xref: /freebsd/sys/kern/kern_descrip.c (revision f7c32ed617858bcd22f8d1b03199099d50125721)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/poll.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sbuf.h>
71 #include <sys/signalvar.h>
72 #include <sys/kdb.h>
73 #include <sys/smr.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
80 #include <sys/user.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97     "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
100 
101 MALLOC_DECLARE(M_FADVISE);
102 
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
106 VFS_SMR_DECLARE;
107 
108 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
109 		    struct thread *td, bool holdleaders, bool audit);
110 static int	fd_first_free(struct filedesc *fdp, int low, int size);
111 static void	fdgrowtable(struct filedesc *fdp, int nfd);
112 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
113 static void	fdunused(struct filedesc *fdp, int fd);
114 static void	fdused(struct filedesc *fdp, int fd);
115 static int	getmaxfd(struct thread *td);
116 static u_long	*filecaps_copy_prep(const struct filecaps *src);
117 static void	filecaps_copy_finish(const struct filecaps *src,
118 		    struct filecaps *dst, u_long *ioctls);
119 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
120 static void	filecaps_free_finish(u_long *ioctls);
121 
122 static struct pwd *pwd_alloc(void);
123 
124 /*
125  * Each process has:
126  *
127  * - An array of open file descriptors (fd_ofiles)
128  * - An array of file flags (fd_ofileflags)
129  * - A bitmap recording which descriptors are in use (fd_map)
130  *
131  * A process starts out with NDFILE descriptors.  The value of NDFILE has
132  * been selected based the historical limit of 20 open files, and an
133  * assumption that the majority of processes, especially short-lived
134  * processes like shells, will never need more.
135  *
136  * If this initial allocation is exhausted, a larger descriptor table and
137  * map are allocated dynamically, and the pointers in the process's struct
138  * filedesc are updated to point to those.  This is repeated every time
139  * the process runs out of file descriptors (provided it hasn't hit its
140  * resource limit).
141  *
142  * Since threads may hold references to individual descriptor table
143  * entries, the tables are never freed.  Instead, they are placed on a
144  * linked list and freed only when the struct filedesc is released.
145  */
146 #define NDFILE		20
147 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
148 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
149 #define NDSLOT(x)	((x) / NDENTRIES)
150 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
151 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
152 
153 /*
154  * SLIST entry used to keep track of ofiles which must be reclaimed when
155  * the process exits.
156  */
157 struct freetable {
158 	struct fdescenttbl *ft_table;
159 	SLIST_ENTRY(freetable) ft_next;
160 };
161 
162 /*
163  * Initial allocation: a filedesc structure + the head of SLIST used to
164  * keep track of old ofiles + enough space for NDFILE descriptors.
165  */
166 
167 struct fdescenttbl0 {
168 	int	fdt_nfiles;
169 	struct	filedescent fdt_ofiles[NDFILE];
170 };
171 
172 struct filedesc0 {
173 	struct filedesc fd_fd;
174 	SLIST_HEAD(, freetable) fd_free;
175 	struct	fdescenttbl0 fd_dfiles;
176 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
177 };
178 
179 /*
180  * Descriptor management.
181  */
182 static int __exclusive_cache_line openfiles; /* actual number of open files */
183 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
184 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
185 
186 /*
187  * If low >= size, just return low. Otherwise find the first zero bit in the
188  * given bitmap, starting at low and not exceeding size - 1. Return size if
189  * not found.
190  */
191 static int
192 fd_first_free(struct filedesc *fdp, int low, int size)
193 {
194 	NDSLOTTYPE *map = fdp->fd_map;
195 	NDSLOTTYPE mask;
196 	int off, maxoff;
197 
198 	if (low >= size)
199 		return (low);
200 
201 	off = NDSLOT(low);
202 	if (low % NDENTRIES) {
203 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
204 		if ((mask &= ~map[off]) != 0UL)
205 			return (off * NDENTRIES + ffsl(mask) - 1);
206 		++off;
207 	}
208 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
209 		if (map[off] != ~0UL)
210 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
211 	return (size);
212 }
213 
214 /*
215  * Find the last used fd.
216  *
217  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
218  * Otherwise use fdlastfile.
219  */
220 int
221 fdlastfile_single(struct filedesc *fdp)
222 {
223 	NDSLOTTYPE *map = fdp->fd_map;
224 	int off, minoff;
225 
226 	off = NDSLOT(fdp->fd_nfiles - 1);
227 	for (minoff = NDSLOT(0); off >= minoff; --off)
228 		if (map[off] != 0)
229 			return (off * NDENTRIES + flsl(map[off]) - 1);
230 	return (-1);
231 }
232 
233 int
234 fdlastfile(struct filedesc *fdp)
235 {
236 
237 	FILEDESC_LOCK_ASSERT(fdp);
238 	return (fdlastfile_single(fdp));
239 }
240 
241 static int
242 fdisused(struct filedesc *fdp, int fd)
243 {
244 
245 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
246 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
247 
248 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
249 }
250 
251 /*
252  * Mark a file descriptor as used.
253  */
254 static void
255 fdused_init(struct filedesc *fdp, int fd)
256 {
257 
258 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
259 
260 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
261 }
262 
263 static void
264 fdused(struct filedesc *fdp, int fd)
265 {
266 
267 	FILEDESC_XLOCK_ASSERT(fdp);
268 
269 	fdused_init(fdp, fd);
270 	if (fd == fdp->fd_freefile)
271 		fdp->fd_freefile++;
272 }
273 
274 /*
275  * Mark a file descriptor as unused.
276  */
277 static void
278 fdunused(struct filedesc *fdp, int fd)
279 {
280 
281 	FILEDESC_XLOCK_ASSERT(fdp);
282 
283 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
284 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
285 	    ("fd=%d is still in use", fd));
286 
287 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
288 	if (fd < fdp->fd_freefile)
289 		fdp->fd_freefile = fd;
290 }
291 
292 /*
293  * Free a file descriptor.
294  *
295  * Avoid some work if fdp is about to be destroyed.
296  */
297 static inline void
298 fdefree_last(struct filedescent *fde)
299 {
300 
301 	filecaps_free(&fde->fde_caps);
302 }
303 
304 static inline void
305 fdfree(struct filedesc *fdp, int fd)
306 {
307 	struct filedescent *fde;
308 
309 	FILEDESC_XLOCK_ASSERT(fdp);
310 	fde = &fdp->fd_ofiles[fd];
311 #ifdef CAPABILITIES
312 	seqc_write_begin(&fde->fde_seqc);
313 #endif
314 	fde->fde_file = NULL;
315 #ifdef CAPABILITIES
316 	seqc_write_end(&fde->fde_seqc);
317 #endif
318 	fdefree_last(fde);
319 	fdunused(fdp, fd);
320 }
321 
322 /*
323  * System calls on descriptors.
324  */
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
327 	int	dummy;
328 };
329 #endif
330 /* ARGSUSED */
331 int
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
333 {
334 #ifdef	RACCT
335 	uint64_t lim;
336 #endif
337 
338 	td->td_retval[0] = getmaxfd(td);
339 #ifdef	RACCT
340 	PROC_LOCK(td->td_proc);
341 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 	PROC_UNLOCK(td->td_proc);
343 	if (lim < td->td_retval[0])
344 		td->td_retval[0] = lim;
345 #endif
346 	return (0);
347 }
348 
349 /*
350  * Duplicate a file descriptor to a particular value.
351  *
352  * Note: keep in mind that a potential race condition exists when closing
353  * descriptors from a shared descriptor table (via rfork).
354  */
355 #ifndef _SYS_SYSPROTO_H_
356 struct dup2_args {
357 	u_int	from;
358 	u_int	to;
359 };
360 #endif
361 /* ARGSUSED */
362 int
363 sys_dup2(struct thread *td, struct dup2_args *uap)
364 {
365 
366 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
367 }
368 
369 /*
370  * Duplicate a file descriptor.
371  */
372 #ifndef _SYS_SYSPROTO_H_
373 struct dup_args {
374 	u_int	fd;
375 };
376 #endif
377 /* ARGSUSED */
378 int
379 sys_dup(struct thread *td, struct dup_args *uap)
380 {
381 
382 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
383 }
384 
385 /*
386  * The file control system call.
387  */
388 #ifndef _SYS_SYSPROTO_H_
389 struct fcntl_args {
390 	int	fd;
391 	int	cmd;
392 	long	arg;
393 };
394 #endif
395 /* ARGSUSED */
396 int
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
398 {
399 
400 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
401 }
402 
403 int
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
405 {
406 	struct flock fl;
407 	struct __oflock ofl;
408 	intptr_t arg1;
409 	int error, newcmd;
410 
411 	error = 0;
412 	newcmd = cmd;
413 	switch (cmd) {
414 	case F_OGETLK:
415 	case F_OSETLK:
416 	case F_OSETLKW:
417 		/*
418 		 * Convert old flock structure to new.
419 		 */
420 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 		fl.l_start = ofl.l_start;
422 		fl.l_len = ofl.l_len;
423 		fl.l_pid = ofl.l_pid;
424 		fl.l_type = ofl.l_type;
425 		fl.l_whence = ofl.l_whence;
426 		fl.l_sysid = 0;
427 
428 		switch (cmd) {
429 		case F_OGETLK:
430 			newcmd = F_GETLK;
431 			break;
432 		case F_OSETLK:
433 			newcmd = F_SETLK;
434 			break;
435 		case F_OSETLKW:
436 			newcmd = F_SETLKW;
437 			break;
438 		}
439 		arg1 = (intptr_t)&fl;
440 		break;
441 	case F_GETLK:
442 	case F_SETLK:
443 	case F_SETLKW:
444 	case F_SETLK_REMOTE:
445 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 		arg1 = (intptr_t)&fl;
447 		break;
448 	default:
449 		arg1 = arg;
450 		break;
451 	}
452 	if (error)
453 		return (error);
454 	error = kern_fcntl(td, fd, newcmd, arg1);
455 	if (error)
456 		return (error);
457 	if (cmd == F_OGETLK) {
458 		ofl.l_start = fl.l_start;
459 		ofl.l_len = fl.l_len;
460 		ofl.l_pid = fl.l_pid;
461 		ofl.l_type = fl.l_type;
462 		ofl.l_whence = fl.l_whence;
463 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 	} else if (cmd == F_GETLK) {
465 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
466 	}
467 	return (error);
468 }
469 
470 int
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
472 {
473 	struct filedesc *fdp;
474 	struct flock *flp;
475 	struct file *fp, *fp2;
476 	struct filedescent *fde;
477 	struct proc *p;
478 	struct vnode *vp;
479 	struct mount *mp;
480 	int error, flg, seals, tmp;
481 	uint64_t bsize;
482 	off_t foffset;
483 
484 	error = 0;
485 	flg = F_POSIX;
486 	p = td->td_proc;
487 	fdp = p->p_fd;
488 
489 	AUDIT_ARG_FD(cmd);
490 	AUDIT_ARG_CMD(cmd);
491 	switch (cmd) {
492 	case F_DUPFD:
493 		tmp = arg;
494 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
495 		break;
496 
497 	case F_DUPFD_CLOEXEC:
498 		tmp = arg;
499 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
500 		break;
501 
502 	case F_DUP2FD:
503 		tmp = arg;
504 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
505 		break;
506 
507 	case F_DUP2FD_CLOEXEC:
508 		tmp = arg;
509 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
510 		break;
511 
512 	case F_GETFD:
513 		error = EBADF;
514 		FILEDESC_SLOCK(fdp);
515 		fde = fdeget_locked(fdp, fd);
516 		if (fde != NULL) {
517 			td->td_retval[0] =
518 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
519 			error = 0;
520 		}
521 		FILEDESC_SUNLOCK(fdp);
522 		break;
523 
524 	case F_SETFD:
525 		error = EBADF;
526 		FILEDESC_XLOCK(fdp);
527 		fde = fdeget_locked(fdp, fd);
528 		if (fde != NULL) {
529 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
531 			error = 0;
532 		}
533 		FILEDESC_XUNLOCK(fdp);
534 		break;
535 
536 	case F_GETFL:
537 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
538 		if (error != 0)
539 			break;
540 		td->td_retval[0] = OFLAGS(fp->f_flag);
541 		fdrop(fp, td);
542 		break;
543 
544 	case F_SETFL:
545 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
546 		if (error != 0)
547 			break;
548 		if (fp->f_ops == &path_fileops) {
549 			fdrop(fp, td);
550 			error = EBADF;
551 			break;
552 		}
553 		do {
554 			tmp = flg = fp->f_flag;
555 			tmp &= ~FCNTLFLAGS;
556 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
557 		} while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
558 		tmp = fp->f_flag & FNONBLOCK;
559 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
560 		if (error != 0) {
561 			fdrop(fp, td);
562 			break;
563 		}
564 		tmp = fp->f_flag & FASYNC;
565 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
566 		if (error == 0) {
567 			fdrop(fp, td);
568 			break;
569 		}
570 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
571 		tmp = 0;
572 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
573 		fdrop(fp, td);
574 		break;
575 
576 	case F_GETOWN:
577 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
578 		if (error != 0)
579 			break;
580 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
581 		if (error == 0)
582 			td->td_retval[0] = tmp;
583 		fdrop(fp, td);
584 		break;
585 
586 	case F_SETOWN:
587 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
588 		if (error != 0)
589 			break;
590 		tmp = arg;
591 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
592 		fdrop(fp, td);
593 		break;
594 
595 	case F_SETLK_REMOTE:
596 		error = priv_check(td, PRIV_NFS_LOCKD);
597 		if (error != 0)
598 			return (error);
599 		flg = F_REMOTE;
600 		goto do_setlk;
601 
602 	case F_SETLKW:
603 		flg |= F_WAIT;
604 		/* FALLTHROUGH F_SETLK */
605 
606 	case F_SETLK:
607 	do_setlk:
608 		flp = (struct flock *)arg;
609 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
610 			error = EINVAL;
611 			break;
612 		}
613 
614 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
615 		if (error != 0)
616 			break;
617 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
618 			error = EBADF;
619 			fdrop(fp, td);
620 			break;
621 		}
622 
623 		if (flp->l_whence == SEEK_CUR) {
624 			foffset = foffset_get(fp);
625 			if (foffset < 0 ||
626 			    (flp->l_start > 0 &&
627 			     foffset > OFF_MAX - flp->l_start)) {
628 				error = EOVERFLOW;
629 				fdrop(fp, td);
630 				break;
631 			}
632 			flp->l_start += foffset;
633 		}
634 
635 		vp = fp->f_vnode;
636 		switch (flp->l_type) {
637 		case F_RDLCK:
638 			if ((fp->f_flag & FREAD) == 0) {
639 				error = EBADF;
640 				break;
641 			}
642 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
643 				PROC_LOCK(p->p_leader);
644 				p->p_leader->p_flag |= P_ADVLOCK;
645 				PROC_UNLOCK(p->p_leader);
646 			}
647 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
648 			    flp, flg);
649 			break;
650 		case F_WRLCK:
651 			if ((fp->f_flag & FWRITE) == 0) {
652 				error = EBADF;
653 				break;
654 			}
655 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
656 				PROC_LOCK(p->p_leader);
657 				p->p_leader->p_flag |= P_ADVLOCK;
658 				PROC_UNLOCK(p->p_leader);
659 			}
660 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
661 			    flp, flg);
662 			break;
663 		case F_UNLCK:
664 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
665 			    flp, flg);
666 			break;
667 		case F_UNLCKSYS:
668 			if (flg != F_REMOTE) {
669 				error = EINVAL;
670 				break;
671 			}
672 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
673 			    F_UNLCKSYS, flp, flg);
674 			break;
675 		default:
676 			error = EINVAL;
677 			break;
678 		}
679 		if (error != 0 || flp->l_type == F_UNLCK ||
680 		    flp->l_type == F_UNLCKSYS) {
681 			fdrop(fp, td);
682 			break;
683 		}
684 
685 		/*
686 		 * Check for a race with close.
687 		 *
688 		 * The vnode is now advisory locked (or unlocked, but this case
689 		 * is not really important) as the caller requested.
690 		 * We had to drop the filedesc lock, so we need to recheck if
691 		 * the descriptor is still valid, because if it was closed
692 		 * in the meantime we need to remove advisory lock from the
693 		 * vnode - close on any descriptor leading to an advisory
694 		 * locked vnode, removes that lock.
695 		 * We will return 0 on purpose in that case, as the result of
696 		 * successful advisory lock might have been externally visible
697 		 * already. This is fine - effectively we pretend to the caller
698 		 * that the closing thread was a bit slower and that the
699 		 * advisory lock succeeded before the close.
700 		 */
701 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
702 		if (error != 0) {
703 			fdrop(fp, td);
704 			break;
705 		}
706 		if (fp != fp2) {
707 			flp->l_whence = SEEK_SET;
708 			flp->l_start = 0;
709 			flp->l_len = 0;
710 			flp->l_type = F_UNLCK;
711 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
712 			    F_UNLCK, flp, F_POSIX);
713 		}
714 		fdrop(fp, td);
715 		fdrop(fp2, td);
716 		break;
717 
718 	case F_GETLK:
719 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
720 		if (error != 0)
721 			break;
722 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
723 			error = EBADF;
724 			fdrop(fp, td);
725 			break;
726 		}
727 		flp = (struct flock *)arg;
728 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
729 		    flp->l_type != F_UNLCK) {
730 			error = EINVAL;
731 			fdrop(fp, td);
732 			break;
733 		}
734 		if (flp->l_whence == SEEK_CUR) {
735 			foffset = foffset_get(fp);
736 			if ((flp->l_start > 0 &&
737 			    foffset > OFF_MAX - flp->l_start) ||
738 			    (flp->l_start < 0 &&
739 			    foffset < OFF_MIN - flp->l_start)) {
740 				error = EOVERFLOW;
741 				fdrop(fp, td);
742 				break;
743 			}
744 			flp->l_start += foffset;
745 		}
746 		vp = fp->f_vnode;
747 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
748 		    F_POSIX);
749 		fdrop(fp, td);
750 		break;
751 
752 	case F_ADD_SEALS:
753 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
754 		if (error != 0)
755 			break;
756 		error = fo_add_seals(fp, arg);
757 		fdrop(fp, td);
758 		break;
759 
760 	case F_GET_SEALS:
761 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
762 		if (error != 0)
763 			break;
764 		if (fo_get_seals(fp, &seals) == 0)
765 			td->td_retval[0] = seals;
766 		else
767 			error = EINVAL;
768 		fdrop(fp, td);
769 		break;
770 
771 	case F_RDAHEAD:
772 		arg = arg ? 128 * 1024: 0;
773 		/* FALLTHROUGH */
774 	case F_READAHEAD:
775 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
776 		if (error != 0)
777 			break;
778 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
779 			fdrop(fp, td);
780 			error = EBADF;
781 			break;
782 		}
783 		vp = fp->f_vnode;
784 		if (vp->v_type != VREG) {
785 			fdrop(fp, td);
786 			error = ENOTTY;
787 			break;
788 		}
789 
790 		/*
791 		 * Exclusive lock synchronizes against f_seqcount reads and
792 		 * writes in sequential_heuristic().
793 		 */
794 		error = vn_lock(vp, LK_EXCLUSIVE);
795 		if (error != 0) {
796 			fdrop(fp, td);
797 			break;
798 		}
799 		if (arg >= 0) {
800 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
801 			arg = MIN(arg, INT_MAX - bsize + 1);
802 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
803 			    (arg + bsize - 1) / bsize);
804 			atomic_set_int(&fp->f_flag, FRDAHEAD);
805 		} else {
806 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
807 		}
808 		VOP_UNLOCK(vp);
809 		fdrop(fp, td);
810 		break;
811 
812 	case F_ISUNIONSTACK:
813 		/*
814 		 * Check if the vnode is part of a union stack (either the
815 		 * "union" flag from mount(2) or unionfs).
816 		 *
817 		 * Prior to introduction of this op libc's readdir would call
818 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
819 		 * data just to check fs name and a mount flag.
820 		 *
821 		 * Fixing the code to handle everything in the kernel instead
822 		 * is a non-trivial endeavor and has low priority, thus this
823 		 * horrible kludge facilitates the current behavior in a much
824 		 * cheaper manner until someone(tm) sorts this out.
825 		 */
826 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
827 		if (error != 0)
828 			break;
829 		if (fp->f_type != DTYPE_VNODE) {
830 			fdrop(fp, td);
831 			error = EBADF;
832 			break;
833 		}
834 		vp = fp->f_vnode;
835 		/*
836 		 * Since we don't prevent dooming the vnode even non-null mp
837 		 * found can become immediately stale. This is tolerable since
838 		 * mount points are type-stable (providing safe memory access)
839 		 * and any vfs op on this vnode going forward will return an
840 		 * error (meaning return value in this case is meaningless).
841 		 */
842 		mp = atomic_load_ptr(&vp->v_mount);
843 		if (__predict_false(mp == NULL)) {
844 			fdrop(fp, td);
845 			error = EBADF;
846 			break;
847 		}
848 		td->td_retval[0] = 0;
849 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
850 		    mp->mnt_flag & MNT_UNION)
851 			td->td_retval[0] = 1;
852 		fdrop(fp, td);
853 		break;
854 
855 	default:
856 		error = EINVAL;
857 		break;
858 	}
859 	return (error);
860 }
861 
862 static int
863 getmaxfd(struct thread *td)
864 {
865 
866 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
867 }
868 
869 /*
870  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
871  */
872 int
873 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
874 {
875 	struct filedesc *fdp;
876 	struct filedescent *oldfde, *newfde;
877 	struct proc *p;
878 	struct file *delfp, *oldfp;
879 	u_long *oioctls, *nioctls;
880 	int error, maxfd;
881 
882 	p = td->td_proc;
883 	fdp = p->p_fd;
884 	oioctls = NULL;
885 
886 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
887 	MPASS(mode < FDDUP_LASTMODE);
888 
889 	AUDIT_ARG_FD(old);
890 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
891 
892 	/*
893 	 * Verify we have a valid descriptor to dup from and possibly to
894 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
895 	 * return EINVAL when the new descriptor is out of bounds.
896 	 */
897 	if (old < 0)
898 		return (EBADF);
899 	if (new < 0)
900 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
901 	maxfd = getmaxfd(td);
902 	if (new >= maxfd)
903 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
904 
905 	error = EBADF;
906 	FILEDESC_XLOCK(fdp);
907 	if (fget_locked(fdp, old) == NULL)
908 		goto unlock;
909 	if (mode == FDDUP_FIXED && old == new) {
910 		td->td_retval[0] = new;
911 		if (flags & FDDUP_FLAG_CLOEXEC)
912 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
913 		error = 0;
914 		goto unlock;
915 	}
916 
917 	oldfde = &fdp->fd_ofiles[old];
918 	oldfp = oldfde->fde_file;
919 	if (!fhold(oldfp))
920 		goto unlock;
921 
922 	/*
923 	 * If the caller specified a file descriptor, make sure the file
924 	 * table is large enough to hold it, and grab it.  Otherwise, just
925 	 * allocate a new descriptor the usual way.
926 	 */
927 	switch (mode) {
928 	case FDDUP_NORMAL:
929 	case FDDUP_FCNTL:
930 		if ((error = fdalloc(td, new, &new)) != 0) {
931 			fdrop(oldfp, td);
932 			goto unlock;
933 		}
934 		break;
935 	case FDDUP_FIXED:
936 		if (new >= fdp->fd_nfiles) {
937 			/*
938 			 * The resource limits are here instead of e.g.
939 			 * fdalloc(), because the file descriptor table may be
940 			 * shared between processes, so we can't really use
941 			 * racct_add()/racct_sub().  Instead of counting the
942 			 * number of actually allocated descriptors, just put
943 			 * the limit on the size of the file descriptor table.
944 			 */
945 #ifdef RACCT
946 			if (RACCT_ENABLED()) {
947 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
948 				if (error != 0) {
949 					error = EMFILE;
950 					fdrop(oldfp, td);
951 					goto unlock;
952 				}
953 			}
954 #endif
955 			fdgrowtable_exp(fdp, new + 1);
956 		}
957 		if (!fdisused(fdp, new))
958 			fdused(fdp, new);
959 		break;
960 	default:
961 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
962 	}
963 
964 	KASSERT(old != new, ("new fd is same as old"));
965 
966 	/* Refetch oldfde because the table may have grown and old one freed. */
967 	oldfde = &fdp->fd_ofiles[old];
968 	KASSERT(oldfp == oldfde->fde_file,
969 	    ("fdt_ofiles shift from growth observed at fd %d",
970 	    old));
971 
972 	newfde = &fdp->fd_ofiles[new];
973 	delfp = newfde->fde_file;
974 
975 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
976 
977 	/*
978 	 * Duplicate the source descriptor.
979 	 */
980 #ifdef CAPABILITIES
981 	seqc_write_begin(&newfde->fde_seqc);
982 #endif
983 	oioctls = filecaps_free_prep(&newfde->fde_caps);
984 	memcpy(newfde, oldfde, fde_change_size);
985 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
986 	    nioctls);
987 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
988 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
989 	else
990 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
991 #ifdef CAPABILITIES
992 	seqc_write_end(&newfde->fde_seqc);
993 #endif
994 	td->td_retval[0] = new;
995 
996 	error = 0;
997 
998 	if (delfp != NULL) {
999 		(void) closefp(fdp, new, delfp, td, true, false);
1000 		FILEDESC_UNLOCK_ASSERT(fdp);
1001 	} else {
1002 unlock:
1003 		FILEDESC_XUNLOCK(fdp);
1004 	}
1005 
1006 	filecaps_free_finish(oioctls);
1007 	return (error);
1008 }
1009 
1010 static void
1011 sigiofree(struct sigio *sigio)
1012 {
1013 	crfree(sigio->sio_ucred);
1014 	free(sigio, M_SIGIO);
1015 }
1016 
1017 static struct sigio *
1018 funsetown_locked(struct sigio *sigio)
1019 {
1020 	struct proc *p;
1021 	struct pgrp *pg;
1022 
1023 	SIGIO_ASSERT_LOCKED();
1024 
1025 	if (sigio == NULL)
1026 		return (NULL);
1027 	*sigio->sio_myref = NULL;
1028 	if (sigio->sio_pgid < 0) {
1029 		pg = sigio->sio_pgrp;
1030 		PGRP_LOCK(pg);
1031 		SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1032 		PGRP_UNLOCK(pg);
1033 	} else {
1034 		p = sigio->sio_proc;
1035 		PROC_LOCK(p);
1036 		SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1037 		PROC_UNLOCK(p);
1038 	}
1039 	return (sigio);
1040 }
1041 
1042 /*
1043  * If sigio is on the list associated with a process or process group,
1044  * disable signalling from the device, remove sigio from the list and
1045  * free sigio.
1046  */
1047 void
1048 funsetown(struct sigio **sigiop)
1049 {
1050 	struct sigio *sigio;
1051 
1052 	/* Racy check, consumers must provide synchronization. */
1053 	if (*sigiop == NULL)
1054 		return;
1055 
1056 	SIGIO_LOCK();
1057 	sigio = funsetown_locked(*sigiop);
1058 	SIGIO_UNLOCK();
1059 	if (sigio != NULL)
1060 		sigiofree(sigio);
1061 }
1062 
1063 /*
1064  * Free a list of sigio structures.  The caller must ensure that new sigio
1065  * structures cannot be added after this point.  For process groups this is
1066  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1067  * as an interlock.
1068  */
1069 void
1070 funsetownlst(struct sigiolst *sigiolst)
1071 {
1072 	struct proc *p;
1073 	struct pgrp *pg;
1074 	struct sigio *sigio, *tmp;
1075 
1076 	/* Racy check. */
1077 	sigio = SLIST_FIRST(sigiolst);
1078 	if (sigio == NULL)
1079 		return;
1080 
1081 	p = NULL;
1082 	pg = NULL;
1083 
1084 	SIGIO_LOCK();
1085 	sigio = SLIST_FIRST(sigiolst);
1086 	if (sigio == NULL) {
1087 		SIGIO_UNLOCK();
1088 		return;
1089 	}
1090 
1091 	/*
1092 	 * Every entry of the list should belong to a single proc or pgrp.
1093 	 */
1094 	if (sigio->sio_pgid < 0) {
1095 		pg = sigio->sio_pgrp;
1096 		sx_assert(&proctree_lock, SX_XLOCKED);
1097 		PGRP_LOCK(pg);
1098 	} else /* if (sigio->sio_pgid > 0) */ {
1099 		p = sigio->sio_proc;
1100 		PROC_LOCK(p);
1101 		KASSERT((p->p_flag & P_WEXIT) != 0,
1102 		    ("%s: process %p is not exiting", __func__, p));
1103 	}
1104 
1105 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1106 		*sigio->sio_myref = NULL;
1107 		if (pg != NULL) {
1108 			KASSERT(sigio->sio_pgid < 0,
1109 			    ("Proc sigio in pgrp sigio list"));
1110 			KASSERT(sigio->sio_pgrp == pg,
1111 			    ("Bogus pgrp in sigio list"));
1112 		} else /* if (p != NULL) */ {
1113 			KASSERT(sigio->sio_pgid > 0,
1114 			    ("Pgrp sigio in proc sigio list"));
1115 			KASSERT(sigio->sio_proc == p,
1116 			    ("Bogus proc in sigio list"));
1117 		}
1118 	}
1119 
1120 	if (pg != NULL)
1121 		PGRP_UNLOCK(pg);
1122 	else
1123 		PROC_UNLOCK(p);
1124 	SIGIO_UNLOCK();
1125 
1126 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1127 		sigiofree(sigio);
1128 }
1129 
1130 /*
1131  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1132  *
1133  * After permission checking, add a sigio structure to the sigio list for
1134  * the process or process group.
1135  */
1136 int
1137 fsetown(pid_t pgid, struct sigio **sigiop)
1138 {
1139 	struct proc *proc;
1140 	struct pgrp *pgrp;
1141 	struct sigio *osigio, *sigio;
1142 	int ret;
1143 
1144 	if (pgid == 0) {
1145 		funsetown(sigiop);
1146 		return (0);
1147 	}
1148 
1149 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1150 	sigio->sio_pgid = pgid;
1151 	sigio->sio_ucred = crhold(curthread->td_ucred);
1152 	sigio->sio_myref = sigiop;
1153 
1154 	ret = 0;
1155 	if (pgid > 0) {
1156 		ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1157 		SIGIO_LOCK();
1158 		osigio = funsetown_locked(*sigiop);
1159 		if (ret == 0) {
1160 			PROC_LOCK(proc);
1161 			_PRELE(proc);
1162 			if ((proc->p_flag & P_WEXIT) != 0) {
1163 				ret = ESRCH;
1164 			} else if (proc->p_session !=
1165 			    curthread->td_proc->p_session) {
1166 				/*
1167 				 * Policy - Don't allow a process to FSETOWN a
1168 				 * process in another session.
1169 				 *
1170 				 * Remove this test to allow maximum flexibility
1171 				 * or restrict FSETOWN to the current process or
1172 				 * process group for maximum safety.
1173 				 */
1174 				ret = EPERM;
1175 			} else {
1176 				sigio->sio_proc = proc;
1177 				SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1178 				    sio_pgsigio);
1179 			}
1180 			PROC_UNLOCK(proc);
1181 		}
1182 	} else /* if (pgid < 0) */ {
1183 		sx_slock(&proctree_lock);
1184 		SIGIO_LOCK();
1185 		osigio = funsetown_locked(*sigiop);
1186 		pgrp = pgfind(-pgid);
1187 		if (pgrp == NULL) {
1188 			ret = ESRCH;
1189 		} else {
1190 			if (pgrp->pg_session != curthread->td_proc->p_session) {
1191 				/*
1192 				 * Policy - Don't allow a process to FSETOWN a
1193 				 * process in another session.
1194 				 *
1195 				 * Remove this test to allow maximum flexibility
1196 				 * or restrict FSETOWN to the current process or
1197 				 * process group for maximum safety.
1198 				 */
1199 				ret = EPERM;
1200 			} else {
1201 				sigio->sio_pgrp = pgrp;
1202 				SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1203 				    sio_pgsigio);
1204 			}
1205 			PGRP_UNLOCK(pgrp);
1206 		}
1207 		sx_sunlock(&proctree_lock);
1208 	}
1209 	if (ret == 0)
1210 		*sigiop = sigio;
1211 	SIGIO_UNLOCK();
1212 	if (osigio != NULL)
1213 		sigiofree(osigio);
1214 	return (ret);
1215 }
1216 
1217 /*
1218  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1219  */
1220 pid_t
1221 fgetown(struct sigio **sigiop)
1222 {
1223 	pid_t pgid;
1224 
1225 	SIGIO_LOCK();
1226 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1227 	SIGIO_UNLOCK();
1228 	return (pgid);
1229 }
1230 
1231 static int
1232 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1233     bool audit)
1234 {
1235 	int error;
1236 
1237 	FILEDESC_XLOCK_ASSERT(fdp);
1238 
1239 	/*
1240 	 * We now hold the fp reference that used to be owned by the
1241 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1242 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1243 	 * added, and deleteing a knote for the new fd.
1244 	 */
1245 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1246 		knote_fdclose(td, fd);
1247 
1248 	/*
1249 	 * We need to notify mqueue if the object is of type mqueue.
1250 	 */
1251 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1252 		mq_fdclose(td, fd, fp);
1253 	FILEDESC_XUNLOCK(fdp);
1254 
1255 #ifdef AUDIT
1256 	if (AUDITING_TD(td) && audit)
1257 		audit_sysclose(td, fd, fp);
1258 #endif
1259 	error = closef(fp, td);
1260 
1261 	/*
1262 	 * All paths leading up to closefp() will have already removed or
1263 	 * replaced the fd in the filedesc table, so a restart would not
1264 	 * operate on the same file.
1265 	 */
1266 	if (error == ERESTART)
1267 		error = EINTR;
1268 
1269 	return (error);
1270 }
1271 
1272 static int
1273 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1274     bool holdleaders, bool audit)
1275 {
1276 	int error;
1277 
1278 	FILEDESC_XLOCK_ASSERT(fdp);
1279 
1280 	if (holdleaders) {
1281 		if (td->td_proc->p_fdtol != NULL) {
1282 			/*
1283 			 * Ask fdfree() to sleep to ensure that all relevant
1284 			 * process leaders can be traversed in closef().
1285 			 */
1286 			fdp->fd_holdleaderscount++;
1287 		} else {
1288 			holdleaders = false;
1289 		}
1290 	}
1291 
1292 	error = closefp_impl(fdp, fd, fp, td, audit);
1293 	if (holdleaders) {
1294 		FILEDESC_XLOCK(fdp);
1295 		fdp->fd_holdleaderscount--;
1296 		if (fdp->fd_holdleaderscount == 0 &&
1297 		    fdp->fd_holdleaderswakeup != 0) {
1298 			fdp->fd_holdleaderswakeup = 0;
1299 			wakeup(&fdp->fd_holdleaderscount);
1300 		}
1301 		FILEDESC_XUNLOCK(fdp);
1302 	}
1303 	return (error);
1304 }
1305 
1306 static int
1307 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1308     bool holdleaders, bool audit)
1309 {
1310 
1311 	FILEDESC_XLOCK_ASSERT(fdp);
1312 
1313 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1314 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1315 	} else {
1316 		return (closefp_impl(fdp, fd, fp, td, audit));
1317 	}
1318 }
1319 
1320 /*
1321  * Close a file descriptor.
1322  */
1323 #ifndef _SYS_SYSPROTO_H_
1324 struct close_args {
1325 	int     fd;
1326 };
1327 #endif
1328 /* ARGSUSED */
1329 int
1330 sys_close(struct thread *td, struct close_args *uap)
1331 {
1332 
1333 	return (kern_close(td, uap->fd));
1334 }
1335 
1336 int
1337 kern_close(struct thread *td, int fd)
1338 {
1339 	struct filedesc *fdp;
1340 	struct file *fp;
1341 
1342 	fdp = td->td_proc->p_fd;
1343 
1344 	FILEDESC_XLOCK(fdp);
1345 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1346 		FILEDESC_XUNLOCK(fdp);
1347 		return (EBADF);
1348 	}
1349 	fdfree(fdp, fd);
1350 
1351 	/* closefp() drops the FILEDESC lock for us. */
1352 	return (closefp(fdp, fd, fp, td, true, true));
1353 }
1354 
1355 int
1356 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1357 {
1358 	struct filedesc *fdp;
1359 	const struct fdescenttbl *fdt;
1360 	struct file *fp;
1361 	int fd;
1362 
1363 	/*
1364 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1365 	 * open should not be a usage error.  From a close_range() perspective,
1366 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1367 	 * be a usage error as all fd above 3 are in-fact already closed.
1368 	 */
1369 	if (highfd < lowfd) {
1370 		return (EINVAL);
1371 	}
1372 
1373 	fdp = td->td_proc->p_fd;
1374 	FILEDESC_XLOCK(fdp);
1375 	fdt = atomic_load_ptr(&fdp->fd_files);
1376 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1377 	fd = lowfd;
1378 	if (__predict_false(fd > highfd)) {
1379 		goto out_locked;
1380 	}
1381 	for (;;) {
1382 		fp = fdt->fdt_ofiles[fd].fde_file;
1383 		if (fp == NULL) {
1384 			if (fd == highfd)
1385 				goto out_locked;
1386 		} else {
1387 			fdfree(fdp, fd);
1388 			(void) closefp(fdp, fd, fp, td, true, true);
1389 			if (fd == highfd)
1390 				goto out_unlocked;
1391 			FILEDESC_XLOCK(fdp);
1392 			fdt = atomic_load_ptr(&fdp->fd_files);
1393 		}
1394 		fd++;
1395 	}
1396 out_locked:
1397 	FILEDESC_XUNLOCK(fdp);
1398 out_unlocked:
1399 	return (0);
1400 }
1401 
1402 #ifndef _SYS_SYSPROTO_H_
1403 struct close_range_args {
1404 	u_int	lowfd;
1405 	u_int	highfd;
1406 	int	flags;
1407 };
1408 #endif
1409 int
1410 sys_close_range(struct thread *td, struct close_range_args *uap)
1411 {
1412 
1413 	AUDIT_ARG_FD(uap->lowfd);
1414 	AUDIT_ARG_CMD(uap->highfd);
1415 	AUDIT_ARG_FFLAGS(uap->flags);
1416 
1417 	/* No flags currently defined */
1418 	if (uap->flags != 0)
1419 		return (EINVAL);
1420 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1421 }
1422 
1423 #ifdef COMPAT_FREEBSD12
1424 /*
1425  * Close open file descriptors.
1426  */
1427 #ifndef _SYS_SYSPROTO_H_
1428 struct freebsd12_closefrom_args {
1429 	int	lowfd;
1430 };
1431 #endif
1432 /* ARGSUSED */
1433 int
1434 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1435 {
1436 	u_int lowfd;
1437 
1438 	AUDIT_ARG_FD(uap->lowfd);
1439 
1440 	/*
1441 	 * Treat negative starting file descriptor values identical to
1442 	 * closefrom(0) which closes all files.
1443 	 */
1444 	lowfd = MAX(0, uap->lowfd);
1445 	return (kern_close_range(td, lowfd, ~0U));
1446 }
1447 #endif	/* COMPAT_FREEBSD12 */
1448 
1449 #if defined(COMPAT_43)
1450 /*
1451  * Return status information about a file descriptor.
1452  */
1453 #ifndef _SYS_SYSPROTO_H_
1454 struct ofstat_args {
1455 	int	fd;
1456 	struct	ostat *sb;
1457 };
1458 #endif
1459 /* ARGSUSED */
1460 int
1461 ofstat(struct thread *td, struct ofstat_args *uap)
1462 {
1463 	struct ostat oub;
1464 	struct stat ub;
1465 	int error;
1466 
1467 	error = kern_fstat(td, uap->fd, &ub);
1468 	if (error == 0) {
1469 		cvtstat(&ub, &oub);
1470 		error = copyout(&oub, uap->sb, sizeof(oub));
1471 	}
1472 	return (error);
1473 }
1474 #endif /* COMPAT_43 */
1475 
1476 #if defined(COMPAT_FREEBSD11)
1477 int
1478 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1479 {
1480 	struct stat sb;
1481 	struct freebsd11_stat osb;
1482 	int error;
1483 
1484 	error = kern_fstat(td, uap->fd, &sb);
1485 	if (error != 0)
1486 		return (error);
1487 	error = freebsd11_cvtstat(&sb, &osb);
1488 	if (error == 0)
1489 		error = copyout(&osb, uap->sb, sizeof(osb));
1490 	return (error);
1491 }
1492 #endif	/* COMPAT_FREEBSD11 */
1493 
1494 /*
1495  * Return status information about a file descriptor.
1496  */
1497 #ifndef _SYS_SYSPROTO_H_
1498 struct fstat_args {
1499 	int	fd;
1500 	struct	stat *sb;
1501 };
1502 #endif
1503 /* ARGSUSED */
1504 int
1505 sys_fstat(struct thread *td, struct fstat_args *uap)
1506 {
1507 	struct stat ub;
1508 	int error;
1509 
1510 	error = kern_fstat(td, uap->fd, &ub);
1511 	if (error == 0)
1512 		error = copyout(&ub, uap->sb, sizeof(ub));
1513 	return (error);
1514 }
1515 
1516 int
1517 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1518 {
1519 	struct file *fp;
1520 	int error;
1521 
1522 	AUDIT_ARG_FD(fd);
1523 
1524 	error = fget(td, fd, &cap_fstat_rights, &fp);
1525 	if (__predict_false(error != 0))
1526 		return (error);
1527 
1528 	AUDIT_ARG_FILE(td->td_proc, fp);
1529 
1530 	error = fo_stat(fp, sbp, td->td_ucred);
1531 	fdrop(fp, td);
1532 #ifdef __STAT_TIME_T_EXT
1533 	sbp->st_atim_ext = 0;
1534 	sbp->st_mtim_ext = 0;
1535 	sbp->st_ctim_ext = 0;
1536 	sbp->st_btim_ext = 0;
1537 #endif
1538 #ifdef KTRACE
1539 	if (KTRPOINT(td, KTR_STRUCT))
1540 		ktrstat_error(sbp, error);
1541 #endif
1542 	return (error);
1543 }
1544 
1545 #if defined(COMPAT_FREEBSD11)
1546 /*
1547  * Return status information about a file descriptor.
1548  */
1549 #ifndef _SYS_SYSPROTO_H_
1550 struct freebsd11_nfstat_args {
1551 	int	fd;
1552 	struct	nstat *sb;
1553 };
1554 #endif
1555 /* ARGSUSED */
1556 int
1557 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1558 {
1559 	struct nstat nub;
1560 	struct stat ub;
1561 	int error;
1562 
1563 	error = kern_fstat(td, uap->fd, &ub);
1564 	if (error == 0) {
1565 		freebsd11_cvtnstat(&ub, &nub);
1566 		error = copyout(&nub, uap->sb, sizeof(nub));
1567 	}
1568 	return (error);
1569 }
1570 #endif /* COMPAT_FREEBSD11 */
1571 
1572 /*
1573  * Return pathconf information about a file descriptor.
1574  */
1575 #ifndef _SYS_SYSPROTO_H_
1576 struct fpathconf_args {
1577 	int	fd;
1578 	int	name;
1579 };
1580 #endif
1581 /* ARGSUSED */
1582 int
1583 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1584 {
1585 	long value;
1586 	int error;
1587 
1588 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1589 	if (error == 0)
1590 		td->td_retval[0] = value;
1591 	return (error);
1592 }
1593 
1594 int
1595 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1596 {
1597 	struct file *fp;
1598 	struct vnode *vp;
1599 	int error;
1600 
1601 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1602 	if (error != 0)
1603 		return (error);
1604 
1605 	if (name == _PC_ASYNC_IO) {
1606 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1607 		goto out;
1608 	}
1609 	vp = fp->f_vnode;
1610 	if (vp != NULL) {
1611 		vn_lock(vp, LK_SHARED | LK_RETRY);
1612 		error = VOP_PATHCONF(vp, name, valuep);
1613 		VOP_UNLOCK(vp);
1614 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1615 		if (name != _PC_PIPE_BUF) {
1616 			error = EINVAL;
1617 		} else {
1618 			*valuep = PIPE_BUF;
1619 			error = 0;
1620 		}
1621 	} else {
1622 		error = EOPNOTSUPP;
1623 	}
1624 out:
1625 	fdrop(fp, td);
1626 	return (error);
1627 }
1628 
1629 /*
1630  * Copy filecaps structure allocating memory for ioctls array if needed.
1631  *
1632  * The last parameter indicates whether the fdtable is locked. If it is not and
1633  * ioctls are encountered, copying fails and the caller must lock the table.
1634  *
1635  * Note that if the table was not locked, the caller has to check the relevant
1636  * sequence counter to determine whether the operation was successful.
1637  */
1638 bool
1639 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1640 {
1641 	size_t size;
1642 
1643 	if (src->fc_ioctls != NULL && !locked)
1644 		return (false);
1645 	memcpy(dst, src, sizeof(*src));
1646 	if (src->fc_ioctls == NULL)
1647 		return (true);
1648 
1649 	KASSERT(src->fc_nioctls > 0,
1650 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1651 
1652 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1653 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1654 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1655 	return (true);
1656 }
1657 
1658 static u_long *
1659 filecaps_copy_prep(const struct filecaps *src)
1660 {
1661 	u_long *ioctls;
1662 	size_t size;
1663 
1664 	if (__predict_true(src->fc_ioctls == NULL))
1665 		return (NULL);
1666 
1667 	KASSERT(src->fc_nioctls > 0,
1668 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1669 
1670 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1671 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1672 	return (ioctls);
1673 }
1674 
1675 static void
1676 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1677     u_long *ioctls)
1678 {
1679 	size_t size;
1680 
1681 	*dst = *src;
1682 	if (__predict_true(src->fc_ioctls == NULL)) {
1683 		MPASS(ioctls == NULL);
1684 		return;
1685 	}
1686 
1687 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1688 	dst->fc_ioctls = ioctls;
1689 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1690 }
1691 
1692 /*
1693  * Move filecaps structure to the new place and clear the old place.
1694  */
1695 void
1696 filecaps_move(struct filecaps *src, struct filecaps *dst)
1697 {
1698 
1699 	*dst = *src;
1700 	bzero(src, sizeof(*src));
1701 }
1702 
1703 /*
1704  * Fill the given filecaps structure with full rights.
1705  */
1706 static void
1707 filecaps_fill(struct filecaps *fcaps)
1708 {
1709 
1710 	CAP_ALL(&fcaps->fc_rights);
1711 	fcaps->fc_ioctls = NULL;
1712 	fcaps->fc_nioctls = -1;
1713 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1714 }
1715 
1716 /*
1717  * Free memory allocated within filecaps structure.
1718  */
1719 void
1720 filecaps_free(struct filecaps *fcaps)
1721 {
1722 
1723 	free(fcaps->fc_ioctls, M_FILECAPS);
1724 	bzero(fcaps, sizeof(*fcaps));
1725 }
1726 
1727 static u_long *
1728 filecaps_free_prep(struct filecaps *fcaps)
1729 {
1730 	u_long *ioctls;
1731 
1732 	ioctls = fcaps->fc_ioctls;
1733 	bzero(fcaps, sizeof(*fcaps));
1734 	return (ioctls);
1735 }
1736 
1737 static void
1738 filecaps_free_finish(u_long *ioctls)
1739 {
1740 
1741 	free(ioctls, M_FILECAPS);
1742 }
1743 
1744 /*
1745  * Validate the given filecaps structure.
1746  */
1747 static void
1748 filecaps_validate(const struct filecaps *fcaps, const char *func)
1749 {
1750 
1751 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1752 	    ("%s: invalid rights", func));
1753 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1754 	    ("%s: invalid fcntls", func));
1755 	KASSERT(fcaps->fc_fcntls == 0 ||
1756 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1757 	    ("%s: fcntls without CAP_FCNTL", func));
1758 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1759 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1760 	    ("%s: invalid ioctls", func));
1761 	KASSERT(fcaps->fc_nioctls == 0 ||
1762 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1763 	    ("%s: ioctls without CAP_IOCTL", func));
1764 }
1765 
1766 static void
1767 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1768 {
1769 	int nfd1;
1770 
1771 	FILEDESC_XLOCK_ASSERT(fdp);
1772 
1773 	nfd1 = fdp->fd_nfiles * 2;
1774 	if (nfd1 < nfd)
1775 		nfd1 = nfd;
1776 	fdgrowtable(fdp, nfd1);
1777 }
1778 
1779 /*
1780  * Grow the file table to accommodate (at least) nfd descriptors.
1781  */
1782 static void
1783 fdgrowtable(struct filedesc *fdp, int nfd)
1784 {
1785 	struct filedesc0 *fdp0;
1786 	struct freetable *ft;
1787 	struct fdescenttbl *ntable;
1788 	struct fdescenttbl *otable;
1789 	int nnfiles, onfiles;
1790 	NDSLOTTYPE *nmap, *omap;
1791 
1792 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1793 
1794 	/* save old values */
1795 	onfiles = fdp->fd_nfiles;
1796 	otable = fdp->fd_files;
1797 	omap = fdp->fd_map;
1798 
1799 	/* compute the size of the new table */
1800 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1801 	if (nnfiles <= onfiles)
1802 		/* the table is already large enough */
1803 		return;
1804 
1805 	/*
1806 	 * Allocate a new table.  We need enough space for the number of
1807 	 * entries, file entries themselves and the struct freetable we will use
1808 	 * when we decommission the table and place it on the freelist.
1809 	 * We place the struct freetable in the middle so we don't have
1810 	 * to worry about padding.
1811 	 */
1812 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1813 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1814 	    sizeof(struct freetable),
1815 	    M_FILEDESC, M_ZERO | M_WAITOK);
1816 	/* copy the old data */
1817 	ntable->fdt_nfiles = nnfiles;
1818 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1819 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1820 
1821 	/*
1822 	 * Allocate a new map only if the old is not large enough.  It will
1823 	 * grow at a slower rate than the table as it can map more
1824 	 * entries than the table can hold.
1825 	 */
1826 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1827 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1828 		    M_ZERO | M_WAITOK);
1829 		/* copy over the old data and update the pointer */
1830 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1831 		fdp->fd_map = nmap;
1832 	}
1833 
1834 	/*
1835 	 * Make sure that ntable is correctly initialized before we replace
1836 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1837 	 * data.
1838 	 */
1839 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1840 
1841 	/*
1842 	 * Free the old file table when not shared by other threads or processes.
1843 	 * The old file table is considered to be shared when either are true:
1844 	 * - The process has more than one thread.
1845 	 * - The file descriptor table has been shared via fdshare().
1846 	 *
1847 	 * When shared, the old file table will be placed on a freelist
1848 	 * which will be processed when the struct filedesc is released.
1849 	 *
1850 	 * Note that if onfiles == NDFILE, we're dealing with the original
1851 	 * static allocation contained within (struct filedesc0 *)fdp,
1852 	 * which must not be freed.
1853 	 */
1854 	if (onfiles > NDFILE) {
1855 		/*
1856 		 * Note we may be called here from fdinit while allocating a
1857 		 * table for a new process in which case ->p_fd points
1858 		 * elsewhere.
1859 		 */
1860 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1861 			free(otable, M_FILEDESC);
1862 		} else {
1863 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1864 			fdp0 = (struct filedesc0 *)fdp;
1865 			ft->ft_table = otable;
1866 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1867 		}
1868 	}
1869 	/*
1870 	 * The map does not have the same possibility of threads still
1871 	 * holding references to it.  So always free it as long as it
1872 	 * does not reference the original static allocation.
1873 	 */
1874 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1875 		free(omap, M_FILEDESC);
1876 }
1877 
1878 /*
1879  * Allocate a file descriptor for the process.
1880  */
1881 int
1882 fdalloc(struct thread *td, int minfd, int *result)
1883 {
1884 	struct proc *p = td->td_proc;
1885 	struct filedesc *fdp = p->p_fd;
1886 	int fd, maxfd, allocfd;
1887 #ifdef RACCT
1888 	int error;
1889 #endif
1890 
1891 	FILEDESC_XLOCK_ASSERT(fdp);
1892 
1893 	if (fdp->fd_freefile > minfd)
1894 		minfd = fdp->fd_freefile;
1895 
1896 	maxfd = getmaxfd(td);
1897 
1898 	/*
1899 	 * Search the bitmap for a free descriptor starting at minfd.
1900 	 * If none is found, grow the file table.
1901 	 */
1902 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1903 	if (__predict_false(fd >= maxfd))
1904 		return (EMFILE);
1905 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1906 		allocfd = min(fd * 2, maxfd);
1907 #ifdef RACCT
1908 		if (RACCT_ENABLED()) {
1909 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1910 			if (error != 0)
1911 				return (EMFILE);
1912 		}
1913 #endif
1914 		/*
1915 		 * fd is already equal to first free descriptor >= minfd, so
1916 		 * we only need to grow the table and we are done.
1917 		 */
1918 		fdgrowtable_exp(fdp, allocfd);
1919 	}
1920 
1921 	/*
1922 	 * Perform some sanity checks, then mark the file descriptor as
1923 	 * used and return it to the caller.
1924 	 */
1925 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1926 	    ("invalid descriptor %d", fd));
1927 	KASSERT(!fdisused(fdp, fd),
1928 	    ("fd_first_free() returned non-free descriptor"));
1929 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1930 	    ("file descriptor isn't free"));
1931 	fdused(fdp, fd);
1932 	*result = fd;
1933 	return (0);
1934 }
1935 
1936 /*
1937  * Allocate n file descriptors for the process.
1938  */
1939 int
1940 fdallocn(struct thread *td, int minfd, int *fds, int n)
1941 {
1942 	struct proc *p = td->td_proc;
1943 	struct filedesc *fdp = p->p_fd;
1944 	int i;
1945 
1946 	FILEDESC_XLOCK_ASSERT(fdp);
1947 
1948 	for (i = 0; i < n; i++)
1949 		if (fdalloc(td, 0, &fds[i]) != 0)
1950 			break;
1951 
1952 	if (i < n) {
1953 		for (i--; i >= 0; i--)
1954 			fdunused(fdp, fds[i]);
1955 		return (EMFILE);
1956 	}
1957 
1958 	return (0);
1959 }
1960 
1961 /*
1962  * Create a new open file structure and allocate a file descriptor for the
1963  * process that refers to it.  We add one reference to the file for the
1964  * descriptor table and one reference for resultfp. This is to prevent us
1965  * being preempted and the entry in the descriptor table closed after we
1966  * release the FILEDESC lock.
1967  */
1968 int
1969 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1970     struct filecaps *fcaps)
1971 {
1972 	struct file *fp;
1973 	int error, fd;
1974 
1975 	MPASS(resultfp != NULL);
1976 	MPASS(resultfd != NULL);
1977 
1978 	error = _falloc_noinstall(td, &fp, 2);
1979 	if (__predict_false(error != 0)) {
1980 		return (error);
1981 	}
1982 
1983 	error = finstall_refed(td, fp, &fd, flags, fcaps);
1984 	if (__predict_false(error != 0)) {
1985 		falloc_abort(td, fp);
1986 		return (error);
1987 	}
1988 
1989 	*resultfp = fp;
1990 	*resultfd = fd;
1991 
1992 	return (0);
1993 }
1994 
1995 /*
1996  * Create a new open file structure without allocating a file descriptor.
1997  */
1998 int
1999 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2000 {
2001 	struct file *fp;
2002 	int maxuserfiles = maxfiles - (maxfiles / 20);
2003 	int openfiles_new;
2004 	static struct timeval lastfail;
2005 	static int curfail;
2006 
2007 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2008 	MPASS(n > 0);
2009 
2010 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2011 	if ((openfiles_new >= maxuserfiles &&
2012 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2013 	    openfiles_new >= maxfiles) {
2014 		atomic_subtract_int(&openfiles, 1);
2015 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2016 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2017 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2018 		}
2019 		return (ENFILE);
2020 	}
2021 	fp = uma_zalloc(file_zone, M_WAITOK);
2022 	bzero(fp, sizeof(*fp));
2023 	refcount_init(&fp->f_count, n);
2024 	fp->f_cred = crhold(td->td_ucred);
2025 	fp->f_ops = &badfileops;
2026 	*resultfp = fp;
2027 	return (0);
2028 }
2029 
2030 void
2031 falloc_abort(struct thread *td, struct file *fp)
2032 {
2033 
2034 	/*
2035 	 * For assertion purposes.
2036 	 */
2037 	refcount_init(&fp->f_count, 0);
2038 	_fdrop(fp, td);
2039 }
2040 
2041 /*
2042  * Install a file in a file descriptor table.
2043  */
2044 void
2045 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2046     struct filecaps *fcaps)
2047 {
2048 	struct filedescent *fde;
2049 
2050 	MPASS(fp != NULL);
2051 	if (fcaps != NULL)
2052 		filecaps_validate(fcaps, __func__);
2053 	FILEDESC_XLOCK_ASSERT(fdp);
2054 
2055 	fde = &fdp->fd_ofiles[fd];
2056 #ifdef CAPABILITIES
2057 	seqc_write_begin(&fde->fde_seqc);
2058 #endif
2059 	fde->fde_file = fp;
2060 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2061 	if (fcaps != NULL)
2062 		filecaps_move(fcaps, &fde->fde_caps);
2063 	else
2064 		filecaps_fill(&fde->fde_caps);
2065 #ifdef CAPABILITIES
2066 	seqc_write_end(&fde->fde_seqc);
2067 #endif
2068 }
2069 
2070 int
2071 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2072     struct filecaps *fcaps)
2073 {
2074 	struct filedesc *fdp = td->td_proc->p_fd;
2075 	int error;
2076 
2077 	MPASS(fd != NULL);
2078 
2079 	FILEDESC_XLOCK(fdp);
2080 	error = fdalloc(td, 0, fd);
2081 	if (__predict_true(error == 0)) {
2082 		_finstall(fdp, fp, *fd, flags, fcaps);
2083 	}
2084 	FILEDESC_XUNLOCK(fdp);
2085 	return (error);
2086 }
2087 
2088 int
2089 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2090     struct filecaps *fcaps)
2091 {
2092 	int error;
2093 
2094 	MPASS(fd != NULL);
2095 
2096 	if (!fhold(fp))
2097 		return (EBADF);
2098 	error = finstall_refed(td, fp, fd, flags, fcaps);
2099 	if (__predict_false(error != 0)) {
2100 		fdrop(fp, td);
2101 	}
2102 	return (error);
2103 }
2104 
2105 /*
2106  * Build a new filedesc structure from another.
2107  *
2108  * If fdp is not NULL, return with it shared locked.
2109  */
2110 struct filedesc *
2111 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2112 {
2113 	struct filedesc0 *newfdp0;
2114 	struct filedesc *newfdp;
2115 
2116 	if (prepfiles)
2117 		MPASS(lastfile != NULL);
2118 	else
2119 		MPASS(lastfile == NULL);
2120 
2121 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2122 	newfdp = &newfdp0->fd_fd;
2123 
2124 	/* Create the file descriptor table. */
2125 	FILEDESC_LOCK_INIT(newfdp);
2126 	refcount_init(&newfdp->fd_refcnt, 1);
2127 	refcount_init(&newfdp->fd_holdcnt, 1);
2128 	newfdp->fd_map = newfdp0->fd_dmap;
2129 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2130 	newfdp->fd_files->fdt_nfiles = NDFILE;
2131 
2132 	if (fdp == NULL)
2133 		return (newfdp);
2134 
2135 	FILEDESC_SLOCK(fdp);
2136 	if (!prepfiles) {
2137 		FILEDESC_SUNLOCK(fdp);
2138 		return (newfdp);
2139 	}
2140 
2141 	for (;;) {
2142 		*lastfile = fdlastfile(fdp);
2143 		if (*lastfile < newfdp->fd_nfiles)
2144 			break;
2145 		FILEDESC_SUNLOCK(fdp);
2146 		fdgrowtable(newfdp, *lastfile + 1);
2147 		FILEDESC_SLOCK(fdp);
2148 	}
2149 
2150 	return (newfdp);
2151 }
2152 
2153 /*
2154  * Build a pwddesc structure from another.
2155  * Copy the current, root, and jail root vnode references.
2156  *
2157  * If pdp is not NULL, return with it shared locked.
2158  */
2159 struct pwddesc *
2160 pdinit(struct pwddesc *pdp, bool keeplock)
2161 {
2162 	struct pwddesc *newpdp;
2163 	struct pwd *newpwd;
2164 
2165 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2166 
2167 	PWDDESC_LOCK_INIT(newpdp);
2168 	refcount_init(&newpdp->pd_refcount, 1);
2169 	newpdp->pd_cmask = CMASK;
2170 
2171 	if (pdp == NULL) {
2172 		newpwd = pwd_alloc();
2173 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2174 		return (newpdp);
2175 	}
2176 
2177 	PWDDESC_XLOCK(pdp);
2178 	newpwd = pwd_hold_pwddesc(pdp);
2179 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2180 	if (!keeplock)
2181 		PWDDESC_XUNLOCK(pdp);
2182 	return (newpdp);
2183 }
2184 
2185 /*
2186  * Hold either filedesc or pwddesc of the passed process.
2187  *
2188  * The process lock is used to synchronize against the target exiting and
2189  * freeing the data.
2190  *
2191  * Clearing can be ilustrated in 3 steps:
2192  * 1. set the pointer to NULL. Either routine can race against it, hence
2193  *   atomic_load_ptr.
2194  * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2195  *   race to either still see the pointer or find NULL. It is still safe to
2196  *   grab a reference as clearing is stalled.
2197  * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2198  *   guaranteed to see NULL, making it safe to finish clearing
2199  */
2200 static struct filedesc *
2201 fdhold(struct proc *p)
2202 {
2203 	struct filedesc *fdp;
2204 
2205 	PROC_LOCK_ASSERT(p, MA_OWNED);
2206 	fdp = atomic_load_ptr(&p->p_fd);
2207 	if (fdp != NULL)
2208 		refcount_acquire(&fdp->fd_holdcnt);
2209 	return (fdp);
2210 }
2211 
2212 static struct pwddesc *
2213 pdhold(struct proc *p)
2214 {
2215 	struct pwddesc *pdp;
2216 
2217 	PROC_LOCK_ASSERT(p, MA_OWNED);
2218 	pdp = atomic_load_ptr(&p->p_pd);
2219 	if (pdp != NULL)
2220 		refcount_acquire(&pdp->pd_refcount);
2221 	return (pdp);
2222 }
2223 
2224 static void
2225 fddrop(struct filedesc *fdp)
2226 {
2227 
2228 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2229 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2230 			return;
2231 	}
2232 
2233 	FILEDESC_LOCK_DESTROY(fdp);
2234 	uma_zfree(filedesc0_zone, fdp);
2235 }
2236 
2237 static void
2238 pddrop(struct pwddesc *pdp)
2239 {
2240 	struct pwd *pwd;
2241 
2242 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2243 		return;
2244 
2245 	PWDDESC_XLOCK(pdp);
2246 	if (refcount_release(&pdp->pd_refcount) == 0) {
2247 		PWDDESC_XUNLOCK(pdp);
2248 		return;
2249 	}
2250 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2251 	pwd_set(pdp, NULL);
2252 	PWDDESC_XUNLOCK(pdp);
2253 	pwd_drop(pwd);
2254 
2255 	PWDDESC_LOCK_DESTROY(pdp);
2256 	free(pdp, M_PWDDESC);
2257 }
2258 
2259 /*
2260  * Share a filedesc structure.
2261  */
2262 struct filedesc *
2263 fdshare(struct filedesc *fdp)
2264 {
2265 
2266 	refcount_acquire(&fdp->fd_refcnt);
2267 	return (fdp);
2268 }
2269 
2270 /*
2271  * Share a pwddesc structure.
2272  */
2273 struct pwddesc *
2274 pdshare(struct pwddesc *pdp)
2275 {
2276 	refcount_acquire(&pdp->pd_refcount);
2277 	return (pdp);
2278 }
2279 
2280 /*
2281  * Unshare a filedesc structure, if necessary by making a copy
2282  */
2283 void
2284 fdunshare(struct thread *td)
2285 {
2286 	struct filedesc *tmp;
2287 	struct proc *p = td->td_proc;
2288 
2289 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2290 		return;
2291 
2292 	tmp = fdcopy(p->p_fd);
2293 	fdescfree(td);
2294 	p->p_fd = tmp;
2295 }
2296 
2297 /*
2298  * Unshare a pwddesc structure.
2299  */
2300 void
2301 pdunshare(struct thread *td)
2302 {
2303 	struct pwddesc *pdp;
2304 	struct proc *p;
2305 
2306 	p = td->td_proc;
2307 	/* Not shared. */
2308 	if (p->p_pd->pd_refcount == 1)
2309 		return;
2310 
2311 	pdp = pdcopy(p->p_pd);
2312 	pdescfree(td);
2313 	p->p_pd = pdp;
2314 }
2315 
2316 /*
2317  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2318  * this is to ease callers, not catch errors.
2319  */
2320 struct filedesc *
2321 fdcopy(struct filedesc *fdp)
2322 {
2323 	struct filedesc *newfdp;
2324 	struct filedescent *nfde, *ofde;
2325 	int i, lastfile;
2326 
2327 	MPASS(fdp != NULL);
2328 
2329 	newfdp = fdinit(fdp, true, &lastfile);
2330 	/* copy all passable descriptors (i.e. not kqueue) */
2331 	newfdp->fd_freefile = -1;
2332 	for (i = 0; i <= lastfile; ++i) {
2333 		ofde = &fdp->fd_ofiles[i];
2334 		if (ofde->fde_file == NULL ||
2335 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2336 		    !fhold(ofde->fde_file)) {
2337 			if (newfdp->fd_freefile == -1)
2338 				newfdp->fd_freefile = i;
2339 			continue;
2340 		}
2341 		nfde = &newfdp->fd_ofiles[i];
2342 		*nfde = *ofde;
2343 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2344 		fdused_init(newfdp, i);
2345 	}
2346 	if (newfdp->fd_freefile == -1)
2347 		newfdp->fd_freefile = i;
2348 	FILEDESC_SUNLOCK(fdp);
2349 	return (newfdp);
2350 }
2351 
2352 /*
2353  * Copy a pwddesc structure.
2354  */
2355 struct pwddesc *
2356 pdcopy(struct pwddesc *pdp)
2357 {
2358 	struct pwddesc *newpdp;
2359 
2360 	MPASS(pdp != NULL);
2361 
2362 	newpdp = pdinit(pdp, true);
2363 	newpdp->pd_cmask = pdp->pd_cmask;
2364 	PWDDESC_XUNLOCK(pdp);
2365 	return (newpdp);
2366 }
2367 
2368 /*
2369  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2370  * one of processes using it exits) and the table used to be shared.
2371  */
2372 static void
2373 fdclearlocks(struct thread *td)
2374 {
2375 	struct filedesc *fdp;
2376 	struct filedesc_to_leader *fdtol;
2377 	struct flock lf;
2378 	struct file *fp;
2379 	struct proc *p;
2380 	struct vnode *vp;
2381 	int i, lastfile;
2382 
2383 	p = td->td_proc;
2384 	fdp = p->p_fd;
2385 	fdtol = p->p_fdtol;
2386 	MPASS(fdtol != NULL);
2387 
2388 	FILEDESC_XLOCK(fdp);
2389 	KASSERT(fdtol->fdl_refcount > 0,
2390 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2391 	    fdtol->fdl_refcount));
2392 	if (fdtol->fdl_refcount == 1 &&
2393 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2394 		lastfile = fdlastfile(fdp);
2395 		for (i = 0; i <= lastfile; i++) {
2396 			fp = fdp->fd_ofiles[i].fde_file;
2397 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2398 			    !fhold(fp))
2399 				continue;
2400 			FILEDESC_XUNLOCK(fdp);
2401 			lf.l_whence = SEEK_SET;
2402 			lf.l_start = 0;
2403 			lf.l_len = 0;
2404 			lf.l_type = F_UNLCK;
2405 			vp = fp->f_vnode;
2406 			(void) VOP_ADVLOCK(vp,
2407 			    (caddr_t)p->p_leader, F_UNLCK,
2408 			    &lf, F_POSIX);
2409 			FILEDESC_XLOCK(fdp);
2410 			fdrop(fp, td);
2411 		}
2412 	}
2413 retry:
2414 	if (fdtol->fdl_refcount == 1) {
2415 		if (fdp->fd_holdleaderscount > 0 &&
2416 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2417 			/*
2418 			 * close() or kern_dup() has cleared a reference
2419 			 * in a shared file descriptor table.
2420 			 */
2421 			fdp->fd_holdleaderswakeup = 1;
2422 			sx_sleep(&fdp->fd_holdleaderscount,
2423 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2424 			goto retry;
2425 		}
2426 		if (fdtol->fdl_holdcount > 0) {
2427 			/*
2428 			 * Ensure that fdtol->fdl_leader remains
2429 			 * valid in closef().
2430 			 */
2431 			fdtol->fdl_wakeup = 1;
2432 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2433 			    "fdlhold", 0);
2434 			goto retry;
2435 		}
2436 	}
2437 	fdtol->fdl_refcount--;
2438 	if (fdtol->fdl_refcount == 0 &&
2439 	    fdtol->fdl_holdcount == 0) {
2440 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2441 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2442 	} else
2443 		fdtol = NULL;
2444 	p->p_fdtol = NULL;
2445 	FILEDESC_XUNLOCK(fdp);
2446 	if (fdtol != NULL)
2447 		free(fdtol, M_FILEDESC_TO_LEADER);
2448 }
2449 
2450 /*
2451  * Release a filedesc structure.
2452  */
2453 static void
2454 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2455 {
2456 	struct filedesc0 *fdp0;
2457 	struct freetable *ft, *tft;
2458 	struct filedescent *fde;
2459 	struct file *fp;
2460 	int i, lastfile;
2461 
2462 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2463 	    ("%s: fd table %p carries references", __func__, fdp));
2464 
2465 	/*
2466 	 * Serialize with threads iterating over the table, if any.
2467 	 */
2468 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2469 		FILEDESC_XLOCK(fdp);
2470 		FILEDESC_XUNLOCK(fdp);
2471 	}
2472 
2473 	lastfile = fdlastfile_single(fdp);
2474 	for (i = 0; i <= lastfile; i++) {
2475 		fde = &fdp->fd_ofiles[i];
2476 		fp = fde->fde_file;
2477 		if (fp != NULL) {
2478 			fdefree_last(fde);
2479 			(void) closef(fp, td);
2480 		}
2481 	}
2482 
2483 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2484 		free(fdp->fd_map, M_FILEDESC);
2485 	if (fdp->fd_nfiles > NDFILE)
2486 		free(fdp->fd_files, M_FILEDESC);
2487 
2488 	fdp0 = (struct filedesc0 *)fdp;
2489 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2490 		free(ft->ft_table, M_FILEDESC);
2491 
2492 	fddrop(fdp);
2493 }
2494 
2495 void
2496 fdescfree(struct thread *td)
2497 {
2498 	struct proc *p;
2499 	struct filedesc *fdp;
2500 
2501 	p = td->td_proc;
2502 	fdp = p->p_fd;
2503 	MPASS(fdp != NULL);
2504 
2505 #ifdef RACCT
2506 	if (RACCT_ENABLED())
2507 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2508 #endif
2509 
2510 	if (p->p_fdtol != NULL)
2511 		fdclearlocks(td);
2512 
2513 	/*
2514 	 * Check fdhold for an explanation.
2515 	 */
2516 	atomic_store_ptr(&p->p_fd, NULL);
2517 	atomic_thread_fence_seq_cst();
2518 	PROC_WAIT_UNLOCKED(p);
2519 
2520 	if (refcount_release(&fdp->fd_refcnt) == 0)
2521 		return;
2522 
2523 	fdescfree_fds(td, fdp);
2524 }
2525 
2526 void
2527 pdescfree(struct thread *td)
2528 {
2529 	struct proc *p;
2530 	struct pwddesc *pdp;
2531 
2532 	p = td->td_proc;
2533 	pdp = p->p_pd;
2534 	MPASS(pdp != NULL);
2535 
2536 	/*
2537 	 * Check pdhold for an explanation.
2538 	 */
2539 	atomic_store_ptr(&p->p_pd, NULL);
2540 	atomic_thread_fence_seq_cst();
2541 	PROC_WAIT_UNLOCKED(p);
2542 
2543 	pddrop(pdp);
2544 }
2545 
2546 /*
2547  * For setugid programs, we don't want to people to use that setugidness
2548  * to generate error messages which write to a file which otherwise would
2549  * otherwise be off-limits to the process.  We check for filesystems where
2550  * the vnode can change out from under us after execve (like [lin]procfs).
2551  *
2552  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2553  * sufficient.  We also don't check for setugidness since we know we are.
2554  */
2555 static bool
2556 is_unsafe(struct file *fp)
2557 {
2558 	struct vnode *vp;
2559 
2560 	if (fp->f_type != DTYPE_VNODE)
2561 		return (false);
2562 
2563 	vp = fp->f_vnode;
2564 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2565 }
2566 
2567 /*
2568  * Make this setguid thing safe, if at all possible.
2569  */
2570 void
2571 fdsetugidsafety(struct thread *td)
2572 {
2573 	struct filedesc *fdp;
2574 	struct file *fp;
2575 	int i;
2576 
2577 	fdp = td->td_proc->p_fd;
2578 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2579 	    ("the fdtable should not be shared"));
2580 	MPASS(fdp->fd_nfiles >= 3);
2581 	for (i = 0; i <= 2; i++) {
2582 		fp = fdp->fd_ofiles[i].fde_file;
2583 		if (fp != NULL && is_unsafe(fp)) {
2584 			FILEDESC_XLOCK(fdp);
2585 			knote_fdclose(td, i);
2586 			/*
2587 			 * NULL-out descriptor prior to close to avoid
2588 			 * a race while close blocks.
2589 			 */
2590 			fdfree(fdp, i);
2591 			FILEDESC_XUNLOCK(fdp);
2592 			(void) closef(fp, td);
2593 		}
2594 	}
2595 }
2596 
2597 /*
2598  * If a specific file object occupies a specific file descriptor, close the
2599  * file descriptor entry and drop a reference on the file object.  This is a
2600  * convenience function to handle a subsequent error in a function that calls
2601  * falloc() that handles the race that another thread might have closed the
2602  * file descriptor out from under the thread creating the file object.
2603  */
2604 void
2605 fdclose(struct thread *td, struct file *fp, int idx)
2606 {
2607 	struct filedesc *fdp = td->td_proc->p_fd;
2608 
2609 	FILEDESC_XLOCK(fdp);
2610 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2611 		fdfree(fdp, idx);
2612 		FILEDESC_XUNLOCK(fdp);
2613 		fdrop(fp, td);
2614 	} else
2615 		FILEDESC_XUNLOCK(fdp);
2616 }
2617 
2618 /*
2619  * Close any files on exec?
2620  */
2621 void
2622 fdcloseexec(struct thread *td)
2623 {
2624 	struct filedesc *fdp;
2625 	struct filedescent *fde;
2626 	struct file *fp;
2627 	int i, lastfile;
2628 
2629 	fdp = td->td_proc->p_fd;
2630 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2631 	    ("the fdtable should not be shared"));
2632 	lastfile = fdlastfile_single(fdp);
2633 	for (i = 0; i <= lastfile; i++) {
2634 		fde = &fdp->fd_ofiles[i];
2635 		fp = fde->fde_file;
2636 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2637 		    (fde->fde_flags & UF_EXCLOSE))) {
2638 			FILEDESC_XLOCK(fdp);
2639 			fdfree(fdp, i);
2640 			(void) closefp(fdp, i, fp, td, false, false);
2641 			FILEDESC_UNLOCK_ASSERT(fdp);
2642 		}
2643 	}
2644 }
2645 
2646 /*
2647  * It is unsafe for set[ug]id processes to be started with file
2648  * descriptors 0..2 closed, as these descriptors are given implicit
2649  * significance in the Standard C library.  fdcheckstd() will create a
2650  * descriptor referencing /dev/null for each of stdin, stdout, and
2651  * stderr that is not already open.
2652  */
2653 int
2654 fdcheckstd(struct thread *td)
2655 {
2656 	struct filedesc *fdp;
2657 	register_t save;
2658 	int i, error, devnull;
2659 
2660 	fdp = td->td_proc->p_fd;
2661 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2662 	    ("the fdtable should not be shared"));
2663 	MPASS(fdp->fd_nfiles >= 3);
2664 	devnull = -1;
2665 	for (i = 0; i <= 2; i++) {
2666 		if (fdp->fd_ofiles[i].fde_file != NULL)
2667 			continue;
2668 
2669 		save = td->td_retval[0];
2670 		if (devnull != -1) {
2671 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2672 		} else {
2673 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2674 			    UIO_SYSSPACE, O_RDWR, 0);
2675 			if (error == 0) {
2676 				devnull = td->td_retval[0];
2677 				KASSERT(devnull == i, ("we didn't get our fd"));
2678 			}
2679 		}
2680 		td->td_retval[0] = save;
2681 		if (error != 0)
2682 			return (error);
2683 	}
2684 	return (0);
2685 }
2686 
2687 /*
2688  * Internal form of close.  Decrement reference count on file structure.
2689  * Note: td may be NULL when closing a file that was being passed in a
2690  * message.
2691  */
2692 int
2693 closef(struct file *fp, struct thread *td)
2694 {
2695 	struct vnode *vp;
2696 	struct flock lf;
2697 	struct filedesc_to_leader *fdtol;
2698 	struct filedesc *fdp;
2699 
2700 	MPASS(td != NULL);
2701 
2702 	/*
2703 	 * POSIX record locking dictates that any close releases ALL
2704 	 * locks owned by this process.  This is handled by setting
2705 	 * a flag in the unlock to free ONLY locks obeying POSIX
2706 	 * semantics, and not to free BSD-style file locks.
2707 	 * If the descriptor was in a message, POSIX-style locks
2708 	 * aren't passed with the descriptor, and the thread pointer
2709 	 * will be NULL.  Callers should be careful only to pass a
2710 	 * NULL thread pointer when there really is no owning
2711 	 * context that might have locks, or the locks will be
2712 	 * leaked.
2713 	 */
2714 	if (fp->f_type == DTYPE_VNODE) {
2715 		vp = fp->f_vnode;
2716 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2717 			lf.l_whence = SEEK_SET;
2718 			lf.l_start = 0;
2719 			lf.l_len = 0;
2720 			lf.l_type = F_UNLCK;
2721 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2722 			    F_UNLCK, &lf, F_POSIX);
2723 		}
2724 		fdtol = td->td_proc->p_fdtol;
2725 		if (fdtol != NULL) {
2726 			/*
2727 			 * Handle special case where file descriptor table is
2728 			 * shared between multiple process leaders.
2729 			 */
2730 			fdp = td->td_proc->p_fd;
2731 			FILEDESC_XLOCK(fdp);
2732 			for (fdtol = fdtol->fdl_next;
2733 			    fdtol != td->td_proc->p_fdtol;
2734 			    fdtol = fdtol->fdl_next) {
2735 				if ((fdtol->fdl_leader->p_flag &
2736 				    P_ADVLOCK) == 0)
2737 					continue;
2738 				fdtol->fdl_holdcount++;
2739 				FILEDESC_XUNLOCK(fdp);
2740 				lf.l_whence = SEEK_SET;
2741 				lf.l_start = 0;
2742 				lf.l_len = 0;
2743 				lf.l_type = F_UNLCK;
2744 				vp = fp->f_vnode;
2745 				(void) VOP_ADVLOCK(vp,
2746 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2747 				    F_POSIX);
2748 				FILEDESC_XLOCK(fdp);
2749 				fdtol->fdl_holdcount--;
2750 				if (fdtol->fdl_holdcount == 0 &&
2751 				    fdtol->fdl_wakeup != 0) {
2752 					fdtol->fdl_wakeup = 0;
2753 					wakeup(fdtol);
2754 				}
2755 			}
2756 			FILEDESC_XUNLOCK(fdp);
2757 		}
2758 	}
2759 	return (fdrop_close(fp, td));
2760 }
2761 
2762 /*
2763  * Hack for file descriptor passing code.
2764  */
2765 void
2766 closef_nothread(struct file *fp)
2767 {
2768 
2769 	fdrop(fp, NULL);
2770 }
2771 
2772 /*
2773  * Initialize the file pointer with the specified properties.
2774  *
2775  * The ops are set with release semantics to be certain that the flags, type,
2776  * and data are visible when ops is.  This is to prevent ops methods from being
2777  * called with bad data.
2778  */
2779 void
2780 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2781 {
2782 	fp->f_data = data;
2783 	fp->f_flag = flag;
2784 	fp->f_type = type;
2785 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2786 }
2787 
2788 void
2789 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2790 {
2791 	fp->f_seqcount[UIO_READ] = 1;
2792 	fp->f_seqcount[UIO_WRITE] = 1;
2793 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2794 	    data, ops);
2795 }
2796 
2797 int
2798 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2799     struct file **fpp, struct filecaps *havecapsp)
2800 {
2801 	struct filedescent *fde;
2802 	int error;
2803 
2804 	FILEDESC_LOCK_ASSERT(fdp);
2805 
2806 	fde = fdeget_locked(fdp, fd);
2807 	if (fde == NULL) {
2808 		error = EBADF;
2809 		goto out;
2810 	}
2811 
2812 #ifdef CAPABILITIES
2813 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2814 	if (error != 0)
2815 		goto out;
2816 #endif
2817 
2818 	if (havecapsp != NULL)
2819 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2820 
2821 	*fpp = fde->fde_file;
2822 
2823 	error = 0;
2824 out:
2825 	return (error);
2826 }
2827 
2828 int
2829 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2830     struct file **fpp, struct filecaps *havecapsp)
2831 {
2832 	struct filedesc *fdp = td->td_proc->p_fd;
2833 	int error;
2834 #ifndef CAPABILITIES
2835 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2836 	if (havecapsp != NULL && error == 0)
2837 		filecaps_fill(havecapsp);
2838 #else
2839 	struct file *fp;
2840 	seqc_t seq;
2841 
2842 	*fpp = NULL;
2843 	for (;;) {
2844 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2845 		if (error != 0)
2846 			return (error);
2847 
2848 		if (havecapsp != NULL) {
2849 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2850 			    havecapsp, false)) {
2851 				fdrop(fp, td);
2852 				goto get_locked;
2853 			}
2854 		}
2855 
2856 		if (!fd_modified(fdp, fd, seq))
2857 			break;
2858 		fdrop(fp, td);
2859 	}
2860 
2861 	*fpp = fp;
2862 	return (0);
2863 
2864 get_locked:
2865 	FILEDESC_SLOCK(fdp);
2866 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2867 	if (error == 0 && !fhold(*fpp))
2868 		error = EBADF;
2869 	FILEDESC_SUNLOCK(fdp);
2870 #endif
2871 	return (error);
2872 }
2873 
2874 #ifdef CAPABILITIES
2875 int
2876 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2877 {
2878 	const struct filedescent *fde;
2879 	const struct fdescenttbl *fdt;
2880 	struct filedesc *fdp;
2881 	struct file *fp;
2882 	struct vnode *vp;
2883 	const cap_rights_t *haverights;
2884 	cap_rights_t rights;
2885 	seqc_t seq;
2886 
2887 	VFS_SMR_ASSERT_ENTERED();
2888 
2889 	rights = *ndp->ni_rightsneeded;
2890 	cap_rights_set_one(&rights, CAP_LOOKUP);
2891 
2892 	fdp = curproc->p_fd;
2893 	fdt = fdp->fd_files;
2894 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2895 		return (EBADF);
2896 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2897 	fde = &fdt->fdt_ofiles[fd];
2898 	haverights = cap_rights_fde_inline(fde);
2899 	fp = fde->fde_file;
2900 	if (__predict_false(fp == NULL))
2901 		return (EAGAIN);
2902 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2903 		return (EAGAIN);
2904 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2905 	vp = fp->f_vnode;
2906 	if (__predict_false(vp == NULL)) {
2907 		return (EAGAIN);
2908 	}
2909 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2910 		return (EAGAIN);
2911 	}
2912 	/*
2913 	 * Use an acquire barrier to force re-reading of fdt so it is
2914 	 * refreshed for verification.
2915 	 */
2916 	atomic_thread_fence_acq();
2917 	fdt = fdp->fd_files;
2918 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2919 		return (EAGAIN);
2920 	/*
2921 	 * If file descriptor doesn't have all rights,
2922 	 * all lookups relative to it must also be
2923 	 * strictly relative.
2924 	 *
2925 	 * Not yet supported by fast path.
2926 	 */
2927 	CAP_ALL(&rights);
2928 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2929 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2930 	    ndp->ni_filecaps.fc_nioctls != -1) {
2931 #ifdef notyet
2932 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2933 #else
2934 		return (EAGAIN);
2935 #endif
2936 	}
2937 	*vpp = vp;
2938 	return (0);
2939 }
2940 #else
2941 int
2942 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2943 {
2944 	const struct fdescenttbl *fdt;
2945 	struct filedesc *fdp;
2946 	struct file *fp;
2947 	struct vnode *vp;
2948 
2949 	VFS_SMR_ASSERT_ENTERED();
2950 
2951 	fdp = curproc->p_fd;
2952 	fdt = fdp->fd_files;
2953 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2954 		return (EBADF);
2955 	fp = fdt->fdt_ofiles[fd].fde_file;
2956 	if (__predict_false(fp == NULL))
2957 		return (EAGAIN);
2958 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2959 	vp = fp->f_vnode;
2960 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2961 		return (EAGAIN);
2962 	}
2963 	/*
2964 	 * Use an acquire barrier to force re-reading of fdt so it is
2965 	 * refreshed for verification.
2966 	 */
2967 	atomic_thread_fence_acq();
2968 	fdt = fdp->fd_files;
2969 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2970 		return (EAGAIN);
2971 	filecaps_fill(&ndp->ni_filecaps);
2972 	*vpp = vp;
2973 	return (0);
2974 }
2975 #endif
2976 
2977 int
2978 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2979     struct file **fpp, seqc_t *seqp)
2980 {
2981 #ifdef CAPABILITIES
2982 	const struct filedescent *fde;
2983 #endif
2984 	const struct fdescenttbl *fdt;
2985 	struct file *fp;
2986 #ifdef CAPABILITIES
2987 	seqc_t seq;
2988 	cap_rights_t haverights;
2989 	int error;
2990 #endif
2991 
2992 	fdt = fdp->fd_files;
2993 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2994 		return (EBADF);
2995 	/*
2996 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2997 	 * never raising a refcount above 0.  To accomplish this we have
2998 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2999 	 * must be re-verified once we acquire a reference to be certain
3000 	 * that the identity is still correct and we did not lose a race
3001 	 * due to preemption.
3002 	 */
3003 	for (;;) {
3004 #ifdef CAPABILITIES
3005 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3006 		fde = &fdt->fdt_ofiles[fd];
3007 		haverights = *cap_rights_fde_inline(fde);
3008 		fp = fde->fde_file;
3009 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3010 			continue;
3011 #else
3012 		fp = fdt->fdt_ofiles[fd].fde_file;
3013 #endif
3014 		if (fp == NULL)
3015 			return (EBADF);
3016 #ifdef CAPABILITIES
3017 		error = cap_check_inline(&haverights, needrightsp);
3018 		if (error != 0)
3019 			return (error);
3020 #endif
3021 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3022 			/*
3023 			 * Force a reload. Other thread could reallocate the
3024 			 * table before this fd was closed, so it is possible
3025 			 * that there is a stale fp pointer in cached version.
3026 			 */
3027 			fdt = atomic_load_ptr(&fdp->fd_files);
3028 			continue;
3029 		}
3030 		/*
3031 		 * Use an acquire barrier to force re-reading of fdt so it is
3032 		 * refreshed for verification.
3033 		 */
3034 		atomic_thread_fence_acq();
3035 		fdt = fdp->fd_files;
3036 #ifdef	CAPABILITIES
3037 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3038 #else
3039 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3040 #endif
3041 			break;
3042 		fdrop(fp, curthread);
3043 	}
3044 	*fpp = fp;
3045 	if (seqp != NULL) {
3046 #ifdef CAPABILITIES
3047 		*seqp = seq;
3048 #endif
3049 	}
3050 	return (0);
3051 }
3052 
3053 /*
3054  * See the comments in fget_unlocked_seq for an explanation of how this works.
3055  *
3056  * This is a simplified variant which bails out to the aforementioned routine
3057  * if anything goes wrong. In practice this only happens when userspace is
3058  * racing with itself.
3059  */
3060 int
3061 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3062     struct file **fpp)
3063 {
3064 #ifdef CAPABILITIES
3065 	const struct filedescent *fde;
3066 #endif
3067 	const struct fdescenttbl *fdt;
3068 	struct file *fp;
3069 #ifdef CAPABILITIES
3070 	seqc_t seq;
3071 	const cap_rights_t *haverights;
3072 #endif
3073 
3074 	fdt = fdp->fd_files;
3075 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3076 		return (EBADF);
3077 #ifdef CAPABILITIES
3078 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3079 	fde = &fdt->fdt_ofiles[fd];
3080 	haverights = cap_rights_fde_inline(fde);
3081 	fp = fde->fde_file;
3082 #else
3083 	fp = fdt->fdt_ofiles[fd].fde_file;
3084 #endif
3085 	if (__predict_false(fp == NULL))
3086 		goto out_fallback;
3087 #ifdef CAPABILITIES
3088 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3089 		goto out_fallback;
3090 #endif
3091 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3092 		goto out_fallback;
3093 
3094 	/*
3095 	 * Use an acquire barrier to force re-reading of fdt so it is
3096 	 * refreshed for verification.
3097 	 */
3098 	atomic_thread_fence_acq();
3099 	fdt = fdp->fd_files;
3100 #ifdef	CAPABILITIES
3101 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3102 #else
3103 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3104 #endif
3105 		goto out_fdrop;
3106 	*fpp = fp;
3107 	return (0);
3108 out_fdrop:
3109 	fdrop(fp, curthread);
3110 out_fallback:
3111 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3112 }
3113 
3114 /*
3115  * Translate fd -> file when the caller guarantees the file descriptor table
3116  * can't be changed by others.
3117  *
3118  * Note this does not mean the file object itself is only visible to the caller,
3119  * merely that it wont disappear without having to be referenced.
3120  *
3121  * Must be paired with fput_only_user.
3122  */
3123 #ifdef	CAPABILITIES
3124 int
3125 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3126     struct file **fpp)
3127 {
3128 	const struct filedescent *fde;
3129 	const struct fdescenttbl *fdt;
3130 	const cap_rights_t *haverights;
3131 	struct file *fp;
3132 	int error;
3133 
3134 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3135 
3136 	if (__predict_false(fd >= fdp->fd_nfiles))
3137 		return (EBADF);
3138 
3139 	fdt = fdp->fd_files;
3140 	fde = &fdt->fdt_ofiles[fd];
3141 	fp = fde->fde_file;
3142 	if (__predict_false(fp == NULL))
3143 		return (EBADF);
3144 	MPASS(refcount_load(&fp->f_count) > 0);
3145 	haverights = cap_rights_fde_inline(fde);
3146 	error = cap_check_inline(haverights, needrightsp);
3147 	if (__predict_false(error != 0))
3148 		return (error);
3149 	*fpp = fp;
3150 	return (0);
3151 }
3152 #else
3153 int
3154 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3155     struct file **fpp)
3156 {
3157 	struct file *fp;
3158 
3159 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3160 
3161 	if (__predict_false(fd >= fdp->fd_nfiles))
3162 		return (EBADF);
3163 
3164 	fp = fdp->fd_ofiles[fd].fde_file;
3165 	if (__predict_false(fp == NULL))
3166 		return (EBADF);
3167 
3168 	MPASS(refcount_load(&fp->f_count) > 0);
3169 	*fpp = fp;
3170 	return (0);
3171 }
3172 #endif
3173 
3174 /*
3175  * Extract the file pointer associated with the specified descriptor for the
3176  * current user process.
3177  *
3178  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3179  * returned.
3180  *
3181  * File's rights will be checked against the capability rights mask.
3182  *
3183  * If an error occurred the non-zero error is returned and *fpp is set to
3184  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3185  * responsible for fdrop().
3186  */
3187 static __inline int
3188 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3189     cap_rights_t *needrightsp)
3190 {
3191 	struct filedesc *fdp;
3192 	struct file *fp;
3193 	int error;
3194 
3195 	*fpp = NULL;
3196 	fdp = td->td_proc->p_fd;
3197 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3198 	if (__predict_false(error != 0))
3199 		return (error);
3200 	if (__predict_false(fp->f_ops == &badfileops)) {
3201 		fdrop(fp, td);
3202 		return (EBADF);
3203 	}
3204 
3205 	/*
3206 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3207 	 */
3208 	error = 0;
3209 	switch (flags) {
3210 	case FREAD:
3211 	case FWRITE:
3212 		if ((fp->f_flag & flags) == 0)
3213 			error = EBADF;
3214 		break;
3215 	case FEXEC:
3216 		if (fp->f_ops != &path_fileops &&
3217 		    ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3218 		    (fp->f_flag & FWRITE) != 0))
3219 			error = EBADF;
3220 		break;
3221 	case 0:
3222 		break;
3223 	default:
3224 		KASSERT(0, ("wrong flags"));
3225 	}
3226 
3227 	if (error != 0) {
3228 		fdrop(fp, td);
3229 		return (error);
3230 	}
3231 
3232 	*fpp = fp;
3233 	return (0);
3234 }
3235 
3236 int
3237 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3238 {
3239 
3240 	return (_fget(td, fd, fpp, 0, rightsp));
3241 }
3242 
3243 int
3244 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3245     struct file **fpp)
3246 {
3247 	int error;
3248 #ifndef CAPABILITIES
3249 	error = _fget(td, fd, fpp, 0, rightsp);
3250 	if (maxprotp != NULL)
3251 		*maxprotp = VM_PROT_ALL;
3252 	return (error);
3253 #else
3254 	cap_rights_t fdrights;
3255 	struct filedesc *fdp;
3256 	struct file *fp;
3257 	seqc_t seq;
3258 
3259 	*fpp = NULL;
3260 	fdp = td->td_proc->p_fd;
3261 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3262 	for (;;) {
3263 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3264 		if (__predict_false(error != 0))
3265 			return (error);
3266 		if (__predict_false(fp->f_ops == &badfileops)) {
3267 			fdrop(fp, td);
3268 			return (EBADF);
3269 		}
3270 		if (maxprotp != NULL)
3271 			fdrights = *cap_rights(fdp, fd);
3272 		if (!fd_modified(fdp, fd, seq))
3273 			break;
3274 		fdrop(fp, td);
3275 	}
3276 
3277 	/*
3278 	 * If requested, convert capability rights to access flags.
3279 	 */
3280 	if (maxprotp != NULL)
3281 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3282 	*fpp = fp;
3283 	return (0);
3284 #endif
3285 }
3286 
3287 int
3288 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3289 {
3290 
3291 	return (_fget(td, fd, fpp, FREAD, rightsp));
3292 }
3293 
3294 int
3295 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3296 {
3297 
3298 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3299 }
3300 
3301 int
3302 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3303     struct file **fpp)
3304 {
3305 	struct filedesc *fdp = td->td_proc->p_fd;
3306 #ifndef CAPABILITIES
3307 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3308 #else
3309 	struct file *fp;
3310 	int error;
3311 	seqc_t seq;
3312 
3313 	*fpp = NULL;
3314 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3315 	for (;;) {
3316 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3317 		if (error != 0)
3318 			return (error);
3319 		error = cap_fcntl_check(fdp, fd, needfcntl);
3320 		if (!fd_modified(fdp, fd, seq))
3321 			break;
3322 		fdrop(fp, td);
3323 	}
3324 	if (error != 0) {
3325 		fdrop(fp, td);
3326 		return (error);
3327 	}
3328 	*fpp = fp;
3329 	return (0);
3330 #endif
3331 }
3332 
3333 /*
3334  * Like fget() but loads the underlying vnode, or returns an error if the
3335  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3336  * never have VM objects.  The returned vnode will be vref()'d.
3337  *
3338  * XXX: what about the unused flags ?
3339  */
3340 static __inline int
3341 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3342     struct vnode **vpp)
3343 {
3344 	struct file *fp;
3345 	int error;
3346 
3347 	*vpp = NULL;
3348 	error = _fget(td, fd, &fp, flags, needrightsp);
3349 	if (error != 0)
3350 		return (error);
3351 	if (fp->f_vnode == NULL) {
3352 		error = EINVAL;
3353 	} else {
3354 		*vpp = fp->f_vnode;
3355 		vref(*vpp);
3356 	}
3357 	fdrop(fp, td);
3358 
3359 	return (error);
3360 }
3361 
3362 int
3363 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3364 {
3365 
3366 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3367 }
3368 
3369 int
3370 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3371     struct filecaps *havecaps, struct vnode **vpp)
3372 {
3373 	struct filecaps caps;
3374 	struct file *fp;
3375 	int error;
3376 
3377 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3378 	if (error != 0)
3379 		return (error);
3380 	if (fp->f_ops == &badfileops) {
3381 		error = EBADF;
3382 		goto out;
3383 	}
3384 	if (fp->f_vnode == NULL) {
3385 		error = EINVAL;
3386 		goto out;
3387 	}
3388 
3389 	*havecaps = caps;
3390 	*vpp = fp->f_vnode;
3391 	vref(*vpp);
3392 	fdrop(fp, td);
3393 
3394 	return (0);
3395 out:
3396 	filecaps_free(&caps);
3397 	fdrop(fp, td);
3398 	return (error);
3399 }
3400 
3401 int
3402 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3403 {
3404 
3405 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3406 }
3407 
3408 int
3409 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3410 {
3411 
3412 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3413 }
3414 
3415 #ifdef notyet
3416 int
3417 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3418     struct vnode **vpp)
3419 {
3420 
3421 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3422 }
3423 #endif
3424 
3425 /*
3426  * Handle the last reference to a file being closed.
3427  *
3428  * Without the noinline attribute clang keeps inlining the func thorough this
3429  * file when fdrop is used.
3430  */
3431 int __noinline
3432 _fdrop(struct file *fp, struct thread *td)
3433 {
3434 	int error;
3435 #ifdef INVARIANTS
3436 	int count;
3437 
3438 	count = refcount_load(&fp->f_count);
3439 	if (count != 0)
3440 		panic("fdrop: fp %p count %d", fp, count);
3441 #endif
3442 	error = fo_close(fp, td);
3443 	atomic_subtract_int(&openfiles, 1);
3444 	crfree(fp->f_cred);
3445 	free(fp->f_advice, M_FADVISE);
3446 	uma_zfree(file_zone, fp);
3447 
3448 	return (error);
3449 }
3450 
3451 /*
3452  * Apply an advisory lock on a file descriptor.
3453  *
3454  * Just attempt to get a record lock of the requested type on the entire file
3455  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3456  */
3457 #ifndef _SYS_SYSPROTO_H_
3458 struct flock_args {
3459 	int	fd;
3460 	int	how;
3461 };
3462 #endif
3463 /* ARGSUSED */
3464 int
3465 sys_flock(struct thread *td, struct flock_args *uap)
3466 {
3467 	struct file *fp;
3468 	struct vnode *vp;
3469 	struct flock lf;
3470 	int error;
3471 
3472 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3473 	if (error != 0)
3474 		return (error);
3475 	error = EOPNOTSUPP;
3476 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3477 		goto done;
3478 	}
3479 	if (fp->f_ops == &path_fileops) {
3480 		goto done;
3481 	}
3482 
3483 	error = 0;
3484 	vp = fp->f_vnode;
3485 	lf.l_whence = SEEK_SET;
3486 	lf.l_start = 0;
3487 	lf.l_len = 0;
3488 	if (uap->how & LOCK_UN) {
3489 		lf.l_type = F_UNLCK;
3490 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3491 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3492 		goto done;
3493 	}
3494 	if (uap->how & LOCK_EX)
3495 		lf.l_type = F_WRLCK;
3496 	else if (uap->how & LOCK_SH)
3497 		lf.l_type = F_RDLCK;
3498 	else {
3499 		error = EBADF;
3500 		goto done;
3501 	}
3502 	atomic_set_int(&fp->f_flag, FHASLOCK);
3503 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3504 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3505 done:
3506 	fdrop(fp, td);
3507 	return (error);
3508 }
3509 /*
3510  * Duplicate the specified descriptor to a free descriptor.
3511  */
3512 int
3513 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3514     int openerror, int *indxp)
3515 {
3516 	struct filedescent *newfde, *oldfde;
3517 	struct file *fp;
3518 	u_long *ioctls;
3519 	int error, indx;
3520 
3521 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3522 	    ("unexpected error %d in %s", openerror, __func__));
3523 
3524 	/*
3525 	 * If the to-be-dup'd fd number is greater than the allowed number
3526 	 * of file descriptors, or the fd to be dup'd has already been
3527 	 * closed, then reject.
3528 	 */
3529 	FILEDESC_XLOCK(fdp);
3530 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3531 		FILEDESC_XUNLOCK(fdp);
3532 		return (EBADF);
3533 	}
3534 
3535 	error = fdalloc(td, 0, &indx);
3536 	if (error != 0) {
3537 		FILEDESC_XUNLOCK(fdp);
3538 		return (error);
3539 	}
3540 
3541 	/*
3542 	 * There are two cases of interest here.
3543 	 *
3544 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3545 	 *
3546 	 * For ENXIO steal away the file structure from (dfd) and store it in
3547 	 * (indx).  (dfd) is effectively closed by this operation.
3548 	 */
3549 	switch (openerror) {
3550 	case ENODEV:
3551 		/*
3552 		 * Check that the mode the file is being opened for is a
3553 		 * subset of the mode of the existing descriptor.
3554 		 */
3555 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3556 			fdunused(fdp, indx);
3557 			FILEDESC_XUNLOCK(fdp);
3558 			return (EACCES);
3559 		}
3560 		if (!fhold(fp)) {
3561 			fdunused(fdp, indx);
3562 			FILEDESC_XUNLOCK(fdp);
3563 			return (EBADF);
3564 		}
3565 		newfde = &fdp->fd_ofiles[indx];
3566 		oldfde = &fdp->fd_ofiles[dfd];
3567 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3568 #ifdef CAPABILITIES
3569 		seqc_write_begin(&newfde->fde_seqc);
3570 #endif
3571 		memcpy(newfde, oldfde, fde_change_size);
3572 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3573 		    ioctls);
3574 #ifdef CAPABILITIES
3575 		seqc_write_end(&newfde->fde_seqc);
3576 #endif
3577 		break;
3578 	case ENXIO:
3579 		/*
3580 		 * Steal away the file pointer from dfd and stuff it into indx.
3581 		 */
3582 		newfde = &fdp->fd_ofiles[indx];
3583 		oldfde = &fdp->fd_ofiles[dfd];
3584 #ifdef CAPABILITIES
3585 		seqc_write_begin(&newfde->fde_seqc);
3586 #endif
3587 		memcpy(newfde, oldfde, fde_change_size);
3588 		oldfde->fde_file = NULL;
3589 		fdunused(fdp, dfd);
3590 #ifdef CAPABILITIES
3591 		seqc_write_end(&newfde->fde_seqc);
3592 #endif
3593 		break;
3594 	}
3595 	FILEDESC_XUNLOCK(fdp);
3596 	*indxp = indx;
3597 	return (0);
3598 }
3599 
3600 /*
3601  * This sysctl determines if we will allow a process to chroot(2) if it
3602  * has a directory open:
3603  *	0: disallowed for all processes.
3604  *	1: allowed for processes that were not already chroot(2)'ed.
3605  *	2: allowed for all processes.
3606  */
3607 
3608 static int chroot_allow_open_directories = 1;
3609 
3610 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3611     &chroot_allow_open_directories, 0,
3612     "Allow a process to chroot(2) if it has a directory open");
3613 
3614 /*
3615  * Helper function for raised chroot(2) security function:  Refuse if
3616  * any filedescriptors are open directories.
3617  */
3618 static int
3619 chroot_refuse_vdir_fds(struct filedesc *fdp)
3620 {
3621 	struct vnode *vp;
3622 	struct file *fp;
3623 	int fd, lastfile;
3624 
3625 	FILEDESC_LOCK_ASSERT(fdp);
3626 
3627 	lastfile = fdlastfile(fdp);
3628 	for (fd = 0; fd <= lastfile; fd++) {
3629 		fp = fget_locked(fdp, fd);
3630 		if (fp == NULL)
3631 			continue;
3632 		if (fp->f_type == DTYPE_VNODE) {
3633 			vp = fp->f_vnode;
3634 			if (vp->v_type == VDIR)
3635 				return (EPERM);
3636 		}
3637 	}
3638 	return (0);
3639 }
3640 
3641 static void
3642 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3643 {
3644 
3645 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3646 		vrefact(oldpwd->pwd_cdir);
3647 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3648 	}
3649 
3650 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3651 		vrefact(oldpwd->pwd_rdir);
3652 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3653 	}
3654 
3655 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3656 		vrefact(oldpwd->pwd_jdir);
3657 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3658 	}
3659 }
3660 
3661 struct pwd *
3662 pwd_hold_pwddesc(struct pwddesc *pdp)
3663 {
3664 	struct pwd *pwd;
3665 
3666 	PWDDESC_ASSERT_XLOCKED(pdp);
3667 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3668 	if (pwd != NULL)
3669 		refcount_acquire(&pwd->pwd_refcount);
3670 	return (pwd);
3671 }
3672 
3673 bool
3674 pwd_hold_smr(struct pwd *pwd)
3675 {
3676 
3677 	MPASS(pwd != NULL);
3678 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3679 		return (true);
3680 	}
3681 	return (false);
3682 }
3683 
3684 struct pwd *
3685 pwd_hold(struct thread *td)
3686 {
3687 	struct pwddesc *pdp;
3688 	struct pwd *pwd;
3689 
3690 	pdp = td->td_proc->p_pd;
3691 
3692 	vfs_smr_enter();
3693 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3694 	if (pwd_hold_smr(pwd)) {
3695 		vfs_smr_exit();
3696 		return (pwd);
3697 	}
3698 	vfs_smr_exit();
3699 	PWDDESC_XLOCK(pdp);
3700 	pwd = pwd_hold_pwddesc(pdp);
3701 	MPASS(pwd != NULL);
3702 	PWDDESC_XUNLOCK(pdp);
3703 	return (pwd);
3704 }
3705 
3706 struct pwd *
3707 pwd_hold_proc(struct proc *p)
3708 {
3709 	struct pwddesc *pdp;
3710 	struct pwd *pwd;
3711 
3712 	PROC_ASSERT_HELD(p);
3713 	PROC_LOCK(p);
3714 	pdp = pdhold(p);
3715 	MPASS(pdp != NULL);
3716 	PROC_UNLOCK(p);
3717 
3718 	PWDDESC_XLOCK(pdp);
3719 	pwd = pwd_hold_pwddesc(pdp);
3720 	MPASS(pwd != NULL);
3721 	PWDDESC_XUNLOCK(pdp);
3722 	pddrop(pdp);
3723 	return (pwd);
3724 }
3725 
3726 static struct pwd *
3727 pwd_alloc(void)
3728 {
3729 	struct pwd *pwd;
3730 
3731 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3732 	bzero(pwd, sizeof(*pwd));
3733 	refcount_init(&pwd->pwd_refcount, 1);
3734 	return (pwd);
3735 }
3736 
3737 void
3738 pwd_drop(struct pwd *pwd)
3739 {
3740 
3741 	if (!refcount_release(&pwd->pwd_refcount))
3742 		return;
3743 
3744 	if (pwd->pwd_cdir != NULL)
3745 		vrele(pwd->pwd_cdir);
3746 	if (pwd->pwd_rdir != NULL)
3747 		vrele(pwd->pwd_rdir);
3748 	if (pwd->pwd_jdir != NULL)
3749 		vrele(pwd->pwd_jdir);
3750 	uma_zfree_smr(pwd_zone, pwd);
3751 }
3752 
3753 /*
3754 * The caller is responsible for invoking priv_check() and
3755 * mac_vnode_check_chroot() to authorize this operation.
3756 */
3757 int
3758 pwd_chroot(struct thread *td, struct vnode *vp)
3759 {
3760 	struct pwddesc *pdp;
3761 	struct filedesc *fdp;
3762 	struct pwd *newpwd, *oldpwd;
3763 	int error;
3764 
3765 	fdp = td->td_proc->p_fd;
3766 	pdp = td->td_proc->p_pd;
3767 	newpwd = pwd_alloc();
3768 	FILEDESC_SLOCK(fdp);
3769 	PWDDESC_XLOCK(pdp);
3770 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3771 	if (chroot_allow_open_directories == 0 ||
3772 	    (chroot_allow_open_directories == 1 &&
3773 	    oldpwd->pwd_rdir != rootvnode)) {
3774 		error = chroot_refuse_vdir_fds(fdp);
3775 		FILEDESC_SUNLOCK(fdp);
3776 		if (error != 0) {
3777 			PWDDESC_XUNLOCK(pdp);
3778 			pwd_drop(newpwd);
3779 			return (error);
3780 		}
3781 	} else {
3782 		FILEDESC_SUNLOCK(fdp);
3783 	}
3784 
3785 	vrefact(vp);
3786 	newpwd->pwd_rdir = vp;
3787 	if (oldpwd->pwd_jdir == NULL) {
3788 		vrefact(vp);
3789 		newpwd->pwd_jdir = vp;
3790 	}
3791 	pwd_fill(oldpwd, newpwd);
3792 	pwd_set(pdp, newpwd);
3793 	PWDDESC_XUNLOCK(pdp);
3794 	pwd_drop(oldpwd);
3795 	return (0);
3796 }
3797 
3798 void
3799 pwd_chdir(struct thread *td, struct vnode *vp)
3800 {
3801 	struct pwddesc *pdp;
3802 	struct pwd *newpwd, *oldpwd;
3803 
3804 	VNPASS(vp->v_usecount > 0, vp);
3805 
3806 	newpwd = pwd_alloc();
3807 	pdp = td->td_proc->p_pd;
3808 	PWDDESC_XLOCK(pdp);
3809 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3810 	newpwd->pwd_cdir = vp;
3811 	pwd_fill(oldpwd, newpwd);
3812 	pwd_set(pdp, newpwd);
3813 	PWDDESC_XUNLOCK(pdp);
3814 	pwd_drop(oldpwd);
3815 }
3816 
3817 /*
3818  * jail_attach(2) changes both root and working directories.
3819  */
3820 int
3821 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3822 {
3823 	struct pwddesc *pdp;
3824 	struct filedesc *fdp;
3825 	struct pwd *newpwd, *oldpwd;
3826 	int error;
3827 
3828 	fdp = td->td_proc->p_fd;
3829 	pdp = td->td_proc->p_pd;
3830 	newpwd = pwd_alloc();
3831 	FILEDESC_SLOCK(fdp);
3832 	PWDDESC_XLOCK(pdp);
3833 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3834 	error = chroot_refuse_vdir_fds(fdp);
3835 	FILEDESC_SUNLOCK(fdp);
3836 	if (error != 0) {
3837 		PWDDESC_XUNLOCK(pdp);
3838 		pwd_drop(newpwd);
3839 		return (error);
3840 	}
3841 
3842 	vrefact(vp);
3843 	newpwd->pwd_rdir = vp;
3844 	vrefact(vp);
3845 	newpwd->pwd_cdir = vp;
3846 	if (oldpwd->pwd_jdir == NULL) {
3847 		vrefact(vp);
3848 		newpwd->pwd_jdir = vp;
3849 	}
3850 	pwd_fill(oldpwd, newpwd);
3851 	pwd_set(pdp, newpwd);
3852 	PWDDESC_XUNLOCK(pdp);
3853 	pwd_drop(oldpwd);
3854 	return (0);
3855 }
3856 
3857 void
3858 pwd_ensure_dirs(void)
3859 {
3860 	struct pwddesc *pdp;
3861 	struct pwd *oldpwd, *newpwd;
3862 
3863 	pdp = curproc->p_pd;
3864 	PWDDESC_XLOCK(pdp);
3865 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3866 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3867 		PWDDESC_XUNLOCK(pdp);
3868 		return;
3869 	}
3870 	PWDDESC_XUNLOCK(pdp);
3871 
3872 	newpwd = pwd_alloc();
3873 	PWDDESC_XLOCK(pdp);
3874 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3875 	pwd_fill(oldpwd, newpwd);
3876 	if (newpwd->pwd_cdir == NULL) {
3877 		vrefact(rootvnode);
3878 		newpwd->pwd_cdir = rootvnode;
3879 	}
3880 	if (newpwd->pwd_rdir == NULL) {
3881 		vrefact(rootvnode);
3882 		newpwd->pwd_rdir = rootvnode;
3883 	}
3884 	pwd_set(pdp, newpwd);
3885 	PWDDESC_XUNLOCK(pdp);
3886 	pwd_drop(oldpwd);
3887 }
3888 
3889 void
3890 pwd_set_rootvnode(void)
3891 {
3892 	struct pwddesc *pdp;
3893 	struct pwd *oldpwd, *newpwd;
3894 
3895 	pdp = curproc->p_pd;
3896 
3897 	newpwd = pwd_alloc();
3898 	PWDDESC_XLOCK(pdp);
3899 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3900 	vrefact(rootvnode);
3901 	newpwd->pwd_cdir = rootvnode;
3902 	vrefact(rootvnode);
3903 	newpwd->pwd_rdir = rootvnode;
3904 	pwd_fill(oldpwd, newpwd);
3905 	pwd_set(pdp, newpwd);
3906 	PWDDESC_XUNLOCK(pdp);
3907 	pwd_drop(oldpwd);
3908 }
3909 
3910 /*
3911  * Scan all active processes and prisons to see if any of them have a current
3912  * or root directory of `olddp'. If so, replace them with the new mount point.
3913  */
3914 void
3915 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3916 {
3917 	struct pwddesc *pdp;
3918 	struct pwd *newpwd, *oldpwd;
3919 	struct prison *pr;
3920 	struct proc *p;
3921 	int nrele;
3922 
3923 	if (vrefcnt(olddp) == 1)
3924 		return;
3925 	nrele = 0;
3926 	newpwd = pwd_alloc();
3927 	sx_slock(&allproc_lock);
3928 	FOREACH_PROC_IN_SYSTEM(p) {
3929 		PROC_LOCK(p);
3930 		pdp = pdhold(p);
3931 		PROC_UNLOCK(p);
3932 		if (pdp == NULL)
3933 			continue;
3934 		PWDDESC_XLOCK(pdp);
3935 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3936 		if (oldpwd == NULL ||
3937 		    (oldpwd->pwd_cdir != olddp &&
3938 		    oldpwd->pwd_rdir != olddp &&
3939 		    oldpwd->pwd_jdir != olddp)) {
3940 			PWDDESC_XUNLOCK(pdp);
3941 			pddrop(pdp);
3942 			continue;
3943 		}
3944 		if (oldpwd->pwd_cdir == olddp) {
3945 			vrefact(newdp);
3946 			newpwd->pwd_cdir = newdp;
3947 		}
3948 		if (oldpwd->pwd_rdir == olddp) {
3949 			vrefact(newdp);
3950 			newpwd->pwd_rdir = newdp;
3951 		}
3952 		if (oldpwd->pwd_jdir == olddp) {
3953 			vrefact(newdp);
3954 			newpwd->pwd_jdir = newdp;
3955 		}
3956 		pwd_fill(oldpwd, newpwd);
3957 		pwd_set(pdp, newpwd);
3958 		PWDDESC_XUNLOCK(pdp);
3959 		pwd_drop(oldpwd);
3960 		pddrop(pdp);
3961 		newpwd = pwd_alloc();
3962 	}
3963 	sx_sunlock(&allproc_lock);
3964 	pwd_drop(newpwd);
3965 	if (rootvnode == olddp) {
3966 		vrefact(newdp);
3967 		rootvnode = newdp;
3968 		nrele++;
3969 	}
3970 	mtx_lock(&prison0.pr_mtx);
3971 	if (prison0.pr_root == olddp) {
3972 		vrefact(newdp);
3973 		prison0.pr_root = newdp;
3974 		nrele++;
3975 	}
3976 	mtx_unlock(&prison0.pr_mtx);
3977 	sx_slock(&allprison_lock);
3978 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3979 		mtx_lock(&pr->pr_mtx);
3980 		if (pr->pr_root == olddp) {
3981 			vrefact(newdp);
3982 			pr->pr_root = newdp;
3983 			nrele++;
3984 		}
3985 		mtx_unlock(&pr->pr_mtx);
3986 	}
3987 	sx_sunlock(&allprison_lock);
3988 	while (nrele--)
3989 		vrele(olddp);
3990 }
3991 
3992 struct filedesc_to_leader *
3993 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3994 {
3995 	struct filedesc_to_leader *fdtol;
3996 
3997 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3998 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3999 	fdtol->fdl_refcount = 1;
4000 	fdtol->fdl_holdcount = 0;
4001 	fdtol->fdl_wakeup = 0;
4002 	fdtol->fdl_leader = leader;
4003 	if (old != NULL) {
4004 		FILEDESC_XLOCK(fdp);
4005 		fdtol->fdl_next = old->fdl_next;
4006 		fdtol->fdl_prev = old;
4007 		old->fdl_next = fdtol;
4008 		fdtol->fdl_next->fdl_prev = fdtol;
4009 		FILEDESC_XUNLOCK(fdp);
4010 	} else {
4011 		fdtol->fdl_next = fdtol;
4012 		fdtol->fdl_prev = fdtol;
4013 	}
4014 	return (fdtol);
4015 }
4016 
4017 static int
4018 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4019 {
4020 	NDSLOTTYPE *map;
4021 	struct filedesc *fdp;
4022 	u_int namelen;
4023 	int count, off, minoff;
4024 
4025 	namelen = arg2;
4026 	if (namelen != 1)
4027 		return (EINVAL);
4028 
4029 	if (*(int *)arg1 != 0)
4030 		return (EINVAL);
4031 
4032 	fdp = curproc->p_fd;
4033 	count = 0;
4034 	FILEDESC_SLOCK(fdp);
4035 	map = fdp->fd_map;
4036 	off = NDSLOT(fdp->fd_nfiles - 1);
4037 	for (minoff = NDSLOT(0); off >= minoff; --off)
4038 		count += bitcountl(map[off]);
4039 	FILEDESC_SUNLOCK(fdp);
4040 
4041 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4042 }
4043 
4044 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4045     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4046     "Number of open file descriptors");
4047 
4048 /*
4049  * Get file structures globally.
4050  */
4051 static int
4052 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4053 {
4054 	struct xfile xf;
4055 	struct filedesc *fdp;
4056 	struct file *fp;
4057 	struct proc *p;
4058 	int error, n, lastfile;
4059 
4060 	error = sysctl_wire_old_buffer(req, 0);
4061 	if (error != 0)
4062 		return (error);
4063 	if (req->oldptr == NULL) {
4064 		n = 0;
4065 		sx_slock(&allproc_lock);
4066 		FOREACH_PROC_IN_SYSTEM(p) {
4067 			PROC_LOCK(p);
4068 			if (p->p_state == PRS_NEW) {
4069 				PROC_UNLOCK(p);
4070 				continue;
4071 			}
4072 			fdp = fdhold(p);
4073 			PROC_UNLOCK(p);
4074 			if (fdp == NULL)
4075 				continue;
4076 			/* overestimates sparse tables. */
4077 			n += fdp->fd_nfiles;
4078 			fddrop(fdp);
4079 		}
4080 		sx_sunlock(&allproc_lock);
4081 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4082 	}
4083 	error = 0;
4084 	bzero(&xf, sizeof(xf));
4085 	xf.xf_size = sizeof(xf);
4086 	sx_slock(&allproc_lock);
4087 	FOREACH_PROC_IN_SYSTEM(p) {
4088 		PROC_LOCK(p);
4089 		if (p->p_state == PRS_NEW) {
4090 			PROC_UNLOCK(p);
4091 			continue;
4092 		}
4093 		if (p_cansee(req->td, p) != 0) {
4094 			PROC_UNLOCK(p);
4095 			continue;
4096 		}
4097 		xf.xf_pid = p->p_pid;
4098 		xf.xf_uid = p->p_ucred->cr_uid;
4099 		fdp = fdhold(p);
4100 		PROC_UNLOCK(p);
4101 		if (fdp == NULL)
4102 			continue;
4103 		FILEDESC_SLOCK(fdp);
4104 		lastfile = fdlastfile(fdp);
4105 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4106 		    n++) {
4107 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4108 				continue;
4109 			xf.xf_fd = n;
4110 			xf.xf_file = (uintptr_t)fp;
4111 			xf.xf_data = (uintptr_t)fp->f_data;
4112 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4113 			xf.xf_type = (uintptr_t)fp->f_type;
4114 			xf.xf_count = refcount_load(&fp->f_count);
4115 			xf.xf_msgcount = 0;
4116 			xf.xf_offset = foffset_get(fp);
4117 			xf.xf_flag = fp->f_flag;
4118 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4119 			if (error)
4120 				break;
4121 		}
4122 		FILEDESC_SUNLOCK(fdp);
4123 		fddrop(fdp);
4124 		if (error)
4125 			break;
4126 	}
4127 	sx_sunlock(&allproc_lock);
4128 	return (error);
4129 }
4130 
4131 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4132     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4133 
4134 #ifdef KINFO_FILE_SIZE
4135 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4136 #endif
4137 
4138 static int
4139 xlate_fflags(int fflags)
4140 {
4141 	static const struct {
4142 		int	fflag;
4143 		int	kf_fflag;
4144 	} fflags_table[] = {
4145 		{ FAPPEND, KF_FLAG_APPEND },
4146 		{ FASYNC, KF_FLAG_ASYNC },
4147 		{ FFSYNC, KF_FLAG_FSYNC },
4148 		{ FHASLOCK, KF_FLAG_HASLOCK },
4149 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4150 		{ FREAD, KF_FLAG_READ },
4151 		{ FWRITE, KF_FLAG_WRITE },
4152 		{ O_CREAT, KF_FLAG_CREAT },
4153 		{ O_DIRECT, KF_FLAG_DIRECT },
4154 		{ O_EXCL, KF_FLAG_EXCL },
4155 		{ O_EXEC, KF_FLAG_EXEC },
4156 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4157 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4158 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4159 		{ O_TRUNC, KF_FLAG_TRUNC }
4160 	};
4161 	unsigned int i;
4162 	int kflags;
4163 
4164 	kflags = 0;
4165 	for (i = 0; i < nitems(fflags_table); i++)
4166 		if (fflags & fflags_table[i].fflag)
4167 			kflags |=  fflags_table[i].kf_fflag;
4168 	return (kflags);
4169 }
4170 
4171 /* Trim unused data from kf_path by truncating the structure size. */
4172 void
4173 pack_kinfo(struct kinfo_file *kif)
4174 {
4175 
4176 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4177 	    strlen(kif->kf_path) + 1;
4178 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4179 }
4180 
4181 static void
4182 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4183     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4184 {
4185 	int error;
4186 
4187 	bzero(kif, sizeof(*kif));
4188 
4189 	/* Set a default type to allow for empty fill_kinfo() methods. */
4190 	kif->kf_type = KF_TYPE_UNKNOWN;
4191 	kif->kf_flags = xlate_fflags(fp->f_flag);
4192 	if (rightsp != NULL)
4193 		kif->kf_cap_rights = *rightsp;
4194 	else
4195 		cap_rights_init_zero(&kif->kf_cap_rights);
4196 	kif->kf_fd = fd;
4197 	kif->kf_ref_count = refcount_load(&fp->f_count);
4198 	kif->kf_offset = foffset_get(fp);
4199 
4200 	/*
4201 	 * This may drop the filedesc lock, so the 'fp' cannot be
4202 	 * accessed after this call.
4203 	 */
4204 	error = fo_fill_kinfo(fp, kif, fdp);
4205 	if (error == 0)
4206 		kif->kf_status |= KF_ATTR_VALID;
4207 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4208 		pack_kinfo(kif);
4209 	else
4210 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4211 }
4212 
4213 static void
4214 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4215     struct kinfo_file *kif, int flags)
4216 {
4217 	int error;
4218 
4219 	bzero(kif, sizeof(*kif));
4220 
4221 	kif->kf_type = KF_TYPE_VNODE;
4222 	error = vn_fill_kinfo_vnode(vp, kif);
4223 	if (error == 0)
4224 		kif->kf_status |= KF_ATTR_VALID;
4225 	kif->kf_flags = xlate_fflags(fflags);
4226 	cap_rights_init_zero(&kif->kf_cap_rights);
4227 	kif->kf_fd = fd;
4228 	kif->kf_ref_count = -1;
4229 	kif->kf_offset = -1;
4230 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4231 		pack_kinfo(kif);
4232 	else
4233 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4234 	vrele(vp);
4235 }
4236 
4237 struct export_fd_buf {
4238 	struct filedesc		*fdp;
4239 	struct pwddesc	*pdp;
4240 	struct sbuf 		*sb;
4241 	ssize_t			remainder;
4242 	struct kinfo_file	kif;
4243 	int			flags;
4244 };
4245 
4246 static int
4247 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4248 {
4249 	struct kinfo_file *kif;
4250 
4251 	kif = &efbuf->kif;
4252 	if (efbuf->remainder != -1) {
4253 		if (efbuf->remainder < kif->kf_structsize) {
4254 			/* Terminate export. */
4255 			efbuf->remainder = 0;
4256 			return (0);
4257 		}
4258 		efbuf->remainder -= kif->kf_structsize;
4259 	}
4260 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4261 }
4262 
4263 static int
4264 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4265     struct export_fd_buf *efbuf)
4266 {
4267 	int error;
4268 
4269 	if (efbuf->remainder == 0)
4270 		return (0);
4271 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4272 	    efbuf->flags);
4273 	FILEDESC_SUNLOCK(efbuf->fdp);
4274 	error = export_kinfo_to_sb(efbuf);
4275 	FILEDESC_SLOCK(efbuf->fdp);
4276 	return (error);
4277 }
4278 
4279 static int
4280 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4281     struct export_fd_buf *efbuf)
4282 {
4283 	int error;
4284 
4285 	if (efbuf->remainder == 0)
4286 		return (0);
4287 	if (efbuf->pdp != NULL)
4288 		PWDDESC_XUNLOCK(efbuf->pdp);
4289 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4290 	error = export_kinfo_to_sb(efbuf);
4291 	if (efbuf->pdp != NULL)
4292 		PWDDESC_XLOCK(efbuf->pdp);
4293 	return (error);
4294 }
4295 
4296 /*
4297  * Store a process file descriptor information to sbuf.
4298  *
4299  * Takes a locked proc as argument, and returns with the proc unlocked.
4300  */
4301 int
4302 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4303     int flags)
4304 {
4305 	struct file *fp;
4306 	struct filedesc *fdp;
4307 	struct pwddesc *pdp;
4308 	struct export_fd_buf *efbuf;
4309 	struct vnode *cttyvp, *textvp, *tracevp;
4310 	struct pwd *pwd;
4311 	int error, i, lastfile;
4312 	cap_rights_t rights;
4313 
4314 	PROC_LOCK_ASSERT(p, MA_OWNED);
4315 
4316 	/* ktrace vnode */
4317 	tracevp = ktr_get_tracevp(p, true);
4318 	/* text vnode */
4319 	textvp = p->p_textvp;
4320 	if (textvp != NULL)
4321 		vrefact(textvp);
4322 	/* Controlling tty. */
4323 	cttyvp = NULL;
4324 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4325 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4326 		if (cttyvp != NULL)
4327 			vrefact(cttyvp);
4328 	}
4329 	fdp = fdhold(p);
4330 	pdp = pdhold(p);
4331 	PROC_UNLOCK(p);
4332 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4333 	efbuf->fdp = NULL;
4334 	efbuf->pdp = NULL;
4335 	efbuf->sb = sb;
4336 	efbuf->remainder = maxlen;
4337 	efbuf->flags = flags;
4338 	if (tracevp != NULL)
4339 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4340 		    efbuf);
4341 	if (textvp != NULL)
4342 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4343 	if (cttyvp != NULL)
4344 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4345 		    efbuf);
4346 	error = 0;
4347 	if (pdp == NULL || fdp == NULL)
4348 		goto fail;
4349 	efbuf->fdp = fdp;
4350 	efbuf->pdp = pdp;
4351 	PWDDESC_XLOCK(pdp);
4352 	pwd = pwd_hold_pwddesc(pdp);
4353 	if (pwd != NULL) {
4354 		/* working directory */
4355 		if (pwd->pwd_cdir != NULL) {
4356 			vrefact(pwd->pwd_cdir);
4357 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4358 			    FREAD, efbuf);
4359 		}
4360 		/* root directory */
4361 		if (pwd->pwd_rdir != NULL) {
4362 			vrefact(pwd->pwd_rdir);
4363 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4364 			    FREAD, efbuf);
4365 		}
4366 		/* jail directory */
4367 		if (pwd->pwd_jdir != NULL) {
4368 			vrefact(pwd->pwd_jdir);
4369 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4370 			    FREAD, efbuf);
4371 		}
4372 	}
4373 	PWDDESC_XUNLOCK(pdp);
4374 	if (pwd != NULL)
4375 		pwd_drop(pwd);
4376 	FILEDESC_SLOCK(fdp);
4377 	lastfile = fdlastfile(fdp);
4378 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4379 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4380 			continue;
4381 #ifdef CAPABILITIES
4382 		rights = *cap_rights(fdp, i);
4383 #else /* !CAPABILITIES */
4384 		rights = cap_no_rights;
4385 #endif
4386 		/*
4387 		 * Create sysctl entry.  It is OK to drop the filedesc
4388 		 * lock inside of export_file_to_sb() as we will
4389 		 * re-validate and re-evaluate its properties when the
4390 		 * loop continues.
4391 		 */
4392 		error = export_file_to_sb(fp, i, &rights, efbuf);
4393 		if (error != 0 || efbuf->remainder == 0)
4394 			break;
4395 	}
4396 	FILEDESC_SUNLOCK(fdp);
4397 fail:
4398 	if (fdp != NULL)
4399 		fddrop(fdp);
4400 	if (pdp != NULL)
4401 		pddrop(pdp);
4402 	free(efbuf, M_TEMP);
4403 	return (error);
4404 }
4405 
4406 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4407 
4408 /*
4409  * Get per-process file descriptors for use by procstat(1), et al.
4410  */
4411 static int
4412 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4413 {
4414 	struct sbuf sb;
4415 	struct proc *p;
4416 	ssize_t maxlen;
4417 	u_int namelen;
4418 	int error, error2, *name;
4419 
4420 	namelen = arg2;
4421 	if (namelen != 1)
4422 		return (EINVAL);
4423 
4424 	name = (int *)arg1;
4425 
4426 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4427 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4428 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4429 	if (error != 0) {
4430 		sbuf_delete(&sb);
4431 		return (error);
4432 	}
4433 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4434 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4435 	    KERN_FILEDESC_PACK_KINFO);
4436 	error2 = sbuf_finish(&sb);
4437 	sbuf_delete(&sb);
4438 	return (error != 0 ? error : error2);
4439 }
4440 
4441 #ifdef COMPAT_FREEBSD7
4442 #ifdef KINFO_OFILE_SIZE
4443 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4444 #endif
4445 
4446 static void
4447 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4448 {
4449 
4450 	okif->kf_structsize = sizeof(*okif);
4451 	okif->kf_type = kif->kf_type;
4452 	okif->kf_fd = kif->kf_fd;
4453 	okif->kf_ref_count = kif->kf_ref_count;
4454 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4455 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4456 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4457 	okif->kf_offset = kif->kf_offset;
4458 	if (kif->kf_type == KF_TYPE_VNODE)
4459 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4460 	else
4461 		okif->kf_vnode_type = KF_VTYPE_VNON;
4462 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4463 	if (kif->kf_type == KF_TYPE_SOCKET) {
4464 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4465 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4466 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4467 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4468 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4469 	} else {
4470 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4471 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4472 	}
4473 }
4474 
4475 static int
4476 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4477     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4478 {
4479 	int error;
4480 
4481 	vrefact(vp);
4482 	PWDDESC_XUNLOCK(pdp);
4483 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4484 	kinfo_to_okinfo(kif, okif);
4485 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4486 	PWDDESC_XLOCK(pdp);
4487 	return (error);
4488 }
4489 
4490 /*
4491  * Get per-process file descriptors for use by procstat(1), et al.
4492  */
4493 static int
4494 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4495 {
4496 	struct kinfo_ofile *okif;
4497 	struct kinfo_file *kif;
4498 	struct filedesc *fdp;
4499 	struct pwddesc *pdp;
4500 	struct pwd *pwd;
4501 	u_int namelen;
4502 	int error, i, lastfile, *name;
4503 	struct file *fp;
4504 	struct proc *p;
4505 
4506 	namelen = arg2;
4507 	if (namelen != 1)
4508 		return (EINVAL);
4509 
4510 	name = (int *)arg1;
4511 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4512 	if (error != 0)
4513 		return (error);
4514 	fdp = fdhold(p);
4515 	if (fdp != NULL)
4516 		pdp = pdhold(p);
4517 	PROC_UNLOCK(p);
4518 	if (fdp == NULL || pdp == NULL) {
4519 		if (fdp != NULL)
4520 			fddrop(fdp);
4521 		return (ENOENT);
4522 	}
4523 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4524 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4525 	PWDDESC_XLOCK(pdp);
4526 	pwd = pwd_hold_pwddesc(pdp);
4527 	if (pwd != NULL) {
4528 		if (pwd->pwd_cdir != NULL)
4529 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4530 			    okif, pdp, req);
4531 		if (pwd->pwd_rdir != NULL)
4532 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4533 			    okif, pdp, req);
4534 		if (pwd->pwd_jdir != NULL)
4535 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4536 			    okif, pdp, req);
4537 	}
4538 	PWDDESC_XUNLOCK(pdp);
4539 	if (pwd != NULL)
4540 		pwd_drop(pwd);
4541 	FILEDESC_SLOCK(fdp);
4542 	lastfile = fdlastfile(fdp);
4543 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4544 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4545 			continue;
4546 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4547 		    KERN_FILEDESC_PACK_KINFO);
4548 		FILEDESC_SUNLOCK(fdp);
4549 		kinfo_to_okinfo(kif, okif);
4550 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4551 		FILEDESC_SLOCK(fdp);
4552 		if (error)
4553 			break;
4554 	}
4555 	FILEDESC_SUNLOCK(fdp);
4556 	fddrop(fdp);
4557 	pddrop(pdp);
4558 	free(kif, M_TEMP);
4559 	free(okif, M_TEMP);
4560 	return (0);
4561 }
4562 
4563 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4564     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4565     "Process ofiledesc entries");
4566 #endif	/* COMPAT_FREEBSD7 */
4567 
4568 int
4569 vntype_to_kinfo(int vtype)
4570 {
4571 	struct {
4572 		int	vtype;
4573 		int	kf_vtype;
4574 	} vtypes_table[] = {
4575 		{ VBAD, KF_VTYPE_VBAD },
4576 		{ VBLK, KF_VTYPE_VBLK },
4577 		{ VCHR, KF_VTYPE_VCHR },
4578 		{ VDIR, KF_VTYPE_VDIR },
4579 		{ VFIFO, KF_VTYPE_VFIFO },
4580 		{ VLNK, KF_VTYPE_VLNK },
4581 		{ VNON, KF_VTYPE_VNON },
4582 		{ VREG, KF_VTYPE_VREG },
4583 		{ VSOCK, KF_VTYPE_VSOCK }
4584 	};
4585 	unsigned int i;
4586 
4587 	/*
4588 	 * Perform vtype translation.
4589 	 */
4590 	for (i = 0; i < nitems(vtypes_table); i++)
4591 		if (vtypes_table[i].vtype == vtype)
4592 			return (vtypes_table[i].kf_vtype);
4593 
4594 	return (KF_VTYPE_UNKNOWN);
4595 }
4596 
4597 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4598     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4599     "Process filedesc entries");
4600 
4601 /*
4602  * Store a process current working directory information to sbuf.
4603  *
4604  * Takes a locked proc as argument, and returns with the proc unlocked.
4605  */
4606 int
4607 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4608 {
4609 	struct pwddesc *pdp;
4610 	struct pwd *pwd;
4611 	struct export_fd_buf *efbuf;
4612 	struct vnode *cdir;
4613 	int error;
4614 
4615 	PROC_LOCK_ASSERT(p, MA_OWNED);
4616 
4617 	pdp = pdhold(p);
4618 	PROC_UNLOCK(p);
4619 	if (pdp == NULL)
4620 		return (EINVAL);
4621 
4622 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4623 	efbuf->pdp = pdp;
4624 	efbuf->sb = sb;
4625 	efbuf->remainder = maxlen;
4626 
4627 	PWDDESC_XLOCK(pdp);
4628 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4629 	cdir = pwd->pwd_cdir;
4630 	if (cdir == NULL) {
4631 		error = EINVAL;
4632 	} else {
4633 		vrefact(cdir);
4634 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4635 	}
4636 	PWDDESC_XUNLOCK(pdp);
4637 	pddrop(pdp);
4638 	free(efbuf, M_TEMP);
4639 	return (error);
4640 }
4641 
4642 /*
4643  * Get per-process current working directory.
4644  */
4645 static int
4646 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4647 {
4648 	struct sbuf sb;
4649 	struct proc *p;
4650 	ssize_t maxlen;
4651 	u_int namelen;
4652 	int error, error2, *name;
4653 
4654 	namelen = arg2;
4655 	if (namelen != 1)
4656 		return (EINVAL);
4657 
4658 	name = (int *)arg1;
4659 
4660 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4661 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4662 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4663 	if (error != 0) {
4664 		sbuf_delete(&sb);
4665 		return (error);
4666 	}
4667 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4668 	error = kern_proc_cwd_out(p, &sb, maxlen);
4669 	error2 = sbuf_finish(&sb);
4670 	sbuf_delete(&sb);
4671 	return (error != 0 ? error : error2);
4672 }
4673 
4674 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4675     sysctl_kern_proc_cwd, "Process current working directory");
4676 
4677 #ifdef DDB
4678 /*
4679  * For the purposes of debugging, generate a human-readable string for the
4680  * file type.
4681  */
4682 static const char *
4683 file_type_to_name(short type)
4684 {
4685 
4686 	switch (type) {
4687 	case 0:
4688 		return ("zero");
4689 	case DTYPE_VNODE:
4690 		return ("vnode");
4691 	case DTYPE_SOCKET:
4692 		return ("socket");
4693 	case DTYPE_PIPE:
4694 		return ("pipe");
4695 	case DTYPE_FIFO:
4696 		return ("fifo");
4697 	case DTYPE_KQUEUE:
4698 		return ("kqueue");
4699 	case DTYPE_CRYPTO:
4700 		return ("crypto");
4701 	case DTYPE_MQUEUE:
4702 		return ("mqueue");
4703 	case DTYPE_SHM:
4704 		return ("shm");
4705 	case DTYPE_SEM:
4706 		return ("ksem");
4707 	case DTYPE_PTS:
4708 		return ("pts");
4709 	case DTYPE_DEV:
4710 		return ("dev");
4711 	case DTYPE_PROCDESC:
4712 		return ("proc");
4713 	case DTYPE_EVENTFD:
4714 		return ("eventfd");
4715 	case DTYPE_LINUXTFD:
4716 		return ("ltimer");
4717 	default:
4718 		return ("unkn");
4719 	}
4720 }
4721 
4722 /*
4723  * For the purposes of debugging, identify a process (if any, perhaps one of
4724  * many) that references the passed file in its file descriptor array. Return
4725  * NULL if none.
4726  */
4727 static struct proc *
4728 file_to_first_proc(struct file *fp)
4729 {
4730 	struct filedesc *fdp;
4731 	struct proc *p;
4732 	int n;
4733 
4734 	FOREACH_PROC_IN_SYSTEM(p) {
4735 		if (p->p_state == PRS_NEW)
4736 			continue;
4737 		fdp = p->p_fd;
4738 		if (fdp == NULL)
4739 			continue;
4740 		for (n = 0; n < fdp->fd_nfiles; n++) {
4741 			if (fp == fdp->fd_ofiles[n].fde_file)
4742 				return (p);
4743 		}
4744 	}
4745 	return (NULL);
4746 }
4747 
4748 static void
4749 db_print_file(struct file *fp, int header)
4750 {
4751 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4752 	struct proc *p;
4753 
4754 	if (header)
4755 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4756 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4757 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4758 		    "FCmd");
4759 	p = file_to_first_proc(fp);
4760 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4761 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4762 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4763 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4764 
4765 #undef XPTRWIDTH
4766 }
4767 
4768 DB_SHOW_COMMAND(file, db_show_file)
4769 {
4770 	struct file *fp;
4771 
4772 	if (!have_addr) {
4773 		db_printf("usage: show file <addr>\n");
4774 		return;
4775 	}
4776 	fp = (struct file *)addr;
4777 	db_print_file(fp, 1);
4778 }
4779 
4780 DB_SHOW_COMMAND(files, db_show_files)
4781 {
4782 	struct filedesc *fdp;
4783 	struct file *fp;
4784 	struct proc *p;
4785 	int header;
4786 	int n;
4787 
4788 	header = 1;
4789 	FOREACH_PROC_IN_SYSTEM(p) {
4790 		if (p->p_state == PRS_NEW)
4791 			continue;
4792 		if ((fdp = p->p_fd) == NULL)
4793 			continue;
4794 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4795 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4796 				continue;
4797 			db_print_file(fp, header);
4798 			header = 0;
4799 		}
4800 	}
4801 }
4802 #endif
4803 
4804 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4805     &maxfilesperproc, 0, "Maximum files allowed open per process");
4806 
4807 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4808     &maxfiles, 0, "Maximum number of files");
4809 
4810 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4811     &openfiles, 0, "System-wide number of open files");
4812 
4813 /* ARGSUSED*/
4814 static void
4815 filelistinit(void *dummy)
4816 {
4817 
4818 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4819 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4820 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4821 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4822 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4823 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4824 	/*
4825 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4826 	 * the vnode zone.
4827 	 */
4828 	vfs_smr = uma_zone_get_smr(pwd_zone);
4829 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4830 }
4831 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4832 
4833 /*-------------------------------------------------------------------*/
4834 
4835 static int
4836 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4837     int flags, struct thread *td)
4838 {
4839 
4840 	return (EBADF);
4841 }
4842 
4843 static int
4844 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4845     struct thread *td)
4846 {
4847 
4848 	return (EINVAL);
4849 }
4850 
4851 static int
4852 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4853     struct thread *td)
4854 {
4855 
4856 	return (EBADF);
4857 }
4858 
4859 static int
4860 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4861     struct thread *td)
4862 {
4863 
4864 	return (0);
4865 }
4866 
4867 static int
4868 badfo_kqfilter(struct file *fp, struct knote *kn)
4869 {
4870 
4871 	return (EBADF);
4872 }
4873 
4874 static int
4875 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4876 {
4877 
4878 	return (EBADF);
4879 }
4880 
4881 static int
4882 badfo_close(struct file *fp, struct thread *td)
4883 {
4884 
4885 	return (0);
4886 }
4887 
4888 static int
4889 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4890     struct thread *td)
4891 {
4892 
4893 	return (EBADF);
4894 }
4895 
4896 static int
4897 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4898     struct thread *td)
4899 {
4900 
4901 	return (EBADF);
4902 }
4903 
4904 static int
4905 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4906     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4907     struct thread *td)
4908 {
4909 
4910 	return (EBADF);
4911 }
4912 
4913 static int
4914 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4915 {
4916 
4917 	return (0);
4918 }
4919 
4920 struct fileops badfileops = {
4921 	.fo_read = badfo_readwrite,
4922 	.fo_write = badfo_readwrite,
4923 	.fo_truncate = badfo_truncate,
4924 	.fo_ioctl = badfo_ioctl,
4925 	.fo_poll = badfo_poll,
4926 	.fo_kqfilter = badfo_kqfilter,
4927 	.fo_stat = badfo_stat,
4928 	.fo_close = badfo_close,
4929 	.fo_chmod = badfo_chmod,
4930 	.fo_chown = badfo_chown,
4931 	.fo_sendfile = badfo_sendfile,
4932 	.fo_fill_kinfo = badfo_fill_kinfo,
4933 };
4934 
4935 static int
4936 path_poll(struct file *fp, int events, struct ucred *active_cred,
4937     struct thread *td)
4938 {
4939 	return (POLLNVAL);
4940 }
4941 
4942 static int
4943 path_close(struct file *fp, struct thread *td)
4944 {
4945 	MPASS(fp->f_type == DTYPE_VNODE);
4946 	fp->f_ops = &badfileops;
4947 	vdrop(fp->f_vnode);
4948 	return (0);
4949 }
4950 
4951 struct fileops path_fileops = {
4952 	.fo_read = badfo_readwrite,
4953 	.fo_write = badfo_readwrite,
4954 	.fo_truncate = badfo_truncate,
4955 	.fo_ioctl = badfo_ioctl,
4956 	.fo_poll = path_poll,
4957 	.fo_kqfilter = vn_kqfilter_opath,
4958 	.fo_stat = vn_statfile,
4959 	.fo_close = path_close,
4960 	.fo_chmod = badfo_chmod,
4961 	.fo_chown = badfo_chown,
4962 	.fo_sendfile = badfo_sendfile,
4963 	.fo_fill_kinfo = vn_fill_kinfo,
4964 	.fo_flags = DFLAG_PASSABLE,
4965 };
4966 
4967 int
4968 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4969     int flags, struct thread *td)
4970 {
4971 
4972 	return (EOPNOTSUPP);
4973 }
4974 
4975 int
4976 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4977     struct thread *td)
4978 {
4979 
4980 	return (EINVAL);
4981 }
4982 
4983 int
4984 invfo_ioctl(struct file *fp, u_long com, void *data,
4985     struct ucred *active_cred, struct thread *td)
4986 {
4987 
4988 	return (ENOTTY);
4989 }
4990 
4991 int
4992 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4993     struct thread *td)
4994 {
4995 
4996 	return (poll_no_poll(events));
4997 }
4998 
4999 int
5000 invfo_kqfilter(struct file *fp, struct knote *kn)
5001 {
5002 
5003 	return (EINVAL);
5004 }
5005 
5006 int
5007 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5008     struct thread *td)
5009 {
5010 
5011 	return (EINVAL);
5012 }
5013 
5014 int
5015 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5016     struct thread *td)
5017 {
5018 
5019 	return (EINVAL);
5020 }
5021 
5022 int
5023 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5024     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5025     struct thread *td)
5026 {
5027 
5028 	return (EINVAL);
5029 }
5030 
5031 /*-------------------------------------------------------------------*/
5032 
5033 /*
5034  * File Descriptor pseudo-device driver (/dev/fd/).
5035  *
5036  * Opening minor device N dup()s the file (if any) connected to file
5037  * descriptor N belonging to the calling process.  Note that this driver
5038  * consists of only the ``open()'' routine, because all subsequent
5039  * references to this file will be direct to the other driver.
5040  *
5041  * XXX: we could give this one a cloning event handler if necessary.
5042  */
5043 
5044 /* ARGSUSED */
5045 static int
5046 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5047 {
5048 
5049 	/*
5050 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5051 	 * the file descriptor being sought for duplication. The error
5052 	 * return ensures that the vnode for this device will be released
5053 	 * by vn_open. Open will detect this special error and take the
5054 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5055 	 * will simply report the error.
5056 	 */
5057 	td->td_dupfd = dev2unit(dev);
5058 	return (ENODEV);
5059 }
5060 
5061 static struct cdevsw fildesc_cdevsw = {
5062 	.d_version =	D_VERSION,
5063 	.d_open =	fdopen,
5064 	.d_name =	"FD",
5065 };
5066 
5067 static void
5068 fildesc_drvinit(void *unused)
5069 {
5070 	struct cdev *dev;
5071 
5072 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5073 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5074 	make_dev_alias(dev, "stdin");
5075 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5076 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5077 	make_dev_alias(dev, "stdout");
5078 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5079 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5080 	make_dev_alias(dev, "stderr");
5081 }
5082 
5083 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5084