4373 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4374 	efbuf->fdp = NULL;
4375 	efbuf->pdp = NULL;
4376 	efbuf->sb = sb;
4377 	efbuf->remainder = maxlen;
4378 	efbuf->flags = flags;
4379 	if (tracevp != NULL)
4380 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4381 		    efbuf);
4382 	if (textvp != NULL)
4383 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4384 	if (cttyvp != NULL)
4385 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4386 		    efbuf);
4387 	error = 0;
4388 	if (pdp == NULL || fdp == NULL)
4389 		goto fail;
4390 	efbuf->fdp = fdp;
4391 	efbuf->pdp = pdp;
4392 	PWDDESC_XLOCK(pdp);
4393 	pwd = pwd_hold_pwddesc(pdp);
4394 	if (pwd != NULL) {
4395 		/* working directory */
4396 		if (pwd->pwd_cdir != NULL) {
4397 			vrefact(pwd->pwd_cdir);
4398 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4399 		}
4400 		/* root directory */
4401 		if (pwd->pwd_rdir != NULL) {
4402 			vrefact(pwd->pwd_rdir);
4403 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4404 		}
4405 		/* jail directory */
4406 		if (pwd->pwd_jdir != NULL) {
4407 			vrefact(pwd->pwd_jdir);
4408 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4409 		}
4410 	}
4411 	PWDDESC_XUNLOCK(pdp);
4412 	if (pwd != NULL)
4413 		pwd_drop(pwd);
4414 	FILEDESC_SLOCK(fdp);
4415 	lastfile = fdlastfile(fdp);
4416 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4417 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4418 			continue;
4419 #ifdef CAPABILITIES
4420 		rights = *cap_rights(fdp, i);
4421 #else /* !CAPABILITIES */
4422 		rights = cap_no_rights;
4423 #endif
4424 		/*
4425 		 * Create sysctl entry.  It is OK to drop the filedesc
4426 		 * lock inside of export_file_to_sb() as we will
4427 		 * re-validate and re-evaluate its properties when the
4428 		 * loop continues.
4429 		 */
4430 		error = export_file_to_sb(fp, i, &rights, efbuf);
4431 		if (error != 0 || efbuf->remainder == 0)
4432 			break;
4433 	}
4434 	FILEDESC_SUNLOCK(fdp);
4435 fail:
4436 	if (fdp != NULL)
4437 		fddrop(fdp);
4438 	if (pdp != NULL)
4439 		pddrop(pdp);
4440 	free(efbuf, M_TEMP);
4441 	return (error);
4442 }
4443 
4444 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4445 
4446 /*
4447  * Get per-process file descriptors for use by procstat(1), et al.
4448  */
4449 static int
4450 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4451 {
4452 	struct sbuf sb;
4453 	struct proc *p;
4454 	ssize_t maxlen;
4455 	int error, error2, *name;
4456 
4457 	name = (int *)arg1;
4458 
4459 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4460 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4461 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4462 	if (error != 0) {
4463 		sbuf_delete(&sb);
4464 		return (error);
4465 	}
4466 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4467 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4468 	    KERN_FILEDESC_PACK_KINFO);
4469 	error2 = sbuf_finish(&sb);
4470 	sbuf_delete(&sb);
4471 	return (error != 0 ? error : error2);
4472 }
4473 
4474 #ifdef COMPAT_FREEBSD7
4475 #ifdef KINFO_OFILE_SIZE
4476 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4477 #endif
4478 
4479 static void
4480 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4481 {
4482 
4483 	okif->kf_structsize = sizeof(*okif);
4484 	okif->kf_type = kif->kf_type;
4485 	okif->kf_fd = kif->kf_fd;
4486 	okif->kf_ref_count = kif->kf_ref_count;
4487 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4488 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4489 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4490 	okif->kf_offset = kif->kf_offset;
4491 	if (kif->kf_type == KF_TYPE_VNODE)
4492 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4493 	else
4494 		okif->kf_vnode_type = KF_VTYPE_VNON;
4495 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4496 	if (kif->kf_type == KF_TYPE_SOCKET) {
4497 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4498 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4499 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4500 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4501 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4502 	} else {
4503 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4504 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4505 	}
4506 }
4507 
4508 static int
4509 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4510     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4511 {
4512 	int error;
4513 
4514 	vrefact(vp);
4515 	PWDDESC_XUNLOCK(pdp);
4516 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4517 	kinfo_to_okinfo(kif, okif);
4518 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4519 	PWDDESC_XLOCK(pdp);
4520 	return (error);
4521 }
4522 
4523 /*
4524  * Get per-process file descriptors for use by procstat(1), et al.
4525  */
4526 static int
4527 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4528 {
4529 	struct kinfo_ofile *okif;
4530 	struct kinfo_file *kif;
4531 	struct filedesc *fdp;
4532 	struct pwddesc *pdp;
4533 	struct pwd *pwd;
4534 	int error, i, lastfile, *name;
4535 	struct file *fp;
4536 	struct proc *p;
4537 
4538 	name = (int *)arg1;
4539 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4540 	if (error != 0)
4541 		return (error);
4542 	fdp = fdhold(p);
4543 	if (fdp != NULL)
4544 		pdp = pdhold(p);
4545 	PROC_UNLOCK(p);
4546 	if (fdp == NULL || pdp == NULL) {
4547 		if (fdp != NULL)
4548 			fddrop(fdp);
4549 		return (ENOENT);
4550 	}
4551 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4552 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4553 	PWDDESC_XLOCK(pdp);
4554 	pwd = pwd_hold_pwddesc(pdp);
4555 	if (pwd != NULL) {
4556 		if (pwd->pwd_cdir != NULL)
4557 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4558 			    okif, pdp, req);
4559 		if (pwd->pwd_rdir != NULL)
4560 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4561 			    okif, pdp, req);
4562 		if (pwd->pwd_jdir != NULL)
4563 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4564 			    okif, pdp, req);
4565 	}
4566 	PWDDESC_XUNLOCK(pdp);
4567 	if (pwd != NULL)
4568 		pwd_drop(pwd);
4569 	FILEDESC_SLOCK(fdp);
4570 	lastfile = fdlastfile(fdp);
4571 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4572 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4573 			continue;
4574 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4575 		    KERN_FILEDESC_PACK_KINFO);
4576 		FILEDESC_SUNLOCK(fdp);
4577 		kinfo_to_okinfo(kif, okif);
4578 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4579 		FILEDESC_SLOCK(fdp);
4580 		if (error)
4581 			break;
4582 	}
4583 	FILEDESC_SUNLOCK(fdp);
4584 	fddrop(fdp);
4585 	pddrop(pdp);
4586 	free(kif, M_TEMP);
4587 	free(okif, M_TEMP);
4588 	return (0);
4589 }
4590 
4591 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4592     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4593     "Process ofiledesc entries");
4594 #endif	/* COMPAT_FREEBSD7 */
4595 
4596 int
4597 vntype_to_kinfo(int vtype)
4598 {
4599 	struct {
4600 		int	vtype;
4601 		int	kf_vtype;
4602 	} vtypes_table[] = {
4603 		{ VBAD, KF_VTYPE_VBAD },
4604 		{ VBLK, KF_VTYPE_VBLK },
4605 		{ VCHR, KF_VTYPE_VCHR },
4606 		{ VDIR, KF_VTYPE_VDIR },
4607 		{ VFIFO, KF_VTYPE_VFIFO },
4608 		{ VLNK, KF_VTYPE_VLNK },
4609 		{ VNON, KF_VTYPE_VNON },
4610 		{ VREG, KF_VTYPE_VREG },
4611 		{ VSOCK, KF_VTYPE_VSOCK }
4612 	};
4613 	unsigned int i;
4614 
4615 	/*
4616 	 * Perform vtype translation.
4617 	 */
4618 	for (i = 0; i < nitems(vtypes_table); i++)
4619 		if (vtypes_table[i].vtype == vtype)
4620 			return (vtypes_table[i].kf_vtype);
4621 
4622 	return (KF_VTYPE_UNKNOWN);
4623 }
4624 
4625 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4626     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4627     "Process filedesc entries");
4628 
4629 /*
4630  * Store a process current working directory information to sbuf.
4631  *
4632  * Takes a locked proc as argument, and returns with the proc unlocked.
4633  */
4634 int
4635 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4636 {
4637 	struct pwddesc *pdp;
4638 	struct pwd *pwd;
4639 	struct export_fd_buf *efbuf;
4640 	struct vnode *cdir;
4641 	int error;
4642 
4643 	PROC_LOCK_ASSERT(p, MA_OWNED);
4644 
4645 	pdp = pdhold(p);
4646 	PROC_UNLOCK(p);
4647 	if (pdp == NULL)
4648 		return (EINVAL);
4649 
4650 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4651 	efbuf->pdp = pdp;
4652 	efbuf->sb = sb;
4653 	efbuf->remainder = maxlen;
4654 
4655 	PWDDESC_XLOCK(pdp);
4656 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4657 	cdir = pwd->pwd_cdir;
4658 	if (cdir == NULL) {
4659 		error = EINVAL;
4660 	} else {
4661 		vrefact(cdir);
4662 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4663 	}
4664 	PWDDESC_XUNLOCK(pdp);
4665 	pddrop(pdp);
4666 	free(efbuf, M_TEMP);
4667 	return (error);
4668 }
4669 
4670 /*
4671  * Get per-process current working directory.
4672  */
4673 static int
4674 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4675 {
4676 	struct sbuf sb;
4677 	struct proc *p;
4678 	ssize_t maxlen;
4679 	int error, error2, *name;
4680 
4681 	name = (int *)arg1;
4682 
4683 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4684 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4685 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4686 	if (error != 0) {
4687 		sbuf_delete(&sb);
4688 		return (error);
4689 	}
4690 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4691 	error = kern_proc_cwd_out(p, &sb, maxlen);
4692 	error2 = sbuf_finish(&sb);
4693 	sbuf_delete(&sb);
4694 	return (error != 0 ? error : error2);
4695 }
4696 
4697 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4698     sysctl_kern_proc_cwd, "Process current working directory");
4699 
4700 #ifdef DDB
4701 /*
4702  * For the purposes of debugging, generate a human-readable string for the
4703  * file type.
4704  */
4705 static const char *
4706 file_type_to_name(short type)
4707 {
4708 
4709 	switch (type) {
4710 	case 0:
4711 		return ("zero");
4712 	case DTYPE_VNODE:
4713 		return ("vnode");
4714 	case DTYPE_SOCKET:
4715 		return ("socket");
4716 	case DTYPE_PIPE:
4717 		return ("pipe");
4718 	case DTYPE_FIFO:
4719 		return ("fifo");
4720 	case DTYPE_KQUEUE:
4721 		return ("kqueue");
4722 	case DTYPE_CRYPTO:
4723 		return ("crypto");
4724 	case DTYPE_MQUEUE:
4725 		return ("mqueue");
4726 	case DTYPE_SHM:
4727 		return ("shm");
4728 	case DTYPE_SEM:
4729 		return ("ksem");
4730 	case DTYPE_PTS:
4731 		return ("pts");
4732 	case DTYPE_DEV:
4733 		return ("dev");
4734 	case DTYPE_PROCDESC:
4735 		return ("proc");
4736 	case DTYPE_EVENTFD:
4737 		return ("eventfd");
4738 	case DTYPE_LINUXTFD:
4739 		return ("ltimer");
4740 	default:
4741 		return ("unkn");
4742 	}
4743 }
4744 
4745 /*
4746  * For the purposes of debugging, identify a process (if any, perhaps one of
4747  * many) that references the passed file in its file descriptor array. Return
4748  * NULL if none.
4749  */
4750 static struct proc *
4751 file_to_first_proc(struct file *fp)
4752 {
4753 	struct filedesc *fdp;
4754 	struct proc *p;
4755 	int n;
4756 
4757 	FOREACH_PROC_IN_SYSTEM(p) {
4758 		if (p->p_state == PRS_NEW)
4759 			continue;
4760 		fdp = p->p_fd;
4761 		if (fdp == NULL)
4762 			continue;
4763 		for (n = 0; n < fdp->fd_nfiles; n++) {
4764 			if (fp == fdp->fd_ofiles[n].fde_file)
4765 				return (p);
4766 		}
4767 	}
4768 	return (NULL);
4769 }
4770 
4771 static void
4772 db_print_file(struct file *fp, int header)
4773 {
4774 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4775 	struct proc *p;
4776 
4777 	if (header)
4778 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4779 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4780 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4781 		    "FCmd");
4782 	p = file_to_first_proc(fp);
4783 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4784 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4785 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4786 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4787 
4788 #undef XPTRWIDTH
4789 }
4790 
4791 DB_SHOW_COMMAND(file, db_show_file)
4792 {
4793 	struct file *fp;
4794 
4795 	if (!have_addr) {
4796 		db_printf("usage: show file <addr>\n");
4797 		return;
4798 	}
4799 	fp = (struct file *)addr;
4800 	db_print_file(fp, 1);
4801 }
4802 
4803 DB_SHOW_COMMAND(files, db_show_files)
4804 {
4805 	struct filedesc *fdp;
4806 	struct file *fp;
4807 	struct proc *p;
4808 	int header;
4809 	int n;
4810 
4811 	header = 1;
4812 	FOREACH_PROC_IN_SYSTEM(p) {
4813 		if (p->p_state == PRS_NEW)
4814 			continue;
4815 		if ((fdp = p->p_fd) == NULL)
4816 			continue;
4817 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4818 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4819 				continue;
4820 			db_print_file(fp, header);
4821 			header = 0;
4822 		}
4823 	}
4824 }
4825 #endif
4826 
4827 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4828     &maxfilesperproc, 0, "Maximum files allowed open per process");
4829 
4830 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4831     &maxfiles, 0, "Maximum number of files");
4832 
4833 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4834     &openfiles, 0, "System-wide number of open files");
4835 
4836 /* ARGSUSED*/
4837 static void
4838 filelistinit(void *dummy)
4839 {
4840 
4841 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4842 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4843 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4844 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4845 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4846 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4847 	/*
4848 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4849 	 * the vnode zone.
4850 	 */
4851 	vfs_smr = uma_zone_get_smr(pwd_zone);
4852 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4853 }
4854 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4855 
4856 /*-------------------------------------------------------------------*/
4857 
4858 static int
4859 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4860     int flags, struct thread *td)
4861 {
4862 
4863 	return (EBADF);
4864 }
4865 
4866 static int
4867 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4868     struct thread *td)
4869 {
4870 
4871 	return (EINVAL);
4872 }
4873 
4874 static int
4875 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4876     struct thread *td)
4877 {
4878 
4879 	return (EBADF);
4880 }
4881 
4882 static int
4883 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4884     struct thread *td)
4885 {
4886 
4887 	return (0);
4888 }
4889 
4890 static int
4891 badfo_kqfilter(struct file *fp, struct knote *kn)
4892 {
4893 
4894 	return (EBADF);
4895 }
4896 
4897 static int
4898 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4899     struct thread *td)
4900 {
4901 
4902 	return (EBADF);
4903 }
4904 
4905 static int
4906 badfo_close(struct file *fp, struct thread *td)
4907 {
4908 
4909 	return (0);
4910 }
4911 
4912 static int
4913 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4914     struct thread *td)
4915 {
4916 
4917 	return (EBADF);
4918 }
4919 
4920 static int
4921 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4922     struct thread *td)
4923 {
4924 
4925 	return (EBADF);
4926 }
4927 
4928 static int
4929 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4930     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4931     struct thread *td)
4932 {
4933 
4934 	return (EBADF);
4935 }
4936 
4937 static int
4938 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4939 {
4940 
4941 	return (0);
4942 }
4943 
4944 struct fileops badfileops = {
4945 	.fo_read = badfo_readwrite,
4946 	.fo_write = badfo_readwrite,
4947 	.fo_truncate = badfo_truncate,
4948 	.fo_ioctl = badfo_ioctl,
4949 	.fo_poll = badfo_poll,
4950 	.fo_kqfilter = badfo_kqfilter,
4951 	.fo_stat = badfo_stat,
4952 	.fo_close = badfo_close,
4953 	.fo_chmod = badfo_chmod,
4954 	.fo_chown = badfo_chown,
4955 	.fo_sendfile = badfo_sendfile,
4956 	.fo_fill_kinfo = badfo_fill_kinfo,
4957 };
4958 
4959 int
4960 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4961     int flags, struct thread *td)
4962 {
4963 
4964 	return (EOPNOTSUPP);
4965 }
4966 
4967 int
4968 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4969     struct thread *td)
4970 {
4971 
4972 	return (EINVAL);
4973 }
4974 
4975 int
4976 invfo_ioctl(struct file *fp, u_long com, void *data,
4977     struct ucred *active_cred, struct thread *td)
4978 {
4979 
4980 	return (ENOTTY);
4981 }
4982 
4983 int
4984 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4985     struct thread *td)
4986 {
4987 
4988 	return (poll_no_poll(events));
4989 }
4990 
4991 int
4992 invfo_kqfilter(struct file *fp, struct knote *kn)
4993 {
4994 
4995 	return (EINVAL);
4996 }
4997 
4998 int
4999 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5000     struct thread *td)
5001 {
5002 
5003 	return (EINVAL);
5004 }
5005 
5006 int
5007 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5008     struct thread *td)
5009 {
5010 
5011 	return (EINVAL);
5012 }
5013 
5014 int
5015 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5016     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5017     struct thread *td)
5018 {
5019 
5020 	return (EINVAL);
5021 }
5022 
5023 /*-------------------------------------------------------------------*/
5024 
5025 /*
5026  * File Descriptor pseudo-device driver (/dev/fd/).
5027  *
5028  * Opening minor device N dup()s the file (if any) connected to file
5029  * descriptor N belonging to the calling process.  Note that this driver
5030  * consists of only the ``open()'' routine, because all subsequent
5031  * references to this file will be direct to the other driver.
5032  *
5033  * XXX: we could give this one a cloning event handler if necessary.
5034  */
5035 
5036 /* ARGSUSED */
5037 static int
5038 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5039 {
5040 
5041 	/*
5042 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5043 	 * the file descriptor being sought for duplication. The error
5044 	 * return ensures that the vnode for this device will be released
5045 	 * by vn_open. Open will detect this special error and take the
5046 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5047 	 * will simply report the error.
5048 	 */
5049 	td->td_dupfd = dev2unit(dev);
5050 	return (ENODEV);
5051 }
5052 
5053 static struct cdevsw fildesc_cdevsw = {
5054 	.d_version =	D_VERSION,
5055 	.d_open =	fdopen,
5056 	.d_name =	"FD",
5057 };
5058 
5059 static void
5060 fildesc_drvinit(void *unused)
5061 {
5062 	struct cdev *dev;
5063 
5064 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5065 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5066 	make_dev_alias(dev, "stdin");
5067 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5068 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5069 	make_dev_alias(dev, "stdout");
5070 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5071 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5072 	make_dev_alias(dev, "stderr");
5073 }
5074 
5075 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5076