xref: /freebsd/sys/kern/kern_descrip.c (revision d396c67f26b079c2808002c07212d9df9818a11b)
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 || mode == FDDUP_MUSTREPLACE) && 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_MUSTREPLACE:
936 		/* Target file descriptor must exist. */
937 		if (fget_locked(fdp, new) == NULL) {
938 			fdrop(oldfp, td);
939 			goto unlock;
940 		}
941 		break;
942 	case FDDUP_FIXED:
943 		if (new >= fdp->fd_nfiles) {
944 			/*
945 			 * The resource limits are here instead of e.g.
946 			 * fdalloc(), because the file descriptor table may be
947 			 * shared between processes, so we can't really use
948 			 * racct_add()/racct_sub().  Instead of counting the
949 			 * number of actually allocated descriptors, just put
950 			 * the limit on the size of the file descriptor table.
951 			 */
952 #ifdef RACCT
953 			if (RACCT_ENABLED()) {
954 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
955 				if (error != 0) {
956 					error = EMFILE;
957 					fdrop(oldfp, td);
958 					goto unlock;
959 				}
960 			}
961 #endif
962 			fdgrowtable_exp(fdp, new + 1);
963 		}
964 		if (!fdisused(fdp, new))
965 			fdused(fdp, new);
966 		break;
967 	default:
968 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
969 	}
970 
971 	KASSERT(old != new, ("new fd is same as old"));
972 
973 	/* Refetch oldfde because the table may have grown and old one freed. */
974 	oldfde = &fdp->fd_ofiles[old];
975 	KASSERT(oldfp == oldfde->fde_file,
976 	    ("fdt_ofiles shift from growth observed at fd %d",
977 	    old));
978 
979 	newfde = &fdp->fd_ofiles[new];
980 	delfp = newfde->fde_file;
981 
982 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
983 
984 	/*
985 	 * Duplicate the source descriptor.
986 	 */
987 #ifdef CAPABILITIES
988 	seqc_write_begin(&newfde->fde_seqc);
989 #endif
990 	oioctls = filecaps_free_prep(&newfde->fde_caps);
991 	memcpy(newfde, oldfde, fde_change_size);
992 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
993 	    nioctls);
994 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
995 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
996 	else
997 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
998 #ifdef CAPABILITIES
999 	seqc_write_end(&newfde->fde_seqc);
1000 #endif
1001 	td->td_retval[0] = new;
1002 
1003 	error = 0;
1004 
1005 	if (delfp != NULL) {
1006 		(void) closefp(fdp, new, delfp, td, true, false);
1007 		FILEDESC_UNLOCK_ASSERT(fdp);
1008 	} else {
1009 unlock:
1010 		FILEDESC_XUNLOCK(fdp);
1011 	}
1012 
1013 	filecaps_free_finish(oioctls);
1014 	return (error);
1015 }
1016 
1017 static void
1018 sigiofree(struct sigio *sigio)
1019 {
1020 	crfree(sigio->sio_ucred);
1021 	free(sigio, M_SIGIO);
1022 }
1023 
1024 static struct sigio *
1025 funsetown_locked(struct sigio *sigio)
1026 {
1027 	struct proc *p;
1028 	struct pgrp *pg;
1029 
1030 	SIGIO_ASSERT_LOCKED();
1031 
1032 	if (sigio == NULL)
1033 		return (NULL);
1034 	*sigio->sio_myref = NULL;
1035 	if (sigio->sio_pgid < 0) {
1036 		pg = sigio->sio_pgrp;
1037 		PGRP_LOCK(pg);
1038 		SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1039 		PGRP_UNLOCK(pg);
1040 	} else {
1041 		p = sigio->sio_proc;
1042 		PROC_LOCK(p);
1043 		SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1044 		PROC_UNLOCK(p);
1045 	}
1046 	return (sigio);
1047 }
1048 
1049 /*
1050  * If sigio is on the list associated with a process or process group,
1051  * disable signalling from the device, remove sigio from the list and
1052  * free sigio.
1053  */
1054 void
1055 funsetown(struct sigio **sigiop)
1056 {
1057 	struct sigio *sigio;
1058 
1059 	/* Racy check, consumers must provide synchronization. */
1060 	if (*sigiop == NULL)
1061 		return;
1062 
1063 	SIGIO_LOCK();
1064 	sigio = funsetown_locked(*sigiop);
1065 	SIGIO_UNLOCK();
1066 	if (sigio != NULL)
1067 		sigiofree(sigio);
1068 }
1069 
1070 /*
1071  * Free a list of sigio structures.  The caller must ensure that new sigio
1072  * structures cannot be added after this point.  For process groups this is
1073  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1074  * as an interlock.
1075  */
1076 void
1077 funsetownlst(struct sigiolst *sigiolst)
1078 {
1079 	struct proc *p;
1080 	struct pgrp *pg;
1081 	struct sigio *sigio, *tmp;
1082 
1083 	/* Racy check. */
1084 	sigio = SLIST_FIRST(sigiolst);
1085 	if (sigio == NULL)
1086 		return;
1087 
1088 	p = NULL;
1089 	pg = NULL;
1090 
1091 	SIGIO_LOCK();
1092 	sigio = SLIST_FIRST(sigiolst);
1093 	if (sigio == NULL) {
1094 		SIGIO_UNLOCK();
1095 		return;
1096 	}
1097 
1098 	/*
1099 	 * Every entry of the list should belong to a single proc or pgrp.
1100 	 */
1101 	if (sigio->sio_pgid < 0) {
1102 		pg = sigio->sio_pgrp;
1103 		sx_assert(&proctree_lock, SX_XLOCKED);
1104 		PGRP_LOCK(pg);
1105 	} else /* if (sigio->sio_pgid > 0) */ {
1106 		p = sigio->sio_proc;
1107 		PROC_LOCK(p);
1108 		KASSERT((p->p_flag & P_WEXIT) != 0,
1109 		    ("%s: process %p is not exiting", __func__, p));
1110 	}
1111 
1112 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1113 		*sigio->sio_myref = NULL;
1114 		if (pg != NULL) {
1115 			KASSERT(sigio->sio_pgid < 0,
1116 			    ("Proc sigio in pgrp sigio list"));
1117 			KASSERT(sigio->sio_pgrp == pg,
1118 			    ("Bogus pgrp in sigio list"));
1119 		} else /* if (p != NULL) */ {
1120 			KASSERT(sigio->sio_pgid > 0,
1121 			    ("Pgrp sigio in proc sigio list"));
1122 			KASSERT(sigio->sio_proc == p,
1123 			    ("Bogus proc in sigio list"));
1124 		}
1125 	}
1126 
1127 	if (pg != NULL)
1128 		PGRP_UNLOCK(pg);
1129 	else
1130 		PROC_UNLOCK(p);
1131 	SIGIO_UNLOCK();
1132 
1133 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1134 		sigiofree(sigio);
1135 }
1136 
1137 /*
1138  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1139  *
1140  * After permission checking, add a sigio structure to the sigio list for
1141  * the process or process group.
1142  */
1143 int
1144 fsetown(pid_t pgid, struct sigio **sigiop)
1145 {
1146 	struct proc *proc;
1147 	struct pgrp *pgrp;
1148 	struct sigio *osigio, *sigio;
1149 	int ret;
1150 
1151 	if (pgid == 0) {
1152 		funsetown(sigiop);
1153 		return (0);
1154 	}
1155 
1156 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1157 	sigio->sio_pgid = pgid;
1158 	sigio->sio_ucred = crhold(curthread->td_ucred);
1159 	sigio->sio_myref = sigiop;
1160 
1161 	osigio = NULL;
1162 	ret = 0;
1163 	if (pgid > 0) {
1164 		ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1165 		SIGIO_LOCK();
1166 		if (ret == 0) {
1167 			osigio = funsetown_locked(*sigiop);
1168 
1169 			PROC_LOCK(proc);
1170 			_PRELE(proc);
1171 			if ((proc->p_flag & P_WEXIT) != 0) {
1172 				ret = ESRCH;
1173 			} else if (proc->p_session !=
1174 			    curthread->td_proc->p_session) {
1175 				/*
1176 				 * Policy - Don't allow a process to FSETOWN a
1177 				 * process in another session.
1178 				 *
1179 				 * Remove this test to allow maximum flexibility
1180 				 * or restrict FSETOWN to the current process or
1181 				 * process group for maximum safety.
1182 				 */
1183 				ret = EPERM;
1184 			} else {
1185 				sigio->sio_proc = proc;
1186 				SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1187 				    sio_pgsigio);
1188 			}
1189 			PROC_UNLOCK(proc);
1190 		}
1191 	} else /* if (pgid < 0) */ {
1192 		sx_slock(&proctree_lock);
1193 		SIGIO_LOCK();
1194 		pgrp = pgfind(-pgid);
1195 		if (pgrp == NULL) {
1196 			ret = ESRCH;
1197 		} else {
1198 			osigio = funsetown_locked(*sigiop);
1199 
1200 			if (pgrp->pg_session != curthread->td_proc->p_session) {
1201 				/*
1202 				 * Policy - Don't allow a process to FSETOWN a
1203 				 * process in another session.
1204 				 *
1205 				 * Remove this test to allow maximum flexibility
1206 				 * or restrict FSETOWN to the current process or
1207 				 * process group for maximum safety.
1208 				 */
1209 				ret = EPERM;
1210 			} else {
1211 				sigio->sio_pgrp = pgrp;
1212 				SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1213 				    sio_pgsigio);
1214 			}
1215 			PGRP_UNLOCK(pgrp);
1216 		}
1217 		sx_sunlock(&proctree_lock);
1218 	}
1219 	if (ret == 0)
1220 		*sigiop = sigio;
1221 	SIGIO_UNLOCK();
1222 	if (osigio != NULL)
1223 		sigiofree(osigio);
1224 	return (ret);
1225 }
1226 
1227 /*
1228  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1229  */
1230 pid_t
1231 fgetown(struct sigio **sigiop)
1232 {
1233 	pid_t pgid;
1234 
1235 	SIGIO_LOCK();
1236 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1237 	SIGIO_UNLOCK();
1238 	return (pgid);
1239 }
1240 
1241 static int
1242 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1243     bool audit)
1244 {
1245 	int error;
1246 
1247 	FILEDESC_XLOCK_ASSERT(fdp);
1248 
1249 	/*
1250 	 * We now hold the fp reference that used to be owned by the
1251 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1252 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1253 	 * added, and deleteing a knote for the new fd.
1254 	 */
1255 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1256 		knote_fdclose(td, fd);
1257 
1258 	/*
1259 	 * We need to notify mqueue if the object is of type mqueue.
1260 	 */
1261 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1262 		mq_fdclose(td, fd, fp);
1263 	FILEDESC_XUNLOCK(fdp);
1264 
1265 #ifdef AUDIT
1266 	if (AUDITING_TD(td) && audit)
1267 		audit_sysclose(td, fd, fp);
1268 #endif
1269 	error = closef(fp, td);
1270 
1271 	/*
1272 	 * All paths leading up to closefp() will have already removed or
1273 	 * replaced the fd in the filedesc table, so a restart would not
1274 	 * operate on the same file.
1275 	 */
1276 	if (error == ERESTART)
1277 		error = EINTR;
1278 
1279 	return (error);
1280 }
1281 
1282 static int
1283 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1284     bool holdleaders, bool audit)
1285 {
1286 	int error;
1287 
1288 	FILEDESC_XLOCK_ASSERT(fdp);
1289 
1290 	if (holdleaders) {
1291 		if (td->td_proc->p_fdtol != NULL) {
1292 			/*
1293 			 * Ask fdfree() to sleep to ensure that all relevant
1294 			 * process leaders can be traversed in closef().
1295 			 */
1296 			fdp->fd_holdleaderscount++;
1297 		} else {
1298 			holdleaders = false;
1299 		}
1300 	}
1301 
1302 	error = closefp_impl(fdp, fd, fp, td, audit);
1303 	if (holdleaders) {
1304 		FILEDESC_XLOCK(fdp);
1305 		fdp->fd_holdleaderscount--;
1306 		if (fdp->fd_holdleaderscount == 0 &&
1307 		    fdp->fd_holdleaderswakeup != 0) {
1308 			fdp->fd_holdleaderswakeup = 0;
1309 			wakeup(&fdp->fd_holdleaderscount);
1310 		}
1311 		FILEDESC_XUNLOCK(fdp);
1312 	}
1313 	return (error);
1314 }
1315 
1316 static int
1317 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1318     bool holdleaders, bool audit)
1319 {
1320 
1321 	FILEDESC_XLOCK_ASSERT(fdp);
1322 
1323 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1324 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1325 	} else {
1326 		return (closefp_impl(fdp, fd, fp, td, audit));
1327 	}
1328 }
1329 
1330 /*
1331  * Close a file descriptor.
1332  */
1333 #ifndef _SYS_SYSPROTO_H_
1334 struct close_args {
1335 	int     fd;
1336 };
1337 #endif
1338 /* ARGSUSED */
1339 int
1340 sys_close(struct thread *td, struct close_args *uap)
1341 {
1342 
1343 	return (kern_close(td, uap->fd));
1344 }
1345 
1346 int
1347 kern_close(struct thread *td, int fd)
1348 {
1349 	struct filedesc *fdp;
1350 	struct file *fp;
1351 
1352 	fdp = td->td_proc->p_fd;
1353 
1354 	FILEDESC_XLOCK(fdp);
1355 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1356 		FILEDESC_XUNLOCK(fdp);
1357 		return (EBADF);
1358 	}
1359 	fdfree(fdp, fd);
1360 
1361 	/* closefp() drops the FILEDESC lock for us. */
1362 	return (closefp(fdp, fd, fp, td, true, true));
1363 }
1364 
1365 int
1366 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1367 {
1368 	struct filedesc *fdp;
1369 	const struct fdescenttbl *fdt;
1370 	struct file *fp;
1371 	int fd;
1372 
1373 	/*
1374 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1375 	 * open should not be a usage error.  From a close_range() perspective,
1376 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1377 	 * be a usage error as all fd above 3 are in-fact already closed.
1378 	 */
1379 	if (highfd < lowfd) {
1380 		return (EINVAL);
1381 	}
1382 
1383 	fdp = td->td_proc->p_fd;
1384 	FILEDESC_XLOCK(fdp);
1385 	fdt = atomic_load_ptr(&fdp->fd_files);
1386 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1387 	fd = lowfd;
1388 	if (__predict_false(fd > highfd)) {
1389 		goto out_locked;
1390 	}
1391 	for (;;) {
1392 		fp = fdt->fdt_ofiles[fd].fde_file;
1393 		if (fp == NULL) {
1394 			if (fd == highfd)
1395 				goto out_locked;
1396 		} else {
1397 			fdfree(fdp, fd);
1398 			(void) closefp(fdp, fd, fp, td, true, true);
1399 			if (fd == highfd)
1400 				goto out_unlocked;
1401 			FILEDESC_XLOCK(fdp);
1402 			fdt = atomic_load_ptr(&fdp->fd_files);
1403 		}
1404 		fd++;
1405 	}
1406 out_locked:
1407 	FILEDESC_XUNLOCK(fdp);
1408 out_unlocked:
1409 	return (0);
1410 }
1411 
1412 #ifndef _SYS_SYSPROTO_H_
1413 struct close_range_args {
1414 	u_int	lowfd;
1415 	u_int	highfd;
1416 	int	flags;
1417 };
1418 #endif
1419 int
1420 sys_close_range(struct thread *td, struct close_range_args *uap)
1421 {
1422 
1423 	AUDIT_ARG_FD(uap->lowfd);
1424 	AUDIT_ARG_CMD(uap->highfd);
1425 	AUDIT_ARG_FFLAGS(uap->flags);
1426 
1427 	/* No flags currently defined */
1428 	if (uap->flags != 0)
1429 		return (EINVAL);
1430 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1431 }
1432 
1433 #ifdef COMPAT_FREEBSD12
1434 /*
1435  * Close open file descriptors.
1436  */
1437 #ifndef _SYS_SYSPROTO_H_
1438 struct freebsd12_closefrom_args {
1439 	int	lowfd;
1440 };
1441 #endif
1442 /* ARGSUSED */
1443 int
1444 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1445 {
1446 	u_int lowfd;
1447 
1448 	AUDIT_ARG_FD(uap->lowfd);
1449 
1450 	/*
1451 	 * Treat negative starting file descriptor values identical to
1452 	 * closefrom(0) which closes all files.
1453 	 */
1454 	lowfd = MAX(0, uap->lowfd);
1455 	return (kern_close_range(td, lowfd, ~0U));
1456 }
1457 #endif	/* COMPAT_FREEBSD12 */
1458 
1459 #if defined(COMPAT_43)
1460 /*
1461  * Return status information about a file descriptor.
1462  */
1463 #ifndef _SYS_SYSPROTO_H_
1464 struct ofstat_args {
1465 	int	fd;
1466 	struct	ostat *sb;
1467 };
1468 #endif
1469 /* ARGSUSED */
1470 int
1471 ofstat(struct thread *td, struct ofstat_args *uap)
1472 {
1473 	struct ostat oub;
1474 	struct stat ub;
1475 	int error;
1476 
1477 	error = kern_fstat(td, uap->fd, &ub);
1478 	if (error == 0) {
1479 		cvtstat(&ub, &oub);
1480 		error = copyout(&oub, uap->sb, sizeof(oub));
1481 	}
1482 	return (error);
1483 }
1484 #endif /* COMPAT_43 */
1485 
1486 #if defined(COMPAT_FREEBSD11)
1487 int
1488 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1489 {
1490 	struct stat sb;
1491 	struct freebsd11_stat osb;
1492 	int error;
1493 
1494 	error = kern_fstat(td, uap->fd, &sb);
1495 	if (error != 0)
1496 		return (error);
1497 	error = freebsd11_cvtstat(&sb, &osb);
1498 	if (error == 0)
1499 		error = copyout(&osb, uap->sb, sizeof(osb));
1500 	return (error);
1501 }
1502 #endif	/* COMPAT_FREEBSD11 */
1503 
1504 /*
1505  * Return status information about a file descriptor.
1506  */
1507 #ifndef _SYS_SYSPROTO_H_
1508 struct fstat_args {
1509 	int	fd;
1510 	struct	stat *sb;
1511 };
1512 #endif
1513 /* ARGSUSED */
1514 int
1515 sys_fstat(struct thread *td, struct fstat_args *uap)
1516 {
1517 	struct stat ub;
1518 	int error;
1519 
1520 	error = kern_fstat(td, uap->fd, &ub);
1521 	if (error == 0)
1522 		error = copyout(&ub, uap->sb, sizeof(ub));
1523 	return (error);
1524 }
1525 
1526 int
1527 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1528 {
1529 	struct file *fp;
1530 	int error;
1531 
1532 	AUDIT_ARG_FD(fd);
1533 
1534 	error = fget(td, fd, &cap_fstat_rights, &fp);
1535 	if (__predict_false(error != 0))
1536 		return (error);
1537 
1538 	AUDIT_ARG_FILE(td->td_proc, fp);
1539 
1540 	error = fo_stat(fp, sbp, td->td_ucred, td);
1541 	fdrop(fp, td);
1542 #ifdef __STAT_TIME_T_EXT
1543 	sbp->st_atim_ext = 0;
1544 	sbp->st_mtim_ext = 0;
1545 	sbp->st_ctim_ext = 0;
1546 	sbp->st_btim_ext = 0;
1547 #endif
1548 #ifdef KTRACE
1549 	if (KTRPOINT(td, KTR_STRUCT))
1550 		ktrstat_error(sbp, error);
1551 #endif
1552 	return (error);
1553 }
1554 
1555 #if defined(COMPAT_FREEBSD11)
1556 /*
1557  * Return status information about a file descriptor.
1558  */
1559 #ifndef _SYS_SYSPROTO_H_
1560 struct freebsd11_nfstat_args {
1561 	int	fd;
1562 	struct	nstat *sb;
1563 };
1564 #endif
1565 /* ARGSUSED */
1566 int
1567 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1568 {
1569 	struct nstat nub;
1570 	struct stat ub;
1571 	int error;
1572 
1573 	error = kern_fstat(td, uap->fd, &ub);
1574 	if (error == 0) {
1575 		freebsd11_cvtnstat(&ub, &nub);
1576 		error = copyout(&nub, uap->sb, sizeof(nub));
1577 	}
1578 	return (error);
1579 }
1580 #endif /* COMPAT_FREEBSD11 */
1581 
1582 /*
1583  * Return pathconf information about a file descriptor.
1584  */
1585 #ifndef _SYS_SYSPROTO_H_
1586 struct fpathconf_args {
1587 	int	fd;
1588 	int	name;
1589 };
1590 #endif
1591 /* ARGSUSED */
1592 int
1593 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1594 {
1595 	long value;
1596 	int error;
1597 
1598 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1599 	if (error == 0)
1600 		td->td_retval[0] = value;
1601 	return (error);
1602 }
1603 
1604 int
1605 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1606 {
1607 	struct file *fp;
1608 	struct vnode *vp;
1609 	int error;
1610 
1611 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1612 	if (error != 0)
1613 		return (error);
1614 
1615 	if (name == _PC_ASYNC_IO) {
1616 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1617 		goto out;
1618 	}
1619 	vp = fp->f_vnode;
1620 	if (vp != NULL) {
1621 		vn_lock(vp, LK_SHARED | LK_RETRY);
1622 		error = VOP_PATHCONF(vp, name, valuep);
1623 		VOP_UNLOCK(vp);
1624 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1625 		if (name != _PC_PIPE_BUF) {
1626 			error = EINVAL;
1627 		} else {
1628 			*valuep = PIPE_BUF;
1629 			error = 0;
1630 		}
1631 	} else {
1632 		error = EOPNOTSUPP;
1633 	}
1634 out:
1635 	fdrop(fp, td);
1636 	return (error);
1637 }
1638 
1639 /*
1640  * Copy filecaps structure allocating memory for ioctls array if needed.
1641  *
1642  * The last parameter indicates whether the fdtable is locked. If it is not and
1643  * ioctls are encountered, copying fails and the caller must lock the table.
1644  *
1645  * Note that if the table was not locked, the caller has to check the relevant
1646  * sequence counter to determine whether the operation was successful.
1647  */
1648 bool
1649 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1650 {
1651 	size_t size;
1652 
1653 	if (src->fc_ioctls != NULL && !locked)
1654 		return (false);
1655 	memcpy(dst, src, sizeof(*src));
1656 	if (src->fc_ioctls == NULL)
1657 		return (true);
1658 
1659 	KASSERT(src->fc_nioctls > 0,
1660 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1661 
1662 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1663 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1664 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1665 	return (true);
1666 }
1667 
1668 static u_long *
1669 filecaps_copy_prep(const struct filecaps *src)
1670 {
1671 	u_long *ioctls;
1672 	size_t size;
1673 
1674 	if (__predict_true(src->fc_ioctls == NULL))
1675 		return (NULL);
1676 
1677 	KASSERT(src->fc_nioctls > 0,
1678 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1679 
1680 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1681 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1682 	return (ioctls);
1683 }
1684 
1685 static void
1686 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1687     u_long *ioctls)
1688 {
1689 	size_t size;
1690 
1691 	*dst = *src;
1692 	if (__predict_true(src->fc_ioctls == NULL)) {
1693 		MPASS(ioctls == NULL);
1694 		return;
1695 	}
1696 
1697 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1698 	dst->fc_ioctls = ioctls;
1699 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1700 }
1701 
1702 /*
1703  * Move filecaps structure to the new place and clear the old place.
1704  */
1705 void
1706 filecaps_move(struct filecaps *src, struct filecaps *dst)
1707 {
1708 
1709 	*dst = *src;
1710 	bzero(src, sizeof(*src));
1711 }
1712 
1713 /*
1714  * Fill the given filecaps structure with full rights.
1715  */
1716 static void
1717 filecaps_fill(struct filecaps *fcaps)
1718 {
1719 
1720 	CAP_ALL(&fcaps->fc_rights);
1721 	fcaps->fc_ioctls = NULL;
1722 	fcaps->fc_nioctls = -1;
1723 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1724 }
1725 
1726 /*
1727  * Free memory allocated within filecaps structure.
1728  */
1729 void
1730 filecaps_free(struct filecaps *fcaps)
1731 {
1732 
1733 	free(fcaps->fc_ioctls, M_FILECAPS);
1734 	bzero(fcaps, sizeof(*fcaps));
1735 }
1736 
1737 static u_long *
1738 filecaps_free_prep(struct filecaps *fcaps)
1739 {
1740 	u_long *ioctls;
1741 
1742 	ioctls = fcaps->fc_ioctls;
1743 	bzero(fcaps, sizeof(*fcaps));
1744 	return (ioctls);
1745 }
1746 
1747 static void
1748 filecaps_free_finish(u_long *ioctls)
1749 {
1750 
1751 	free(ioctls, M_FILECAPS);
1752 }
1753 
1754 /*
1755  * Validate the given filecaps structure.
1756  */
1757 static void
1758 filecaps_validate(const struct filecaps *fcaps, const char *func)
1759 {
1760 
1761 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1762 	    ("%s: invalid rights", func));
1763 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1764 	    ("%s: invalid fcntls", func));
1765 	KASSERT(fcaps->fc_fcntls == 0 ||
1766 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1767 	    ("%s: fcntls without CAP_FCNTL", func));
1768 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1769 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1770 	    ("%s: invalid ioctls", func));
1771 	KASSERT(fcaps->fc_nioctls == 0 ||
1772 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1773 	    ("%s: ioctls without CAP_IOCTL", func));
1774 }
1775 
1776 static void
1777 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1778 {
1779 	int nfd1;
1780 
1781 	FILEDESC_XLOCK_ASSERT(fdp);
1782 
1783 	nfd1 = fdp->fd_nfiles * 2;
1784 	if (nfd1 < nfd)
1785 		nfd1 = nfd;
1786 	fdgrowtable(fdp, nfd1);
1787 }
1788 
1789 /*
1790  * Grow the file table to accommodate (at least) nfd descriptors.
1791  */
1792 static void
1793 fdgrowtable(struct filedesc *fdp, int nfd)
1794 {
1795 	struct filedesc0 *fdp0;
1796 	struct freetable *ft;
1797 	struct fdescenttbl *ntable;
1798 	struct fdescenttbl *otable;
1799 	int nnfiles, onfiles;
1800 	NDSLOTTYPE *nmap, *omap;
1801 
1802 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1803 
1804 	/* save old values */
1805 	onfiles = fdp->fd_nfiles;
1806 	otable = fdp->fd_files;
1807 	omap = fdp->fd_map;
1808 
1809 	/* compute the size of the new table */
1810 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1811 	if (nnfiles <= onfiles)
1812 		/* the table is already large enough */
1813 		return;
1814 
1815 	/*
1816 	 * Allocate a new table.  We need enough space for the number of
1817 	 * entries, file entries themselves and the struct freetable we will use
1818 	 * when we decommission the table and place it on the freelist.
1819 	 * We place the struct freetable in the middle so we don't have
1820 	 * to worry about padding.
1821 	 */
1822 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1823 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1824 	    sizeof(struct freetable),
1825 	    M_FILEDESC, M_ZERO | M_WAITOK);
1826 	/* copy the old data */
1827 	ntable->fdt_nfiles = nnfiles;
1828 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1829 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1830 
1831 	/*
1832 	 * Allocate a new map only if the old is not large enough.  It will
1833 	 * grow at a slower rate than the table as it can map more
1834 	 * entries than the table can hold.
1835 	 */
1836 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1837 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1838 		    M_ZERO | M_WAITOK);
1839 		/* copy over the old data and update the pointer */
1840 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1841 		fdp->fd_map = nmap;
1842 	}
1843 
1844 	/*
1845 	 * Make sure that ntable is correctly initialized before we replace
1846 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1847 	 * data.
1848 	 */
1849 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1850 
1851 	/*
1852 	 * Free the old file table when not shared by other threads or processes.
1853 	 * The old file table is considered to be shared when either are true:
1854 	 * - The process has more than one thread.
1855 	 * - The file descriptor table has been shared via fdshare().
1856 	 *
1857 	 * When shared, the old file table will be placed on a freelist
1858 	 * which will be processed when the struct filedesc is released.
1859 	 *
1860 	 * Note that if onfiles == NDFILE, we're dealing with the original
1861 	 * static allocation contained within (struct filedesc0 *)fdp,
1862 	 * which must not be freed.
1863 	 */
1864 	if (onfiles > NDFILE) {
1865 		/*
1866 		 * Note we may be called here from fdinit while allocating a
1867 		 * table for a new process in which case ->p_fd points
1868 		 * elsewhere.
1869 		 */
1870 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1871 			free(otable, M_FILEDESC);
1872 		} else {
1873 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1874 			fdp0 = (struct filedesc0 *)fdp;
1875 			ft->ft_table = otable;
1876 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1877 		}
1878 	}
1879 	/*
1880 	 * The map does not have the same possibility of threads still
1881 	 * holding references to it.  So always free it as long as it
1882 	 * does not reference the original static allocation.
1883 	 */
1884 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1885 		free(omap, M_FILEDESC);
1886 }
1887 
1888 /*
1889  * Allocate a file descriptor for the process.
1890  */
1891 int
1892 fdalloc(struct thread *td, int minfd, int *result)
1893 {
1894 	struct proc *p = td->td_proc;
1895 	struct filedesc *fdp = p->p_fd;
1896 	int fd, maxfd, allocfd;
1897 #ifdef RACCT
1898 	int error;
1899 #endif
1900 
1901 	FILEDESC_XLOCK_ASSERT(fdp);
1902 
1903 	if (fdp->fd_freefile > minfd)
1904 		minfd = fdp->fd_freefile;
1905 
1906 	maxfd = getmaxfd(td);
1907 
1908 	/*
1909 	 * Search the bitmap for a free descriptor starting at minfd.
1910 	 * If none is found, grow the file table.
1911 	 */
1912 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1913 	if (__predict_false(fd >= maxfd))
1914 		return (EMFILE);
1915 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1916 		allocfd = min(fd * 2, maxfd);
1917 #ifdef RACCT
1918 		if (RACCT_ENABLED()) {
1919 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1920 			if (error != 0)
1921 				return (EMFILE);
1922 		}
1923 #endif
1924 		/*
1925 		 * fd is already equal to first free descriptor >= minfd, so
1926 		 * we only need to grow the table and we are done.
1927 		 */
1928 		fdgrowtable_exp(fdp, allocfd);
1929 	}
1930 
1931 	/*
1932 	 * Perform some sanity checks, then mark the file descriptor as
1933 	 * used and return it to the caller.
1934 	 */
1935 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1936 	    ("invalid descriptor %d", fd));
1937 	KASSERT(!fdisused(fdp, fd),
1938 	    ("fd_first_free() returned non-free descriptor"));
1939 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1940 	    ("file descriptor isn't free"));
1941 	fdused(fdp, fd);
1942 	*result = fd;
1943 	return (0);
1944 }
1945 
1946 /*
1947  * Allocate n file descriptors for the process.
1948  */
1949 int
1950 fdallocn(struct thread *td, int minfd, int *fds, int n)
1951 {
1952 	struct proc *p = td->td_proc;
1953 	struct filedesc *fdp = p->p_fd;
1954 	int i;
1955 
1956 	FILEDESC_XLOCK_ASSERT(fdp);
1957 
1958 	for (i = 0; i < n; i++)
1959 		if (fdalloc(td, 0, &fds[i]) != 0)
1960 			break;
1961 
1962 	if (i < n) {
1963 		for (i--; i >= 0; i--)
1964 			fdunused(fdp, fds[i]);
1965 		return (EMFILE);
1966 	}
1967 
1968 	return (0);
1969 }
1970 
1971 /*
1972  * Create a new open file structure and allocate a file descriptor for the
1973  * process that refers to it.  We add one reference to the file for the
1974  * descriptor table and one reference for resultfp. This is to prevent us
1975  * being preempted and the entry in the descriptor table closed after we
1976  * release the FILEDESC lock.
1977  */
1978 int
1979 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1980     struct filecaps *fcaps)
1981 {
1982 	struct file *fp;
1983 	int error, fd;
1984 
1985 	MPASS(resultfp != NULL);
1986 	MPASS(resultfd != NULL);
1987 
1988 	error = _falloc_noinstall(td, &fp, 2);
1989 	if (__predict_false(error != 0)) {
1990 		return (error);
1991 	}
1992 
1993 	error = finstall_refed(td, fp, &fd, flags, fcaps);
1994 	if (__predict_false(error != 0)) {
1995 		falloc_abort(td, fp);
1996 		return (error);
1997 	}
1998 
1999 	*resultfp = fp;
2000 	*resultfd = fd;
2001 
2002 	return (0);
2003 }
2004 
2005 /*
2006  * Create a new open file structure without allocating a file descriptor.
2007  */
2008 int
2009 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2010 {
2011 	struct file *fp;
2012 	int maxuserfiles = maxfiles - (maxfiles / 20);
2013 	int openfiles_new;
2014 	static struct timeval lastfail;
2015 	static int curfail;
2016 
2017 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2018 	MPASS(n > 0);
2019 
2020 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2021 	if ((openfiles_new >= maxuserfiles &&
2022 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2023 	    openfiles_new >= maxfiles) {
2024 		atomic_subtract_int(&openfiles, 1);
2025 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2026 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2027 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2028 		}
2029 		return (ENFILE);
2030 	}
2031 	fp = uma_zalloc(file_zone, M_WAITOK);
2032 	bzero(fp, sizeof(*fp));
2033 	refcount_init(&fp->f_count, n);
2034 	fp->f_cred = crhold(td->td_ucred);
2035 	fp->f_ops = &badfileops;
2036 	*resultfp = fp;
2037 	return (0);
2038 }
2039 
2040 void
2041 falloc_abort(struct thread *td, struct file *fp)
2042 {
2043 
2044 	/*
2045 	 * For assertion purposes.
2046 	 */
2047 	refcount_init(&fp->f_count, 0);
2048 	_fdrop(fp, td);
2049 }
2050 
2051 /*
2052  * Install a file in a file descriptor table.
2053  */
2054 void
2055 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2056     struct filecaps *fcaps)
2057 {
2058 	struct filedescent *fde;
2059 
2060 	MPASS(fp != NULL);
2061 	if (fcaps != NULL)
2062 		filecaps_validate(fcaps, __func__);
2063 	FILEDESC_XLOCK_ASSERT(fdp);
2064 
2065 	fde = &fdp->fd_ofiles[fd];
2066 #ifdef CAPABILITIES
2067 	seqc_write_begin(&fde->fde_seqc);
2068 #endif
2069 	fde->fde_file = fp;
2070 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2071 	if (fcaps != NULL)
2072 		filecaps_move(fcaps, &fde->fde_caps);
2073 	else
2074 		filecaps_fill(&fde->fde_caps);
2075 #ifdef CAPABILITIES
2076 	seqc_write_end(&fde->fde_seqc);
2077 #endif
2078 }
2079 
2080 int
2081 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2082     struct filecaps *fcaps)
2083 {
2084 	struct filedesc *fdp = td->td_proc->p_fd;
2085 	int error;
2086 
2087 	MPASS(fd != NULL);
2088 
2089 	FILEDESC_XLOCK(fdp);
2090 	error = fdalloc(td, 0, fd);
2091 	if (__predict_true(error == 0)) {
2092 		_finstall(fdp, fp, *fd, flags, fcaps);
2093 	}
2094 	FILEDESC_XUNLOCK(fdp);
2095 	return (error);
2096 }
2097 
2098 int
2099 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2100     struct filecaps *fcaps)
2101 {
2102 	int error;
2103 
2104 	MPASS(fd != NULL);
2105 
2106 	if (!fhold(fp))
2107 		return (EBADF);
2108 	error = finstall_refed(td, fp, fd, flags, fcaps);
2109 	if (__predict_false(error != 0)) {
2110 		fdrop(fp, td);
2111 	}
2112 	return (error);
2113 }
2114 
2115 /*
2116  * Build a new filedesc structure from another.
2117  *
2118  * If fdp is not NULL, return with it shared locked.
2119  */
2120 struct filedesc *
2121 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2122 {
2123 	struct filedesc0 *newfdp0;
2124 	struct filedesc *newfdp;
2125 
2126 	if (prepfiles)
2127 		MPASS(lastfile != NULL);
2128 	else
2129 		MPASS(lastfile == NULL);
2130 
2131 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2132 	newfdp = &newfdp0->fd_fd;
2133 
2134 	/* Create the file descriptor table. */
2135 	FILEDESC_LOCK_INIT(newfdp);
2136 	refcount_init(&newfdp->fd_refcnt, 1);
2137 	refcount_init(&newfdp->fd_holdcnt, 1);
2138 	newfdp->fd_map = newfdp0->fd_dmap;
2139 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2140 	newfdp->fd_files->fdt_nfiles = NDFILE;
2141 
2142 	if (fdp == NULL)
2143 		return (newfdp);
2144 
2145 	FILEDESC_SLOCK(fdp);
2146 	if (!prepfiles) {
2147 		FILEDESC_SUNLOCK(fdp);
2148 		return (newfdp);
2149 	}
2150 
2151 	for (;;) {
2152 		*lastfile = fdlastfile(fdp);
2153 		if (*lastfile < newfdp->fd_nfiles)
2154 			break;
2155 		FILEDESC_SUNLOCK(fdp);
2156 		fdgrowtable(newfdp, *lastfile + 1);
2157 		FILEDESC_SLOCK(fdp);
2158 	}
2159 
2160 	return (newfdp);
2161 }
2162 
2163 /*
2164  * Build a pwddesc structure from another.
2165  * Copy the current, root, and jail root vnode references.
2166  *
2167  * If pdp is not NULL, return with it shared locked.
2168  */
2169 struct pwddesc *
2170 pdinit(struct pwddesc *pdp, bool keeplock)
2171 {
2172 	struct pwddesc *newpdp;
2173 	struct pwd *newpwd;
2174 
2175 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2176 
2177 	PWDDESC_LOCK_INIT(newpdp);
2178 	refcount_init(&newpdp->pd_refcount, 1);
2179 	newpdp->pd_cmask = CMASK;
2180 
2181 	if (pdp == NULL) {
2182 		newpwd = pwd_alloc();
2183 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2184 		return (newpdp);
2185 	}
2186 
2187 	PWDDESC_XLOCK(pdp);
2188 	newpwd = pwd_hold_pwddesc(pdp);
2189 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2190 	if (!keeplock)
2191 		PWDDESC_XUNLOCK(pdp);
2192 	return (newpdp);
2193 }
2194 
2195 /*
2196  * Hold either filedesc or pwddesc of the passed process.
2197  *
2198  * The process lock is used to synchronize against the target exiting and
2199  * freeing the data.
2200  *
2201  * Clearing can be ilustrated in 3 steps:
2202  * 1. set the pointer to NULL. Either routine can race against it, hence
2203  *   atomic_load_ptr.
2204  * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2205  *   race to either still see the pointer or find NULL. It is still safe to
2206  *   grab a reference as clearing is stalled.
2207  * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2208  *   guaranteed to see NULL, making it safe to finish clearing
2209  */
2210 static struct filedesc *
2211 fdhold(struct proc *p)
2212 {
2213 	struct filedesc *fdp;
2214 
2215 	PROC_LOCK_ASSERT(p, MA_OWNED);
2216 	fdp = atomic_load_ptr(&p->p_fd);
2217 	if (fdp != NULL)
2218 		refcount_acquire(&fdp->fd_holdcnt);
2219 	return (fdp);
2220 }
2221 
2222 static struct pwddesc *
2223 pdhold(struct proc *p)
2224 {
2225 	struct pwddesc *pdp;
2226 
2227 	PROC_LOCK_ASSERT(p, MA_OWNED);
2228 	pdp = atomic_load_ptr(&p->p_pd);
2229 	if (pdp != NULL)
2230 		refcount_acquire(&pdp->pd_refcount);
2231 	return (pdp);
2232 }
2233 
2234 static void
2235 fddrop(struct filedesc *fdp)
2236 {
2237 
2238 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2239 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2240 			return;
2241 	}
2242 
2243 	FILEDESC_LOCK_DESTROY(fdp);
2244 	uma_zfree(filedesc0_zone, fdp);
2245 }
2246 
2247 static void
2248 pddrop(struct pwddesc *pdp)
2249 {
2250 	struct pwd *pwd;
2251 
2252 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2253 		return;
2254 
2255 	PWDDESC_XLOCK(pdp);
2256 	if (refcount_release(&pdp->pd_refcount) == 0) {
2257 		PWDDESC_XUNLOCK(pdp);
2258 		return;
2259 	}
2260 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2261 	pwd_set(pdp, NULL);
2262 	PWDDESC_XUNLOCK(pdp);
2263 	pwd_drop(pwd);
2264 
2265 	PWDDESC_LOCK_DESTROY(pdp);
2266 	free(pdp, M_PWDDESC);
2267 }
2268 
2269 /*
2270  * Share a filedesc structure.
2271  */
2272 struct filedesc *
2273 fdshare(struct filedesc *fdp)
2274 {
2275 
2276 	refcount_acquire(&fdp->fd_refcnt);
2277 	return (fdp);
2278 }
2279 
2280 /*
2281  * Share a pwddesc structure.
2282  */
2283 struct pwddesc *
2284 pdshare(struct pwddesc *pdp)
2285 {
2286 	refcount_acquire(&pdp->pd_refcount);
2287 	return (pdp);
2288 }
2289 
2290 /*
2291  * Unshare a filedesc structure, if necessary by making a copy
2292  */
2293 void
2294 fdunshare(struct thread *td)
2295 {
2296 	struct filedesc *tmp;
2297 	struct proc *p = td->td_proc;
2298 
2299 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2300 		return;
2301 
2302 	tmp = fdcopy(p->p_fd);
2303 	fdescfree(td);
2304 	p->p_fd = tmp;
2305 }
2306 
2307 /*
2308  * Unshare a pwddesc structure.
2309  */
2310 void
2311 pdunshare(struct thread *td)
2312 {
2313 	struct pwddesc *pdp;
2314 	struct proc *p;
2315 
2316 	p = td->td_proc;
2317 	/* Not shared. */
2318 	if (p->p_pd->pd_refcount == 1)
2319 		return;
2320 
2321 	pdp = pdcopy(p->p_pd);
2322 	pdescfree(td);
2323 	p->p_pd = pdp;
2324 }
2325 
2326 void
2327 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2328 {
2329 
2330 	fdescfree(td);
2331 	td->td_proc->p_fd = fdp;
2332 }
2333 
2334 /*
2335  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2336  * this is to ease callers, not catch errors.
2337  */
2338 struct filedesc *
2339 fdcopy(struct filedesc *fdp)
2340 {
2341 	struct filedesc *newfdp;
2342 	struct filedescent *nfde, *ofde;
2343 	int i, lastfile;
2344 
2345 	MPASS(fdp != NULL);
2346 
2347 	newfdp = fdinit(fdp, true, &lastfile);
2348 	/* copy all passable descriptors (i.e. not kqueue) */
2349 	newfdp->fd_freefile = -1;
2350 	for (i = 0; i <= lastfile; ++i) {
2351 		ofde = &fdp->fd_ofiles[i];
2352 		if (ofde->fde_file == NULL ||
2353 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2354 		    !fhold(ofde->fde_file)) {
2355 			if (newfdp->fd_freefile == -1)
2356 				newfdp->fd_freefile = i;
2357 			continue;
2358 		}
2359 		nfde = &newfdp->fd_ofiles[i];
2360 		*nfde = *ofde;
2361 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2362 		fdused_init(newfdp, i);
2363 	}
2364 	if (newfdp->fd_freefile == -1)
2365 		newfdp->fd_freefile = i;
2366 	FILEDESC_SUNLOCK(fdp);
2367 	return (newfdp);
2368 }
2369 
2370 /*
2371  * Copy a pwddesc structure.
2372  */
2373 struct pwddesc *
2374 pdcopy(struct pwddesc *pdp)
2375 {
2376 	struct pwddesc *newpdp;
2377 
2378 	MPASS(pdp != NULL);
2379 
2380 	newpdp = pdinit(pdp, true);
2381 	newpdp->pd_cmask = pdp->pd_cmask;
2382 	PWDDESC_XUNLOCK(pdp);
2383 	return (newpdp);
2384 }
2385 
2386 /*
2387  * Copies a filedesc structure, while remapping all file descriptors
2388  * stored inside using a translation table.
2389  *
2390  * File descriptors are copied over to the new file descriptor table,
2391  * regardless of whether the close-on-exec flag is set.
2392  */
2393 int
2394 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2395     struct filedesc **ret)
2396 {
2397 	struct filedesc *newfdp;
2398 	struct filedescent *nfde, *ofde;
2399 	int error, i, lastfile;
2400 
2401 	MPASS(fdp != NULL);
2402 
2403 	newfdp = fdinit(fdp, true, &lastfile);
2404 	if (nfds > lastfile + 1) {
2405 		/* New table cannot be larger than the old one. */
2406 		error = E2BIG;
2407 		goto bad;
2408 	}
2409 	/* Copy all passable descriptors (i.e. not kqueue). */
2410 	newfdp->fd_freefile = nfds;
2411 	for (i = 0; i < nfds; ++i) {
2412 		if (fds[i] < 0 || fds[i] > lastfile) {
2413 			/* File descriptor out of bounds. */
2414 			error = EBADF;
2415 			goto bad;
2416 		}
2417 		ofde = &fdp->fd_ofiles[fds[i]];
2418 		if (ofde->fde_file == NULL) {
2419 			/* Unused file descriptor. */
2420 			error = EBADF;
2421 			goto bad;
2422 		}
2423 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2424 			/* File descriptor cannot be passed. */
2425 			error = EINVAL;
2426 			goto bad;
2427 		}
2428 		if (!fhold(ofde->fde_file)) {
2429 			error = EBADF;
2430 			goto bad;
2431 		}
2432 		nfde = &newfdp->fd_ofiles[i];
2433 		*nfde = *ofde;
2434 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2435 		fdused_init(newfdp, i);
2436 	}
2437 	FILEDESC_SUNLOCK(fdp);
2438 	*ret = newfdp;
2439 	return (0);
2440 bad:
2441 	FILEDESC_SUNLOCK(fdp);
2442 	fdescfree_remapped(newfdp);
2443 	return (error);
2444 }
2445 
2446 /*
2447  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2448  * one of processes using it exits) and the table used to be shared.
2449  */
2450 static void
2451 fdclearlocks(struct thread *td)
2452 {
2453 	struct filedesc *fdp;
2454 	struct filedesc_to_leader *fdtol;
2455 	struct flock lf;
2456 	struct file *fp;
2457 	struct proc *p;
2458 	struct vnode *vp;
2459 	int i, lastfile;
2460 
2461 	p = td->td_proc;
2462 	fdp = p->p_fd;
2463 	fdtol = p->p_fdtol;
2464 	MPASS(fdtol != NULL);
2465 
2466 	FILEDESC_XLOCK(fdp);
2467 	KASSERT(fdtol->fdl_refcount > 0,
2468 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2469 	    fdtol->fdl_refcount));
2470 	if (fdtol->fdl_refcount == 1 &&
2471 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2472 		lastfile = fdlastfile(fdp);
2473 		for (i = 0; i <= lastfile; i++) {
2474 			fp = fdp->fd_ofiles[i].fde_file;
2475 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2476 			    !fhold(fp))
2477 				continue;
2478 			FILEDESC_XUNLOCK(fdp);
2479 			lf.l_whence = SEEK_SET;
2480 			lf.l_start = 0;
2481 			lf.l_len = 0;
2482 			lf.l_type = F_UNLCK;
2483 			vp = fp->f_vnode;
2484 			(void) VOP_ADVLOCK(vp,
2485 			    (caddr_t)p->p_leader, F_UNLCK,
2486 			    &lf, F_POSIX);
2487 			FILEDESC_XLOCK(fdp);
2488 			fdrop(fp, td);
2489 		}
2490 	}
2491 retry:
2492 	if (fdtol->fdl_refcount == 1) {
2493 		if (fdp->fd_holdleaderscount > 0 &&
2494 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2495 			/*
2496 			 * close() or kern_dup() has cleared a reference
2497 			 * in a shared file descriptor table.
2498 			 */
2499 			fdp->fd_holdleaderswakeup = 1;
2500 			sx_sleep(&fdp->fd_holdleaderscount,
2501 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2502 			goto retry;
2503 		}
2504 		if (fdtol->fdl_holdcount > 0) {
2505 			/*
2506 			 * Ensure that fdtol->fdl_leader remains
2507 			 * valid in closef().
2508 			 */
2509 			fdtol->fdl_wakeup = 1;
2510 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2511 			    "fdlhold", 0);
2512 			goto retry;
2513 		}
2514 	}
2515 	fdtol->fdl_refcount--;
2516 	if (fdtol->fdl_refcount == 0 &&
2517 	    fdtol->fdl_holdcount == 0) {
2518 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2519 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2520 	} else
2521 		fdtol = NULL;
2522 	p->p_fdtol = NULL;
2523 	FILEDESC_XUNLOCK(fdp);
2524 	if (fdtol != NULL)
2525 		free(fdtol, M_FILEDESC_TO_LEADER);
2526 }
2527 
2528 /*
2529  * Release a filedesc structure.
2530  */
2531 static void
2532 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2533 {
2534 	struct filedesc0 *fdp0;
2535 	struct freetable *ft, *tft;
2536 	struct filedescent *fde;
2537 	struct file *fp;
2538 	int i, lastfile;
2539 
2540 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2541 	    ("%s: fd table %p carries references", __func__, fdp));
2542 
2543 	/*
2544 	 * Serialize with threads iterating over the table, if any.
2545 	 */
2546 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2547 		FILEDESC_XLOCK(fdp);
2548 		FILEDESC_XUNLOCK(fdp);
2549 	}
2550 
2551 	lastfile = fdlastfile_single(fdp);
2552 	for (i = 0; i <= lastfile; i++) {
2553 		fde = &fdp->fd_ofiles[i];
2554 		fp = fde->fde_file;
2555 		if (fp != NULL) {
2556 			fdefree_last(fde);
2557 			if (needclose)
2558 				(void) closef(fp, td);
2559 			else
2560 				fdrop(fp, td);
2561 		}
2562 	}
2563 
2564 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2565 		free(fdp->fd_map, M_FILEDESC);
2566 	if (fdp->fd_nfiles > NDFILE)
2567 		free(fdp->fd_files, M_FILEDESC);
2568 
2569 	fdp0 = (struct filedesc0 *)fdp;
2570 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2571 		free(ft->ft_table, M_FILEDESC);
2572 
2573 	fddrop(fdp);
2574 }
2575 
2576 void
2577 fdescfree(struct thread *td)
2578 {
2579 	struct proc *p;
2580 	struct filedesc *fdp;
2581 
2582 	p = td->td_proc;
2583 	fdp = p->p_fd;
2584 	MPASS(fdp != NULL);
2585 
2586 #ifdef RACCT
2587 	if (RACCT_ENABLED())
2588 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2589 #endif
2590 
2591 	if (p->p_fdtol != NULL)
2592 		fdclearlocks(td);
2593 
2594 	/*
2595 	 * Check fdhold for an explanation.
2596 	 */
2597 	atomic_store_ptr(&p->p_fd, NULL);
2598 	atomic_thread_fence_seq_cst();
2599 	PROC_WAIT_UNLOCKED(p);
2600 
2601 	if (refcount_release(&fdp->fd_refcnt) == 0)
2602 		return;
2603 
2604 	fdescfree_fds(td, fdp, 1);
2605 }
2606 
2607 void
2608 pdescfree(struct thread *td)
2609 {
2610 	struct proc *p;
2611 	struct pwddesc *pdp;
2612 
2613 	p = td->td_proc;
2614 	pdp = p->p_pd;
2615 	MPASS(pdp != NULL);
2616 
2617 	/*
2618 	 * Check pdhold for an explanation.
2619 	 */
2620 	atomic_store_ptr(&p->p_pd, NULL);
2621 	atomic_thread_fence_seq_cst();
2622 	PROC_WAIT_UNLOCKED(p);
2623 
2624 	pddrop(pdp);
2625 }
2626 
2627 void
2628 fdescfree_remapped(struct filedesc *fdp)
2629 {
2630 #ifdef INVARIANTS
2631 	/* fdescfree_fds() asserts that fd_refcnt == 0. */
2632 	if (!refcount_release(&fdp->fd_refcnt))
2633 		panic("%s: fd table %p has extra references", __func__, fdp);
2634 #endif
2635 	fdescfree_fds(curthread, fdp, 0);
2636 }
2637 
2638 /*
2639  * For setugid programs, we don't want to people to use that setugidness
2640  * to generate error messages which write to a file which otherwise would
2641  * otherwise be off-limits to the process.  We check for filesystems where
2642  * the vnode can change out from under us after execve (like [lin]procfs).
2643  *
2644  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2645  * sufficient.  We also don't check for setugidness since we know we are.
2646  */
2647 static bool
2648 is_unsafe(struct file *fp)
2649 {
2650 	struct vnode *vp;
2651 
2652 	if (fp->f_type != DTYPE_VNODE)
2653 		return (false);
2654 
2655 	vp = fp->f_vnode;
2656 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2657 }
2658 
2659 /*
2660  * Make this setguid thing safe, if at all possible.
2661  */
2662 void
2663 fdsetugidsafety(struct thread *td)
2664 {
2665 	struct filedesc *fdp;
2666 	struct file *fp;
2667 	int i;
2668 
2669 	fdp = td->td_proc->p_fd;
2670 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2671 	    ("the fdtable should not be shared"));
2672 	MPASS(fdp->fd_nfiles >= 3);
2673 	for (i = 0; i <= 2; i++) {
2674 		fp = fdp->fd_ofiles[i].fde_file;
2675 		if (fp != NULL && is_unsafe(fp)) {
2676 			FILEDESC_XLOCK(fdp);
2677 			knote_fdclose(td, i);
2678 			/*
2679 			 * NULL-out descriptor prior to close to avoid
2680 			 * a race while close blocks.
2681 			 */
2682 			fdfree(fdp, i);
2683 			FILEDESC_XUNLOCK(fdp);
2684 			(void) closef(fp, td);
2685 		}
2686 	}
2687 }
2688 
2689 /*
2690  * If a specific file object occupies a specific file descriptor, close the
2691  * file descriptor entry and drop a reference on the file object.  This is a
2692  * convenience function to handle a subsequent error in a function that calls
2693  * falloc() that handles the race that another thread might have closed the
2694  * file descriptor out from under the thread creating the file object.
2695  */
2696 void
2697 fdclose(struct thread *td, struct file *fp, int idx)
2698 {
2699 	struct filedesc *fdp = td->td_proc->p_fd;
2700 
2701 	FILEDESC_XLOCK(fdp);
2702 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2703 		fdfree(fdp, idx);
2704 		FILEDESC_XUNLOCK(fdp);
2705 		fdrop(fp, td);
2706 	} else
2707 		FILEDESC_XUNLOCK(fdp);
2708 }
2709 
2710 /*
2711  * Close any files on exec?
2712  */
2713 void
2714 fdcloseexec(struct thread *td)
2715 {
2716 	struct filedesc *fdp;
2717 	struct filedescent *fde;
2718 	struct file *fp;
2719 	int i, lastfile;
2720 
2721 	fdp = td->td_proc->p_fd;
2722 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2723 	    ("the fdtable should not be shared"));
2724 	lastfile = fdlastfile_single(fdp);
2725 	for (i = 0; i <= lastfile; i++) {
2726 		fde = &fdp->fd_ofiles[i];
2727 		fp = fde->fde_file;
2728 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2729 		    (fde->fde_flags & UF_EXCLOSE))) {
2730 			FILEDESC_XLOCK(fdp);
2731 			fdfree(fdp, i);
2732 			(void) closefp(fdp, i, fp, td, false, false);
2733 			FILEDESC_UNLOCK_ASSERT(fdp);
2734 		}
2735 	}
2736 }
2737 
2738 /*
2739  * It is unsafe for set[ug]id processes to be started with file
2740  * descriptors 0..2 closed, as these descriptors are given implicit
2741  * significance in the Standard C library.  fdcheckstd() will create a
2742  * descriptor referencing /dev/null for each of stdin, stdout, and
2743  * stderr that is not already open.
2744  */
2745 int
2746 fdcheckstd(struct thread *td)
2747 {
2748 	struct filedesc *fdp;
2749 	register_t save;
2750 	int i, error, devnull;
2751 
2752 	fdp = td->td_proc->p_fd;
2753 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2754 	    ("the fdtable should not be shared"));
2755 	MPASS(fdp->fd_nfiles >= 3);
2756 	devnull = -1;
2757 	for (i = 0; i <= 2; i++) {
2758 		if (fdp->fd_ofiles[i].fde_file != NULL)
2759 			continue;
2760 
2761 		save = td->td_retval[0];
2762 		if (devnull != -1) {
2763 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2764 		} else {
2765 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2766 			    UIO_SYSSPACE, O_RDWR, 0);
2767 			if (error == 0) {
2768 				devnull = td->td_retval[0];
2769 				KASSERT(devnull == i, ("we didn't get our fd"));
2770 			}
2771 		}
2772 		td->td_retval[0] = save;
2773 		if (error != 0)
2774 			return (error);
2775 	}
2776 	return (0);
2777 }
2778 
2779 /*
2780  * Internal form of close.  Decrement reference count on file structure.
2781  * Note: td may be NULL when closing a file that was being passed in a
2782  * message.
2783  */
2784 int
2785 closef(struct file *fp, struct thread *td)
2786 {
2787 	struct vnode *vp;
2788 	struct flock lf;
2789 	struct filedesc_to_leader *fdtol;
2790 	struct filedesc *fdp;
2791 
2792 	MPASS(td != NULL);
2793 
2794 	/*
2795 	 * POSIX record locking dictates that any close releases ALL
2796 	 * locks owned by this process.  This is handled by setting
2797 	 * a flag in the unlock to free ONLY locks obeying POSIX
2798 	 * semantics, and not to free BSD-style file locks.
2799 	 * If the descriptor was in a message, POSIX-style locks
2800 	 * aren't passed with the descriptor, and the thread pointer
2801 	 * will be NULL.  Callers should be careful only to pass a
2802 	 * NULL thread pointer when there really is no owning
2803 	 * context that might have locks, or the locks will be
2804 	 * leaked.
2805 	 */
2806 	if (fp->f_type == DTYPE_VNODE) {
2807 		vp = fp->f_vnode;
2808 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2809 			lf.l_whence = SEEK_SET;
2810 			lf.l_start = 0;
2811 			lf.l_len = 0;
2812 			lf.l_type = F_UNLCK;
2813 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2814 			    F_UNLCK, &lf, F_POSIX);
2815 		}
2816 		fdtol = td->td_proc->p_fdtol;
2817 		if (fdtol != NULL) {
2818 			/*
2819 			 * Handle special case where file descriptor table is
2820 			 * shared between multiple process leaders.
2821 			 */
2822 			fdp = td->td_proc->p_fd;
2823 			FILEDESC_XLOCK(fdp);
2824 			for (fdtol = fdtol->fdl_next;
2825 			    fdtol != td->td_proc->p_fdtol;
2826 			    fdtol = fdtol->fdl_next) {
2827 				if ((fdtol->fdl_leader->p_flag &
2828 				    P_ADVLOCK) == 0)
2829 					continue;
2830 				fdtol->fdl_holdcount++;
2831 				FILEDESC_XUNLOCK(fdp);
2832 				lf.l_whence = SEEK_SET;
2833 				lf.l_start = 0;
2834 				lf.l_len = 0;
2835 				lf.l_type = F_UNLCK;
2836 				vp = fp->f_vnode;
2837 				(void) VOP_ADVLOCK(vp,
2838 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2839 				    F_POSIX);
2840 				FILEDESC_XLOCK(fdp);
2841 				fdtol->fdl_holdcount--;
2842 				if (fdtol->fdl_holdcount == 0 &&
2843 				    fdtol->fdl_wakeup != 0) {
2844 					fdtol->fdl_wakeup = 0;
2845 					wakeup(fdtol);
2846 				}
2847 			}
2848 			FILEDESC_XUNLOCK(fdp);
2849 		}
2850 	}
2851 	return (fdrop_close(fp, td));
2852 }
2853 
2854 /*
2855  * Hack for file descriptor passing code.
2856  */
2857 void
2858 closef_nothread(struct file *fp)
2859 {
2860 
2861 	fdrop(fp, NULL);
2862 }
2863 
2864 /*
2865  * Initialize the file pointer with the specified properties.
2866  *
2867  * The ops are set with release semantics to be certain that the flags, type,
2868  * and data are visible when ops is.  This is to prevent ops methods from being
2869  * called with bad data.
2870  */
2871 void
2872 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2873 {
2874 	fp->f_data = data;
2875 	fp->f_flag = flag;
2876 	fp->f_type = type;
2877 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2878 }
2879 
2880 void
2881 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2882 {
2883 	fp->f_seqcount[UIO_READ] = 1;
2884 	fp->f_seqcount[UIO_WRITE] = 1;
2885 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2886 	    data, ops);
2887 }
2888 
2889 int
2890 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2891     struct file **fpp, struct filecaps *havecapsp)
2892 {
2893 	struct filedescent *fde;
2894 	int error;
2895 
2896 	FILEDESC_LOCK_ASSERT(fdp);
2897 
2898 	fde = fdeget_locked(fdp, fd);
2899 	if (fde == NULL) {
2900 		error = EBADF;
2901 		goto out;
2902 	}
2903 
2904 #ifdef CAPABILITIES
2905 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2906 	if (error != 0)
2907 		goto out;
2908 #endif
2909 
2910 	if (havecapsp != NULL)
2911 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2912 
2913 	*fpp = fde->fde_file;
2914 
2915 	error = 0;
2916 out:
2917 	return (error);
2918 }
2919 
2920 int
2921 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2922     struct file **fpp, struct filecaps *havecapsp)
2923 {
2924 	struct filedesc *fdp = td->td_proc->p_fd;
2925 	int error;
2926 #ifndef CAPABILITIES
2927 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2928 	if (havecapsp != NULL && error == 0)
2929 		filecaps_fill(havecapsp);
2930 #else
2931 	struct file *fp;
2932 	seqc_t seq;
2933 
2934 	*fpp = NULL;
2935 	for (;;) {
2936 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2937 		if (error != 0)
2938 			return (error);
2939 
2940 		if (havecapsp != NULL) {
2941 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2942 			    havecapsp, false)) {
2943 				fdrop(fp, td);
2944 				goto get_locked;
2945 			}
2946 		}
2947 
2948 		if (!fd_modified(fdp, fd, seq))
2949 			break;
2950 		fdrop(fp, td);
2951 	}
2952 
2953 	*fpp = fp;
2954 	return (0);
2955 
2956 get_locked:
2957 	FILEDESC_SLOCK(fdp);
2958 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2959 	if (error == 0 && !fhold(*fpp))
2960 		error = EBADF;
2961 	FILEDESC_SUNLOCK(fdp);
2962 #endif
2963 	return (error);
2964 }
2965 
2966 #ifdef CAPABILITIES
2967 int
2968 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2969 {
2970 	const struct filedescent *fde;
2971 	const struct fdescenttbl *fdt;
2972 	struct filedesc *fdp;
2973 	struct file *fp;
2974 	struct vnode *vp;
2975 	const cap_rights_t *haverights;
2976 	cap_rights_t rights;
2977 	seqc_t seq;
2978 
2979 	VFS_SMR_ASSERT_ENTERED();
2980 
2981 	rights = *ndp->ni_rightsneeded;
2982 	cap_rights_set_one(&rights, CAP_LOOKUP);
2983 
2984 	fdp = curproc->p_fd;
2985 	fdt = fdp->fd_files;
2986 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2987 		return (EBADF);
2988 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2989 	fde = &fdt->fdt_ofiles[fd];
2990 	haverights = cap_rights_fde_inline(fde);
2991 	fp = fde->fde_file;
2992 	if (__predict_false(fp == NULL))
2993 		return (EAGAIN);
2994 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2995 		return (EAGAIN);
2996 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2997 	vp = fp->f_vnode;
2998 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2999 		return (EAGAIN);
3000 	}
3001 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
3002 		return (EAGAIN);
3003 	}
3004 	/*
3005 	 * Use an acquire barrier to force re-reading of fdt so it is
3006 	 * refreshed for verification.
3007 	 */
3008 	atomic_thread_fence_acq();
3009 	fdt = fdp->fd_files;
3010 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3011 		return (EAGAIN);
3012 	/*
3013 	 * If file descriptor doesn't have all rights,
3014 	 * all lookups relative to it must also be
3015 	 * strictly relative.
3016 	 *
3017 	 * Not yet supported by fast path.
3018 	 */
3019 	CAP_ALL(&rights);
3020 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3021 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3022 	    ndp->ni_filecaps.fc_nioctls != -1) {
3023 #ifdef notyet
3024 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3025 #else
3026 		return (EAGAIN);
3027 #endif
3028 	}
3029 	*vpp = vp;
3030 	return (0);
3031 }
3032 #else
3033 int
3034 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3035 {
3036 	const struct fdescenttbl *fdt;
3037 	struct filedesc *fdp;
3038 	struct file *fp;
3039 	struct vnode *vp;
3040 
3041 	VFS_SMR_ASSERT_ENTERED();
3042 
3043 	fdp = curproc->p_fd;
3044 	fdt = fdp->fd_files;
3045 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3046 		return (EBADF);
3047 	fp = fdt->fdt_ofiles[fd].fde_file;
3048 	if (__predict_false(fp == NULL))
3049 		return (EAGAIN);
3050 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
3051 	vp = fp->f_vnode;
3052 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3053 		return (EAGAIN);
3054 	}
3055 	/*
3056 	 * Use an acquire barrier to force re-reading of fdt so it is
3057 	 * refreshed for verification.
3058 	 */
3059 	atomic_thread_fence_acq();
3060 	fdt = fdp->fd_files;
3061 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3062 		return (EAGAIN);
3063 	filecaps_fill(&ndp->ni_filecaps);
3064 	*vpp = vp;
3065 	return (0);
3066 }
3067 #endif
3068 
3069 int
3070 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3071     struct file **fpp, seqc_t *seqp)
3072 {
3073 #ifdef CAPABILITIES
3074 	const struct filedescent *fde;
3075 #endif
3076 	const struct fdescenttbl *fdt;
3077 	struct file *fp;
3078 #ifdef CAPABILITIES
3079 	seqc_t seq;
3080 	cap_rights_t haverights;
3081 	int error;
3082 #endif
3083 
3084 	fdt = fdp->fd_files;
3085 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3086 		return (EBADF);
3087 	/*
3088 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
3089 	 * never raising a refcount above 0.  To accomplish this we have
3090 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3091 	 * must be re-verified once we acquire a reference to be certain
3092 	 * that the identity is still correct and we did not lose a race
3093 	 * due to preemption.
3094 	 */
3095 	for (;;) {
3096 #ifdef CAPABILITIES
3097 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3098 		fde = &fdt->fdt_ofiles[fd];
3099 		haverights = *cap_rights_fde_inline(fde);
3100 		fp = fde->fde_file;
3101 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3102 			continue;
3103 #else
3104 		fp = fdt->fdt_ofiles[fd].fde_file;
3105 #endif
3106 		if (fp == NULL)
3107 			return (EBADF);
3108 #ifdef CAPABILITIES
3109 		error = cap_check_inline(&haverights, needrightsp);
3110 		if (error != 0)
3111 			return (error);
3112 #endif
3113 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3114 			/*
3115 			 * Force a reload. Other thread could reallocate the
3116 			 * table before this fd was closed, so it is possible
3117 			 * that there is a stale fp pointer in cached version.
3118 			 */
3119 			fdt = atomic_load_ptr(&fdp->fd_files);
3120 			continue;
3121 		}
3122 		/*
3123 		 * Use an acquire barrier to force re-reading of fdt so it is
3124 		 * refreshed for verification.
3125 		 */
3126 		atomic_thread_fence_acq();
3127 		fdt = fdp->fd_files;
3128 #ifdef	CAPABILITIES
3129 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3130 #else
3131 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3132 #endif
3133 			break;
3134 		fdrop(fp, curthread);
3135 	}
3136 	*fpp = fp;
3137 	if (seqp != NULL) {
3138 #ifdef CAPABILITIES
3139 		*seqp = seq;
3140 #endif
3141 	}
3142 	return (0);
3143 }
3144 
3145 /*
3146  * See the comments in fget_unlocked_seq for an explanation of how this works.
3147  *
3148  * This is a simplified variant which bails out to the aforementioned routine
3149  * if anything goes wrong. In practice this only happens when userspace is
3150  * racing with itself.
3151  */
3152 int
3153 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3154     struct file **fpp)
3155 {
3156 #ifdef CAPABILITIES
3157 	const struct filedescent *fde;
3158 #endif
3159 	const struct fdescenttbl *fdt;
3160 	struct file *fp;
3161 #ifdef CAPABILITIES
3162 	seqc_t seq;
3163 	const cap_rights_t *haverights;
3164 #endif
3165 
3166 	fdt = fdp->fd_files;
3167 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3168 		return (EBADF);
3169 #ifdef CAPABILITIES
3170 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3171 	fde = &fdt->fdt_ofiles[fd];
3172 	haverights = cap_rights_fde_inline(fde);
3173 	fp = fde->fde_file;
3174 #else
3175 	fp = fdt->fdt_ofiles[fd].fde_file;
3176 #endif
3177 	if (__predict_false(fp == NULL))
3178 		goto out_fallback;
3179 #ifdef CAPABILITIES
3180 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3181 		goto out_fallback;
3182 #endif
3183 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3184 		goto out_fallback;
3185 
3186 	/*
3187 	 * Use an acquire barrier to force re-reading of fdt so it is
3188 	 * refreshed for verification.
3189 	 */
3190 	atomic_thread_fence_acq();
3191 	fdt = fdp->fd_files;
3192 #ifdef	CAPABILITIES
3193 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3194 #else
3195 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3196 #endif
3197 		goto out_fdrop;
3198 	*fpp = fp;
3199 	return (0);
3200 out_fdrop:
3201 	fdrop(fp, curthread);
3202 out_fallback:
3203 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3204 }
3205 
3206 /*
3207  * Translate fd -> file when the caller guarantees the file descriptor table
3208  * can't be changed by others.
3209  *
3210  * Note this does not mean the file object itself is only visible to the caller,
3211  * merely that it wont disappear without having to be referenced.
3212  *
3213  * Must be paired with fput_only_user.
3214  */
3215 #ifdef	CAPABILITIES
3216 int
3217 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3218     struct file **fpp)
3219 {
3220 	const struct filedescent *fde;
3221 	const struct fdescenttbl *fdt;
3222 	const cap_rights_t *haverights;
3223 	struct file *fp;
3224 	int error;
3225 
3226 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3227 
3228 	if (__predict_false(fd >= fdp->fd_nfiles))
3229 		return (EBADF);
3230 
3231 	fdt = fdp->fd_files;
3232 	fde = &fdt->fdt_ofiles[fd];
3233 	fp = fde->fde_file;
3234 	if (__predict_false(fp == NULL))
3235 		return (EBADF);
3236 	MPASS(refcount_load(&fp->f_count) > 0);
3237 	haverights = cap_rights_fde_inline(fde);
3238 	error = cap_check_inline(haverights, needrightsp);
3239 	if (__predict_false(error != 0))
3240 		return (error);
3241 	*fpp = fp;
3242 	return (0);
3243 }
3244 #else
3245 int
3246 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3247     struct file **fpp)
3248 {
3249 	struct file *fp;
3250 
3251 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3252 
3253 	if (__predict_false(fd >= fdp->fd_nfiles))
3254 		return (EBADF);
3255 
3256 	fp = fdp->fd_ofiles[fd].fde_file;
3257 	if (__predict_false(fp == NULL))
3258 		return (EBADF);
3259 
3260 	MPASS(refcount_load(&fp->f_count) > 0);
3261 	*fpp = fp;
3262 	return (0);
3263 }
3264 #endif
3265 
3266 /*
3267  * Extract the file pointer associated with the specified descriptor for the
3268  * current user process.
3269  *
3270  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3271  * returned.
3272  *
3273  * File's rights will be checked against the capability rights mask.
3274  *
3275  * If an error occurred the non-zero error is returned and *fpp is set to
3276  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3277  * responsible for fdrop().
3278  */
3279 static __inline int
3280 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3281     cap_rights_t *needrightsp)
3282 {
3283 	struct filedesc *fdp;
3284 	struct file *fp;
3285 	int error;
3286 
3287 	*fpp = NULL;
3288 	fdp = td->td_proc->p_fd;
3289 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3290 	if (__predict_false(error != 0))
3291 		return (error);
3292 	if (__predict_false(fp->f_ops == &badfileops)) {
3293 		fdrop(fp, td);
3294 		return (EBADF);
3295 	}
3296 
3297 	/*
3298 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3299 	 */
3300 	error = 0;
3301 	switch (flags) {
3302 	case FREAD:
3303 	case FWRITE:
3304 		if ((fp->f_flag & flags) == 0)
3305 			error = EBADF;
3306 		break;
3307 	case FEXEC:
3308 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3309 		    ((fp->f_flag & FWRITE) != 0))
3310 			error = EBADF;
3311 		break;
3312 	case 0:
3313 		break;
3314 	default:
3315 		KASSERT(0, ("wrong flags"));
3316 	}
3317 
3318 	if (error != 0) {
3319 		fdrop(fp, td);
3320 		return (error);
3321 	}
3322 
3323 	*fpp = fp;
3324 	return (0);
3325 }
3326 
3327 int
3328 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3329 {
3330 
3331 	return (_fget(td, fd, fpp, 0, rightsp));
3332 }
3333 
3334 int
3335 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3336     struct file **fpp)
3337 {
3338 	int error;
3339 #ifndef CAPABILITIES
3340 	error = _fget(td, fd, fpp, 0, rightsp);
3341 	if (maxprotp != NULL)
3342 		*maxprotp = VM_PROT_ALL;
3343 	return (error);
3344 #else
3345 	cap_rights_t fdrights;
3346 	struct filedesc *fdp;
3347 	struct file *fp;
3348 	seqc_t seq;
3349 
3350 	*fpp = NULL;
3351 	fdp = td->td_proc->p_fd;
3352 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3353 	for (;;) {
3354 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3355 		if (__predict_false(error != 0))
3356 			return (error);
3357 		if (__predict_false(fp->f_ops == &badfileops)) {
3358 			fdrop(fp, td);
3359 			return (EBADF);
3360 		}
3361 		if (maxprotp != NULL)
3362 			fdrights = *cap_rights(fdp, fd);
3363 		if (!fd_modified(fdp, fd, seq))
3364 			break;
3365 		fdrop(fp, td);
3366 	}
3367 
3368 	/*
3369 	 * If requested, convert capability rights to access flags.
3370 	 */
3371 	if (maxprotp != NULL)
3372 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3373 	*fpp = fp;
3374 	return (0);
3375 #endif
3376 }
3377 
3378 int
3379 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3380 {
3381 
3382 	return (_fget(td, fd, fpp, FREAD, rightsp));
3383 }
3384 
3385 int
3386 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3387 {
3388 
3389 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3390 }
3391 
3392 int
3393 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3394     struct file **fpp)
3395 {
3396 	struct filedesc *fdp = td->td_proc->p_fd;
3397 #ifndef CAPABILITIES
3398 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3399 #else
3400 	struct file *fp;
3401 	int error;
3402 	seqc_t seq;
3403 
3404 	*fpp = NULL;
3405 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3406 	for (;;) {
3407 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3408 		if (error != 0)
3409 			return (error);
3410 		error = cap_fcntl_check(fdp, fd, needfcntl);
3411 		if (!fd_modified(fdp, fd, seq))
3412 			break;
3413 		fdrop(fp, td);
3414 	}
3415 	if (error != 0) {
3416 		fdrop(fp, td);
3417 		return (error);
3418 	}
3419 	*fpp = fp;
3420 	return (0);
3421 #endif
3422 }
3423 
3424 /*
3425  * Like fget() but loads the underlying vnode, or returns an error if the
3426  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3427  * never have VM objects.  The returned vnode will be vref()'d.
3428  *
3429  * XXX: what about the unused flags ?
3430  */
3431 static __inline int
3432 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3433     struct vnode **vpp)
3434 {
3435 	struct file *fp;
3436 	int error;
3437 
3438 	*vpp = NULL;
3439 	error = _fget(td, fd, &fp, flags, needrightsp);
3440 	if (error != 0)
3441 		return (error);
3442 	if (fp->f_vnode == NULL) {
3443 		error = EINVAL;
3444 	} else {
3445 		*vpp = fp->f_vnode;
3446 		vref(*vpp);
3447 	}
3448 	fdrop(fp, td);
3449 
3450 	return (error);
3451 }
3452 
3453 int
3454 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3455 {
3456 
3457 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3458 }
3459 
3460 int
3461 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3462     struct filecaps *havecaps, struct vnode **vpp)
3463 {
3464 	struct filecaps caps;
3465 	struct file *fp;
3466 	int error;
3467 
3468 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3469 	if (error != 0)
3470 		return (error);
3471 	if (fp->f_ops == &badfileops) {
3472 		error = EBADF;
3473 		goto out;
3474 	}
3475 	if (fp->f_vnode == NULL) {
3476 		error = EINVAL;
3477 		goto out;
3478 	}
3479 
3480 	*havecaps = caps;
3481 	*vpp = fp->f_vnode;
3482 	vref(*vpp);
3483 	fdrop(fp, td);
3484 
3485 	return (0);
3486 out:
3487 	filecaps_free(&caps);
3488 	fdrop(fp, td);
3489 	return (error);
3490 }
3491 
3492 int
3493 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3494 {
3495 
3496 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3497 }
3498 
3499 int
3500 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3501 {
3502 
3503 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3504 }
3505 
3506 #ifdef notyet
3507 int
3508 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3509     struct vnode **vpp)
3510 {
3511 
3512 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3513 }
3514 #endif
3515 
3516 /*
3517  * Handle the last reference to a file being closed.
3518  *
3519  * Without the noinline attribute clang keeps inlining the func thorough this
3520  * file when fdrop is used.
3521  */
3522 int __noinline
3523 _fdrop(struct file *fp, struct thread *td)
3524 {
3525 	int error;
3526 #ifdef INVARIANTS
3527 	int count;
3528 
3529 	count = refcount_load(&fp->f_count);
3530 	if (count != 0)
3531 		panic("fdrop: fp %p count %d", fp, count);
3532 #endif
3533 	error = fo_close(fp, td);
3534 	atomic_subtract_int(&openfiles, 1);
3535 	crfree(fp->f_cred);
3536 	free(fp->f_advice, M_FADVISE);
3537 	uma_zfree(file_zone, fp);
3538 
3539 	return (error);
3540 }
3541 
3542 /*
3543  * Apply an advisory lock on a file descriptor.
3544  *
3545  * Just attempt to get a record lock of the requested type on the entire file
3546  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3547  */
3548 #ifndef _SYS_SYSPROTO_H_
3549 struct flock_args {
3550 	int	fd;
3551 	int	how;
3552 };
3553 #endif
3554 /* ARGSUSED */
3555 int
3556 sys_flock(struct thread *td, struct flock_args *uap)
3557 {
3558 	struct file *fp;
3559 	struct vnode *vp;
3560 	struct flock lf;
3561 	int error;
3562 
3563 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3564 	if (error != 0)
3565 		return (error);
3566 	if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
3567 		fdrop(fp, td);
3568 		return (EOPNOTSUPP);
3569 	}
3570 
3571 	vp = fp->f_vnode;
3572 	lf.l_whence = SEEK_SET;
3573 	lf.l_start = 0;
3574 	lf.l_len = 0;
3575 	if (uap->how & LOCK_UN) {
3576 		lf.l_type = F_UNLCK;
3577 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3578 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3579 		goto done2;
3580 	}
3581 	if (uap->how & LOCK_EX)
3582 		lf.l_type = F_WRLCK;
3583 	else if (uap->how & LOCK_SH)
3584 		lf.l_type = F_RDLCK;
3585 	else {
3586 		error = EBADF;
3587 		goto done2;
3588 	}
3589 	atomic_set_int(&fp->f_flag, FHASLOCK);
3590 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3591 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3592 done2:
3593 	fdrop(fp, td);
3594 	return (error);
3595 }
3596 /*
3597  * Duplicate the specified descriptor to a free descriptor.
3598  */
3599 int
3600 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3601     int openerror, int *indxp)
3602 {
3603 	struct filedescent *newfde, *oldfde;
3604 	struct file *fp;
3605 	u_long *ioctls;
3606 	int error, indx;
3607 
3608 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3609 	    ("unexpected error %d in %s", openerror, __func__));
3610 
3611 	/*
3612 	 * If the to-be-dup'd fd number is greater than the allowed number
3613 	 * of file descriptors, or the fd to be dup'd has already been
3614 	 * closed, then reject.
3615 	 */
3616 	FILEDESC_XLOCK(fdp);
3617 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3618 		FILEDESC_XUNLOCK(fdp);
3619 		return (EBADF);
3620 	}
3621 
3622 	error = fdalloc(td, 0, &indx);
3623 	if (error != 0) {
3624 		FILEDESC_XUNLOCK(fdp);
3625 		return (error);
3626 	}
3627 
3628 	/*
3629 	 * There are two cases of interest here.
3630 	 *
3631 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3632 	 *
3633 	 * For ENXIO steal away the file structure from (dfd) and store it in
3634 	 * (indx).  (dfd) is effectively closed by this operation.
3635 	 */
3636 	switch (openerror) {
3637 	case ENODEV:
3638 		/*
3639 		 * Check that the mode the file is being opened for is a
3640 		 * subset of the mode of the existing descriptor.
3641 		 */
3642 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3643 			fdunused(fdp, indx);
3644 			FILEDESC_XUNLOCK(fdp);
3645 			return (EACCES);
3646 		}
3647 		if (!fhold(fp)) {
3648 			fdunused(fdp, indx);
3649 			FILEDESC_XUNLOCK(fdp);
3650 			return (EBADF);
3651 		}
3652 		newfde = &fdp->fd_ofiles[indx];
3653 		oldfde = &fdp->fd_ofiles[dfd];
3654 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3655 #ifdef CAPABILITIES
3656 		seqc_write_begin(&newfde->fde_seqc);
3657 #endif
3658 		memcpy(newfde, oldfde, fde_change_size);
3659 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3660 		    ioctls);
3661 #ifdef CAPABILITIES
3662 		seqc_write_end(&newfde->fde_seqc);
3663 #endif
3664 		break;
3665 	case ENXIO:
3666 		/*
3667 		 * Steal away the file pointer from dfd and stuff it into indx.
3668 		 */
3669 		newfde = &fdp->fd_ofiles[indx];
3670 		oldfde = &fdp->fd_ofiles[dfd];
3671 #ifdef CAPABILITIES
3672 		seqc_write_begin(&newfde->fde_seqc);
3673 #endif
3674 		memcpy(newfde, oldfde, fde_change_size);
3675 		oldfde->fde_file = NULL;
3676 		fdunused(fdp, dfd);
3677 #ifdef CAPABILITIES
3678 		seqc_write_end(&newfde->fde_seqc);
3679 #endif
3680 		break;
3681 	}
3682 	FILEDESC_XUNLOCK(fdp);
3683 	*indxp = indx;
3684 	return (0);
3685 }
3686 
3687 /*
3688  * This sysctl determines if we will allow a process to chroot(2) if it
3689  * has a directory open:
3690  *	0: disallowed for all processes.
3691  *	1: allowed for processes that were not already chroot(2)'ed.
3692  *	2: allowed for all processes.
3693  */
3694 
3695 static int chroot_allow_open_directories = 1;
3696 
3697 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3698     &chroot_allow_open_directories, 0,
3699     "Allow a process to chroot(2) if it has a directory open");
3700 
3701 /*
3702  * Helper function for raised chroot(2) security function:  Refuse if
3703  * any filedescriptors are open directories.
3704  */
3705 static int
3706 chroot_refuse_vdir_fds(struct filedesc *fdp)
3707 {
3708 	struct vnode *vp;
3709 	struct file *fp;
3710 	int fd, lastfile;
3711 
3712 	FILEDESC_LOCK_ASSERT(fdp);
3713 
3714 	lastfile = fdlastfile(fdp);
3715 	for (fd = 0; fd <= lastfile; fd++) {
3716 		fp = fget_locked(fdp, fd);
3717 		if (fp == NULL)
3718 			continue;
3719 		if (fp->f_type == DTYPE_VNODE) {
3720 			vp = fp->f_vnode;
3721 			if (vp->v_type == VDIR)
3722 				return (EPERM);
3723 		}
3724 	}
3725 	return (0);
3726 }
3727 
3728 static void
3729 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3730 {
3731 
3732 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3733 		vrefact(oldpwd->pwd_cdir);
3734 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3735 	}
3736 
3737 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3738 		vrefact(oldpwd->pwd_rdir);
3739 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3740 	}
3741 
3742 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3743 		vrefact(oldpwd->pwd_jdir);
3744 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3745 	}
3746 }
3747 
3748 struct pwd *
3749 pwd_hold_pwddesc(struct pwddesc *pdp)
3750 {
3751 	struct pwd *pwd;
3752 
3753 	PWDDESC_ASSERT_XLOCKED(pdp);
3754 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3755 	if (pwd != NULL)
3756 		refcount_acquire(&pwd->pwd_refcount);
3757 	return (pwd);
3758 }
3759 
3760 bool
3761 pwd_hold_smr(struct pwd *pwd)
3762 {
3763 
3764 	MPASS(pwd != NULL);
3765 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3766 		return (true);
3767 	}
3768 	return (false);
3769 }
3770 
3771 struct pwd *
3772 pwd_hold(struct thread *td)
3773 {
3774 	struct pwddesc *pdp;
3775 	struct pwd *pwd;
3776 
3777 	pdp = td->td_proc->p_pd;
3778 
3779 	vfs_smr_enter();
3780 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3781 	if (pwd_hold_smr(pwd)) {
3782 		vfs_smr_exit();
3783 		return (pwd);
3784 	}
3785 	vfs_smr_exit();
3786 	PWDDESC_XLOCK(pdp);
3787 	pwd = pwd_hold_pwddesc(pdp);
3788 	MPASS(pwd != NULL);
3789 	PWDDESC_XUNLOCK(pdp);
3790 	return (pwd);
3791 }
3792 
3793 static struct pwd *
3794 pwd_alloc(void)
3795 {
3796 	struct pwd *pwd;
3797 
3798 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3799 	bzero(pwd, sizeof(*pwd));
3800 	refcount_init(&pwd->pwd_refcount, 1);
3801 	return (pwd);
3802 }
3803 
3804 void
3805 pwd_drop(struct pwd *pwd)
3806 {
3807 
3808 	if (!refcount_release(&pwd->pwd_refcount))
3809 		return;
3810 
3811 	if (pwd->pwd_cdir != NULL)
3812 		vrele(pwd->pwd_cdir);
3813 	if (pwd->pwd_rdir != NULL)
3814 		vrele(pwd->pwd_rdir);
3815 	if (pwd->pwd_jdir != NULL)
3816 		vrele(pwd->pwd_jdir);
3817 	uma_zfree_smr(pwd_zone, pwd);
3818 }
3819 
3820 /*
3821 * The caller is responsible for invoking priv_check() and
3822 * mac_vnode_check_chroot() to authorize this operation.
3823 */
3824 int
3825 pwd_chroot(struct thread *td, struct vnode *vp)
3826 {
3827 	struct pwddesc *pdp;
3828 	struct filedesc *fdp;
3829 	struct pwd *newpwd, *oldpwd;
3830 	int error;
3831 
3832 	fdp = td->td_proc->p_fd;
3833 	pdp = td->td_proc->p_pd;
3834 	newpwd = pwd_alloc();
3835 	FILEDESC_SLOCK(fdp);
3836 	PWDDESC_XLOCK(pdp);
3837 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3838 	if (chroot_allow_open_directories == 0 ||
3839 	    (chroot_allow_open_directories == 1 &&
3840 	    oldpwd->pwd_rdir != rootvnode)) {
3841 		error = chroot_refuse_vdir_fds(fdp);
3842 		FILEDESC_SUNLOCK(fdp);
3843 		if (error != 0) {
3844 			PWDDESC_XUNLOCK(pdp);
3845 			pwd_drop(newpwd);
3846 			return (error);
3847 		}
3848 	} else {
3849 		FILEDESC_SUNLOCK(fdp);
3850 	}
3851 
3852 	vrefact(vp);
3853 	newpwd->pwd_rdir = vp;
3854 	if (oldpwd->pwd_jdir == NULL) {
3855 		vrefact(vp);
3856 		newpwd->pwd_jdir = vp;
3857 	}
3858 	pwd_fill(oldpwd, newpwd);
3859 	pwd_set(pdp, newpwd);
3860 	PWDDESC_XUNLOCK(pdp);
3861 	pwd_drop(oldpwd);
3862 	return (0);
3863 }
3864 
3865 void
3866 pwd_chdir(struct thread *td, struct vnode *vp)
3867 {
3868 	struct pwddesc *pdp;
3869 	struct pwd *newpwd, *oldpwd;
3870 
3871 	VNPASS(vp->v_usecount > 0, vp);
3872 
3873 	newpwd = pwd_alloc();
3874 	pdp = td->td_proc->p_pd;
3875 	PWDDESC_XLOCK(pdp);
3876 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3877 	newpwd->pwd_cdir = vp;
3878 	pwd_fill(oldpwd, newpwd);
3879 	pwd_set(pdp, newpwd);
3880 	PWDDESC_XUNLOCK(pdp);
3881 	pwd_drop(oldpwd);
3882 }
3883 
3884 /*
3885  * jail_attach(2) changes both root and working directories.
3886  */
3887 int
3888 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3889 {
3890 	struct pwddesc *pdp;
3891 	struct filedesc *fdp;
3892 	struct pwd *newpwd, *oldpwd;
3893 	int error;
3894 
3895 	fdp = td->td_proc->p_fd;
3896 	pdp = td->td_proc->p_pd;
3897 	newpwd = pwd_alloc();
3898 	FILEDESC_SLOCK(fdp);
3899 	PWDDESC_XLOCK(pdp);
3900 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3901 	error = chroot_refuse_vdir_fds(fdp);
3902 	FILEDESC_SUNLOCK(fdp);
3903 	if (error != 0) {
3904 		PWDDESC_XUNLOCK(pdp);
3905 		pwd_drop(newpwd);
3906 		return (error);
3907 	}
3908 
3909 	vrefact(vp);
3910 	newpwd->pwd_rdir = vp;
3911 	vrefact(vp);
3912 	newpwd->pwd_cdir = vp;
3913 	if (oldpwd->pwd_jdir == NULL) {
3914 		vrefact(vp);
3915 		newpwd->pwd_jdir = vp;
3916 	}
3917 	pwd_fill(oldpwd, newpwd);
3918 	pwd_set(pdp, newpwd);
3919 	PWDDESC_XUNLOCK(pdp);
3920 	pwd_drop(oldpwd);
3921 	return (0);
3922 }
3923 
3924 void
3925 pwd_ensure_dirs(void)
3926 {
3927 	struct pwddesc *pdp;
3928 	struct pwd *oldpwd, *newpwd;
3929 
3930 	pdp = curproc->p_pd;
3931 	PWDDESC_XLOCK(pdp);
3932 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3933 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3934 		PWDDESC_XUNLOCK(pdp);
3935 		return;
3936 	}
3937 	PWDDESC_XUNLOCK(pdp);
3938 
3939 	newpwd = pwd_alloc();
3940 	PWDDESC_XLOCK(pdp);
3941 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3942 	pwd_fill(oldpwd, newpwd);
3943 	if (newpwd->pwd_cdir == NULL) {
3944 		vrefact(rootvnode);
3945 		newpwd->pwd_cdir = rootvnode;
3946 	}
3947 	if (newpwd->pwd_rdir == NULL) {
3948 		vrefact(rootvnode);
3949 		newpwd->pwd_rdir = rootvnode;
3950 	}
3951 	pwd_set(pdp, newpwd);
3952 	PWDDESC_XUNLOCK(pdp);
3953 	pwd_drop(oldpwd);
3954 }
3955 
3956 void
3957 pwd_set_rootvnode(void)
3958 {
3959 	struct pwddesc *pdp;
3960 	struct pwd *oldpwd, *newpwd;
3961 
3962 	pdp = curproc->p_pd;
3963 
3964 	newpwd = pwd_alloc();
3965 	PWDDESC_XLOCK(pdp);
3966 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3967 	vrefact(rootvnode);
3968 	newpwd->pwd_cdir = rootvnode;
3969 	vrefact(rootvnode);
3970 	newpwd->pwd_rdir = rootvnode;
3971 	pwd_fill(oldpwd, newpwd);
3972 	pwd_set(pdp, newpwd);
3973 	PWDDESC_XUNLOCK(pdp);
3974 	pwd_drop(oldpwd);
3975 }
3976 
3977 /*
3978  * Scan all active processes and prisons to see if any of them have a current
3979  * or root directory of `olddp'. If so, replace them with the new mount point.
3980  */
3981 void
3982 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3983 {
3984 	struct pwddesc *pdp;
3985 	struct pwd *newpwd, *oldpwd;
3986 	struct prison *pr;
3987 	struct proc *p;
3988 	int nrele;
3989 
3990 	if (vrefcnt(olddp) == 1)
3991 		return;
3992 	nrele = 0;
3993 	newpwd = pwd_alloc();
3994 	sx_slock(&allproc_lock);
3995 	FOREACH_PROC_IN_SYSTEM(p) {
3996 		PROC_LOCK(p);
3997 		pdp = pdhold(p);
3998 		PROC_UNLOCK(p);
3999 		if (pdp == NULL)
4000 			continue;
4001 		PWDDESC_XLOCK(pdp);
4002 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4003 		if (oldpwd == NULL ||
4004 		    (oldpwd->pwd_cdir != olddp &&
4005 		    oldpwd->pwd_rdir != olddp &&
4006 		    oldpwd->pwd_jdir != olddp)) {
4007 			PWDDESC_XUNLOCK(pdp);
4008 			pddrop(pdp);
4009 			continue;
4010 		}
4011 		if (oldpwd->pwd_cdir == olddp) {
4012 			vrefact(newdp);
4013 			newpwd->pwd_cdir = newdp;
4014 		}
4015 		if (oldpwd->pwd_rdir == olddp) {
4016 			vrefact(newdp);
4017 			newpwd->pwd_rdir = newdp;
4018 		}
4019 		if (oldpwd->pwd_jdir == olddp) {
4020 			vrefact(newdp);
4021 			newpwd->pwd_jdir = newdp;
4022 		}
4023 		pwd_fill(oldpwd, newpwd);
4024 		pwd_set(pdp, newpwd);
4025 		PWDDESC_XUNLOCK(pdp);
4026 		pwd_drop(oldpwd);
4027 		pddrop(pdp);
4028 		newpwd = pwd_alloc();
4029 	}
4030 	sx_sunlock(&allproc_lock);
4031 	pwd_drop(newpwd);
4032 	if (rootvnode == olddp) {
4033 		vrefact(newdp);
4034 		rootvnode = newdp;
4035 		nrele++;
4036 	}
4037 	mtx_lock(&prison0.pr_mtx);
4038 	if (prison0.pr_root == olddp) {
4039 		vrefact(newdp);
4040 		prison0.pr_root = newdp;
4041 		nrele++;
4042 	}
4043 	mtx_unlock(&prison0.pr_mtx);
4044 	sx_slock(&allprison_lock);
4045 	TAILQ_FOREACH(pr, &allprison, pr_list) {
4046 		mtx_lock(&pr->pr_mtx);
4047 		if (pr->pr_root == olddp) {
4048 			vrefact(newdp);
4049 			pr->pr_root = newdp;
4050 			nrele++;
4051 		}
4052 		mtx_unlock(&pr->pr_mtx);
4053 	}
4054 	sx_sunlock(&allprison_lock);
4055 	while (nrele--)
4056 		vrele(olddp);
4057 }
4058 
4059 struct filedesc_to_leader *
4060 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4061 {
4062 	struct filedesc_to_leader *fdtol;
4063 
4064 	fdtol = malloc(sizeof(struct filedesc_to_leader),
4065 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4066 	fdtol->fdl_refcount = 1;
4067 	fdtol->fdl_holdcount = 0;
4068 	fdtol->fdl_wakeup = 0;
4069 	fdtol->fdl_leader = leader;
4070 	if (old != NULL) {
4071 		FILEDESC_XLOCK(fdp);
4072 		fdtol->fdl_next = old->fdl_next;
4073 		fdtol->fdl_prev = old;
4074 		old->fdl_next = fdtol;
4075 		fdtol->fdl_next->fdl_prev = fdtol;
4076 		FILEDESC_XUNLOCK(fdp);
4077 	} else {
4078 		fdtol->fdl_next = fdtol;
4079 		fdtol->fdl_prev = fdtol;
4080 	}
4081 	return (fdtol);
4082 }
4083 
4084 static int
4085 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4086 {
4087 	NDSLOTTYPE *map;
4088 	struct filedesc *fdp;
4089 	u_int namelen;
4090 	int count, off, minoff;
4091 
4092 	namelen = arg2;
4093 	if (namelen != 1)
4094 		return (EINVAL);
4095 
4096 	if (*(int *)arg1 != 0)
4097 		return (EINVAL);
4098 
4099 	fdp = curproc->p_fd;
4100 	count = 0;
4101 	FILEDESC_SLOCK(fdp);
4102 	map = fdp->fd_map;
4103 	off = NDSLOT(fdp->fd_nfiles - 1);
4104 	for (minoff = NDSLOT(0); off >= minoff; --off)
4105 		count += bitcountl(map[off]);
4106 	FILEDESC_SUNLOCK(fdp);
4107 
4108 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4109 }
4110 
4111 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4112     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4113     "Number of open file descriptors");
4114 
4115 /*
4116  * Get file structures globally.
4117  */
4118 static int
4119 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4120 {
4121 	struct xfile xf;
4122 	struct filedesc *fdp;
4123 	struct file *fp;
4124 	struct proc *p;
4125 	int error, n, lastfile;
4126 
4127 	error = sysctl_wire_old_buffer(req, 0);
4128 	if (error != 0)
4129 		return (error);
4130 	if (req->oldptr == NULL) {
4131 		n = 0;
4132 		sx_slock(&allproc_lock);
4133 		FOREACH_PROC_IN_SYSTEM(p) {
4134 			PROC_LOCK(p);
4135 			if (p->p_state == PRS_NEW) {
4136 				PROC_UNLOCK(p);
4137 				continue;
4138 			}
4139 			fdp = fdhold(p);
4140 			PROC_UNLOCK(p);
4141 			if (fdp == NULL)
4142 				continue;
4143 			/* overestimates sparse tables. */
4144 			n += fdp->fd_nfiles;
4145 			fddrop(fdp);
4146 		}
4147 		sx_sunlock(&allproc_lock);
4148 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4149 	}
4150 	error = 0;
4151 	bzero(&xf, sizeof(xf));
4152 	xf.xf_size = sizeof(xf);
4153 	sx_slock(&allproc_lock);
4154 	FOREACH_PROC_IN_SYSTEM(p) {
4155 		PROC_LOCK(p);
4156 		if (p->p_state == PRS_NEW) {
4157 			PROC_UNLOCK(p);
4158 			continue;
4159 		}
4160 		if (p_cansee(req->td, p) != 0) {
4161 			PROC_UNLOCK(p);
4162 			continue;
4163 		}
4164 		xf.xf_pid = p->p_pid;
4165 		xf.xf_uid = p->p_ucred->cr_uid;
4166 		fdp = fdhold(p);
4167 		PROC_UNLOCK(p);
4168 		if (fdp == NULL)
4169 			continue;
4170 		FILEDESC_SLOCK(fdp);
4171 		lastfile = fdlastfile(fdp);
4172 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4173 		    n++) {
4174 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4175 				continue;
4176 			xf.xf_fd = n;
4177 			xf.xf_file = (uintptr_t)fp;
4178 			xf.xf_data = (uintptr_t)fp->f_data;
4179 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4180 			xf.xf_type = (uintptr_t)fp->f_type;
4181 			xf.xf_count = refcount_load(&fp->f_count);
4182 			xf.xf_msgcount = 0;
4183 			xf.xf_offset = foffset_get(fp);
4184 			xf.xf_flag = fp->f_flag;
4185 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4186 			if (error)
4187 				break;
4188 		}
4189 		FILEDESC_SUNLOCK(fdp);
4190 		fddrop(fdp);
4191 		if (error)
4192 			break;
4193 	}
4194 	sx_sunlock(&allproc_lock);
4195 	return (error);
4196 }
4197 
4198 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4199     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4200 
4201 #ifdef KINFO_FILE_SIZE
4202 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4203 #endif
4204 
4205 static int
4206 xlate_fflags(int fflags)
4207 {
4208 	static const struct {
4209 		int	fflag;
4210 		int	kf_fflag;
4211 	} fflags_table[] = {
4212 		{ FAPPEND, KF_FLAG_APPEND },
4213 		{ FASYNC, KF_FLAG_ASYNC },
4214 		{ FFSYNC, KF_FLAG_FSYNC },
4215 		{ FHASLOCK, KF_FLAG_HASLOCK },
4216 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4217 		{ FREAD, KF_FLAG_READ },
4218 		{ FWRITE, KF_FLAG_WRITE },
4219 		{ O_CREAT, KF_FLAG_CREAT },
4220 		{ O_DIRECT, KF_FLAG_DIRECT },
4221 		{ O_EXCL, KF_FLAG_EXCL },
4222 		{ O_EXEC, KF_FLAG_EXEC },
4223 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4224 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4225 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4226 		{ O_TRUNC, KF_FLAG_TRUNC }
4227 	};
4228 	unsigned int i;
4229 	int kflags;
4230 
4231 	kflags = 0;
4232 	for (i = 0; i < nitems(fflags_table); i++)
4233 		if (fflags & fflags_table[i].fflag)
4234 			kflags |=  fflags_table[i].kf_fflag;
4235 	return (kflags);
4236 }
4237 
4238 /* Trim unused data from kf_path by truncating the structure size. */
4239 void
4240 pack_kinfo(struct kinfo_file *kif)
4241 {
4242 
4243 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4244 	    strlen(kif->kf_path) + 1;
4245 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4246 }
4247 
4248 static void
4249 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4250     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4251 {
4252 	int error;
4253 
4254 	bzero(kif, sizeof(*kif));
4255 
4256 	/* Set a default type to allow for empty fill_kinfo() methods. */
4257 	kif->kf_type = KF_TYPE_UNKNOWN;
4258 	kif->kf_flags = xlate_fflags(fp->f_flag);
4259 	if (rightsp != NULL)
4260 		kif->kf_cap_rights = *rightsp;
4261 	else
4262 		cap_rights_init_zero(&kif->kf_cap_rights);
4263 	kif->kf_fd = fd;
4264 	kif->kf_ref_count = refcount_load(&fp->f_count);
4265 	kif->kf_offset = foffset_get(fp);
4266 
4267 	/*
4268 	 * This may drop the filedesc lock, so the 'fp' cannot be
4269 	 * accessed after this call.
4270 	 */
4271 	error = fo_fill_kinfo(fp, kif, fdp);
4272 	if (error == 0)
4273 		kif->kf_status |= KF_ATTR_VALID;
4274 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4275 		pack_kinfo(kif);
4276 	else
4277 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4278 }
4279 
4280 static void
4281 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4282     struct kinfo_file *kif, int flags)
4283 {
4284 	int error;
4285 
4286 	bzero(kif, sizeof(*kif));
4287 
4288 	kif->kf_type = KF_TYPE_VNODE;
4289 	error = vn_fill_kinfo_vnode(vp, kif);
4290 	if (error == 0)
4291 		kif->kf_status |= KF_ATTR_VALID;
4292 	kif->kf_flags = xlate_fflags(fflags);
4293 	cap_rights_init_zero(&kif->kf_cap_rights);
4294 	kif->kf_fd = fd;
4295 	kif->kf_ref_count = -1;
4296 	kif->kf_offset = -1;
4297 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4298 		pack_kinfo(kif);
4299 	else
4300 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4301 	vrele(vp);
4302 }
4303 
4304 struct export_fd_buf {
4305 	struct filedesc		*fdp;
4306 	struct pwddesc	*pdp;
4307 	struct sbuf 		*sb;
4308 	ssize_t			remainder;
4309 	struct kinfo_file	kif;
4310 	int			flags;
4311 };
4312 
4313 static int
4314 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4315 {
4316 	struct kinfo_file *kif;
4317 
4318 	kif = &efbuf->kif;
4319 	if (efbuf->remainder != -1) {
4320 		if (efbuf->remainder < kif->kf_structsize) {
4321 			/* Terminate export. */
4322 			efbuf->remainder = 0;
4323 			return (0);
4324 		}
4325 		efbuf->remainder -= kif->kf_structsize;
4326 	}
4327 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4328 }
4329 
4330 static int
4331 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4332     struct export_fd_buf *efbuf)
4333 {
4334 	int error;
4335 
4336 	if (efbuf->remainder == 0)
4337 		return (0);
4338 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4339 	    efbuf->flags);
4340 	FILEDESC_SUNLOCK(efbuf->fdp);
4341 	error = export_kinfo_to_sb(efbuf);
4342 	FILEDESC_SLOCK(efbuf->fdp);
4343 	return (error);
4344 }
4345 
4346 static int
4347 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4348     struct export_fd_buf *efbuf)
4349 {
4350 	int error;
4351 
4352 	if (efbuf->remainder == 0)
4353 		return (0);
4354 	if (efbuf->pdp != NULL)
4355 		PWDDESC_XUNLOCK(efbuf->pdp);
4356 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4357 	error = export_kinfo_to_sb(efbuf);
4358 	if (efbuf->pdp != NULL)
4359 		PWDDESC_XLOCK(efbuf->pdp);
4360 	return (error);
4361 }
4362 
4363 /*
4364  * Store a process file descriptor information to sbuf.
4365  *
4366  * Takes a locked proc as argument, and returns with the proc unlocked.
4367  */
4368 int
4369 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4370     int flags)
4371 {
4372 	struct file *fp;
4373 	struct filedesc *fdp;
4374 	struct pwddesc *pdp;
4375 	struct export_fd_buf *efbuf;
4376 	struct vnode *cttyvp, *textvp, *tracevp;
4377 	struct pwd *pwd;
4378 	int error, i, lastfile;
4379 	cap_rights_t rights;
4380 
4381 	PROC_LOCK_ASSERT(p, MA_OWNED);
4382 
4383 	/* ktrace vnode */
4384 	tracevp = ktr_get_tracevp(p, true);
4385 	/* text vnode */
4386 	textvp = p->p_textvp;
4387 	if (textvp != NULL)
4388 		vrefact(textvp);
4389 	/* Controlling tty. */
4390 	cttyvp = NULL;
4391 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4392 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4393 		if (cttyvp != NULL)
4394 			vrefact(cttyvp);
4395 	}
4396 	fdp = fdhold(p);
4397 	pdp = pdhold(p);
4398 	PROC_UNLOCK(p);
4399 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4400 	efbuf->fdp = NULL;
4401 	efbuf->pdp = NULL;
4402 	efbuf->sb = sb;
4403 	efbuf->remainder = maxlen;
4404 	efbuf->flags = flags;
4405 	if (tracevp != NULL)
4406 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4407 		    efbuf);
4408 	if (textvp != NULL)
4409 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4410 	if (cttyvp != NULL)
4411 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4412 		    efbuf);
4413 	error = 0;
4414 	if (pdp == NULL || fdp == NULL)
4415 		goto fail;
4416 	efbuf->fdp = fdp;
4417 	efbuf->pdp = pdp;
4418 	PWDDESC_XLOCK(pdp);
4419 	pwd = pwd_hold_pwddesc(pdp);
4420 	if (pwd != NULL) {
4421 		/* working directory */
4422 		if (pwd->pwd_cdir != NULL) {
4423 			vrefact(pwd->pwd_cdir);
4424 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD,
4425 			    FREAD, efbuf);
4426 		}
4427 		/* root directory */
4428 		if (pwd->pwd_rdir != NULL) {
4429 			vrefact(pwd->pwd_rdir);
4430 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT,
4431 			    FREAD, efbuf);
4432 		}
4433 		/* jail directory */
4434 		if (pwd->pwd_jdir != NULL) {
4435 			vrefact(pwd->pwd_jdir);
4436 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL,
4437 			    FREAD, efbuf);
4438 		}
4439 	}
4440 	PWDDESC_XUNLOCK(pdp);
4441 	if (pwd != NULL)
4442 		pwd_drop(pwd);
4443 	FILEDESC_SLOCK(fdp);
4444 	lastfile = fdlastfile(fdp);
4445 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4446 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4447 			continue;
4448 #ifdef CAPABILITIES
4449 		rights = *cap_rights(fdp, i);
4450 #else /* !CAPABILITIES */
4451 		rights = cap_no_rights;
4452 #endif
4453 		/*
4454 		 * Create sysctl entry.  It is OK to drop the filedesc
4455 		 * lock inside of export_file_to_sb() as we will
4456 		 * re-validate and re-evaluate its properties when the
4457 		 * loop continues.
4458 		 */
4459 		error = export_file_to_sb(fp, i, &rights, efbuf);
4460 		if (error != 0 || efbuf->remainder == 0)
4461 			break;
4462 	}
4463 	FILEDESC_SUNLOCK(fdp);
4464 fail:
4465 	if (fdp != NULL)
4466 		fddrop(fdp);
4467 	if (pdp != NULL)
4468 		pddrop(pdp);
4469 	free(efbuf, M_TEMP);
4470 	return (error);
4471 }
4472 
4473 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4474 
4475 /*
4476  * Get per-process file descriptors for use by procstat(1), et al.
4477  */
4478 static int
4479 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4480 {
4481 	struct sbuf sb;
4482 	struct proc *p;
4483 	ssize_t maxlen;
4484 	u_int namelen;
4485 	int error, error2, *name;
4486 
4487 	namelen = arg2;
4488 	if (namelen != 1)
4489 		return (EINVAL);
4490 
4491 	name = (int *)arg1;
4492 
4493 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4494 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4495 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4496 	if (error != 0) {
4497 		sbuf_delete(&sb);
4498 		return (error);
4499 	}
4500 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4501 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4502 	    KERN_FILEDESC_PACK_KINFO);
4503 	error2 = sbuf_finish(&sb);
4504 	sbuf_delete(&sb);
4505 	return (error != 0 ? error : error2);
4506 }
4507 
4508 #ifdef COMPAT_FREEBSD7
4509 #ifdef KINFO_OFILE_SIZE
4510 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4511 #endif
4512 
4513 static void
4514 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4515 {
4516 
4517 	okif->kf_structsize = sizeof(*okif);
4518 	okif->kf_type = kif->kf_type;
4519 	okif->kf_fd = kif->kf_fd;
4520 	okif->kf_ref_count = kif->kf_ref_count;
4521 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4522 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4523 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4524 	okif->kf_offset = kif->kf_offset;
4525 	if (kif->kf_type == KF_TYPE_VNODE)
4526 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4527 	else
4528 		okif->kf_vnode_type = KF_VTYPE_VNON;
4529 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4530 	if (kif->kf_type == KF_TYPE_SOCKET) {
4531 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4532 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4533 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4534 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4535 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4536 	} else {
4537 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4538 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4539 	}
4540 }
4541 
4542 static int
4543 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4544     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4545 {
4546 	int error;
4547 
4548 	vrefact(vp);
4549 	PWDDESC_XUNLOCK(pdp);
4550 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4551 	kinfo_to_okinfo(kif, okif);
4552 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4553 	PWDDESC_XLOCK(pdp);
4554 	return (error);
4555 }
4556 
4557 /*
4558  * Get per-process file descriptors for use by procstat(1), et al.
4559  */
4560 static int
4561 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4562 {
4563 	struct kinfo_ofile *okif;
4564 	struct kinfo_file *kif;
4565 	struct filedesc *fdp;
4566 	struct pwddesc *pdp;
4567 	struct pwd *pwd;
4568 	u_int namelen;
4569 	int error, i, lastfile, *name;
4570 	struct file *fp;
4571 	struct proc *p;
4572 
4573 	namelen = arg2;
4574 	if (namelen != 1)
4575 		return (EINVAL);
4576 
4577 	name = (int *)arg1;
4578 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4579 	if (error != 0)
4580 		return (error);
4581 	fdp = fdhold(p);
4582 	if (fdp != NULL)
4583 		pdp = pdhold(p);
4584 	PROC_UNLOCK(p);
4585 	if (fdp == NULL || pdp == NULL) {
4586 		if (fdp != NULL)
4587 			fddrop(fdp);
4588 		return (ENOENT);
4589 	}
4590 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4591 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4592 	PWDDESC_XLOCK(pdp);
4593 	pwd = pwd_hold_pwddesc(pdp);
4594 	if (pwd != NULL) {
4595 		if (pwd->pwd_cdir != NULL)
4596 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4597 			    okif, pdp, req);
4598 		if (pwd->pwd_rdir != NULL)
4599 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4600 			    okif, pdp, req);
4601 		if (pwd->pwd_jdir != NULL)
4602 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4603 			    okif, pdp, req);
4604 	}
4605 	PWDDESC_XUNLOCK(pdp);
4606 	if (pwd != NULL)
4607 		pwd_drop(pwd);
4608 	FILEDESC_SLOCK(fdp);
4609 	lastfile = fdlastfile(fdp);
4610 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4611 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4612 			continue;
4613 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4614 		    KERN_FILEDESC_PACK_KINFO);
4615 		FILEDESC_SUNLOCK(fdp);
4616 		kinfo_to_okinfo(kif, okif);
4617 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4618 		FILEDESC_SLOCK(fdp);
4619 		if (error)
4620 			break;
4621 	}
4622 	FILEDESC_SUNLOCK(fdp);
4623 	fddrop(fdp);
4624 	pddrop(pdp);
4625 	free(kif, M_TEMP);
4626 	free(okif, M_TEMP);
4627 	return (0);
4628 }
4629 
4630 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4631     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4632     "Process ofiledesc entries");
4633 #endif	/* COMPAT_FREEBSD7 */
4634 
4635 int
4636 vntype_to_kinfo(int vtype)
4637 {
4638 	struct {
4639 		int	vtype;
4640 		int	kf_vtype;
4641 	} vtypes_table[] = {
4642 		{ VBAD, KF_VTYPE_VBAD },
4643 		{ VBLK, KF_VTYPE_VBLK },
4644 		{ VCHR, KF_VTYPE_VCHR },
4645 		{ VDIR, KF_VTYPE_VDIR },
4646 		{ VFIFO, KF_VTYPE_VFIFO },
4647 		{ VLNK, KF_VTYPE_VLNK },
4648 		{ VNON, KF_VTYPE_VNON },
4649 		{ VREG, KF_VTYPE_VREG },
4650 		{ VSOCK, KF_VTYPE_VSOCK }
4651 	};
4652 	unsigned int i;
4653 
4654 	/*
4655 	 * Perform vtype translation.
4656 	 */
4657 	for (i = 0; i < nitems(vtypes_table); i++)
4658 		if (vtypes_table[i].vtype == vtype)
4659 			return (vtypes_table[i].kf_vtype);
4660 
4661 	return (KF_VTYPE_UNKNOWN);
4662 }
4663 
4664 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4665     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4666     "Process filedesc entries");
4667 
4668 /*
4669  * Store a process current working directory information to sbuf.
4670  *
4671  * Takes a locked proc as argument, and returns with the proc unlocked.
4672  */
4673 int
4674 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4675 {
4676 	struct pwddesc *pdp;
4677 	struct pwd *pwd;
4678 	struct export_fd_buf *efbuf;
4679 	struct vnode *cdir;
4680 	int error;
4681 
4682 	PROC_LOCK_ASSERT(p, MA_OWNED);
4683 
4684 	pdp = pdhold(p);
4685 	PROC_UNLOCK(p);
4686 	if (pdp == NULL)
4687 		return (EINVAL);
4688 
4689 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4690 	efbuf->pdp = pdp;
4691 	efbuf->sb = sb;
4692 	efbuf->remainder = maxlen;
4693 
4694 	PWDDESC_XLOCK(pdp);
4695 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4696 	cdir = pwd->pwd_cdir;
4697 	if (cdir == NULL) {
4698 		error = EINVAL;
4699 	} else {
4700 		vrefact(cdir);
4701 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4702 	}
4703 	PWDDESC_XUNLOCK(pdp);
4704 	pddrop(pdp);
4705 	free(efbuf, M_TEMP);
4706 	return (error);
4707 }
4708 
4709 /*
4710  * Get per-process current working directory.
4711  */
4712 static int
4713 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4714 {
4715 	struct sbuf sb;
4716 	struct proc *p;
4717 	ssize_t maxlen;
4718 	u_int namelen;
4719 	int error, error2, *name;
4720 
4721 	namelen = arg2;
4722 	if (namelen != 1)
4723 		return (EINVAL);
4724 
4725 	name = (int *)arg1;
4726 
4727 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4728 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4729 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4730 	if (error != 0) {
4731 		sbuf_delete(&sb);
4732 		return (error);
4733 	}
4734 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4735 	error = kern_proc_cwd_out(p, &sb, maxlen);
4736 	error2 = sbuf_finish(&sb);
4737 	sbuf_delete(&sb);
4738 	return (error != 0 ? error : error2);
4739 }
4740 
4741 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4742     sysctl_kern_proc_cwd, "Process current working directory");
4743 
4744 #ifdef DDB
4745 /*
4746  * For the purposes of debugging, generate a human-readable string for the
4747  * file type.
4748  */
4749 static const char *
4750 file_type_to_name(short type)
4751 {
4752 
4753 	switch (type) {
4754 	case 0:
4755 		return ("zero");
4756 	case DTYPE_VNODE:
4757 		return ("vnode");
4758 	case DTYPE_SOCKET:
4759 		return ("socket");
4760 	case DTYPE_PIPE:
4761 		return ("pipe");
4762 	case DTYPE_FIFO:
4763 		return ("fifo");
4764 	case DTYPE_KQUEUE:
4765 		return ("kqueue");
4766 	case DTYPE_CRYPTO:
4767 		return ("crypto");
4768 	case DTYPE_MQUEUE:
4769 		return ("mqueue");
4770 	case DTYPE_SHM:
4771 		return ("shm");
4772 	case DTYPE_SEM:
4773 		return ("ksem");
4774 	case DTYPE_PTS:
4775 		return ("pts");
4776 	case DTYPE_DEV:
4777 		return ("dev");
4778 	case DTYPE_PROCDESC:
4779 		return ("proc");
4780 	case DTYPE_EVENTFD:
4781 		return ("eventfd");
4782 	case DTYPE_LINUXTFD:
4783 		return ("ltimer");
4784 	default:
4785 		return ("unkn");
4786 	}
4787 }
4788 
4789 /*
4790  * For the purposes of debugging, identify a process (if any, perhaps one of
4791  * many) that references the passed file in its file descriptor array. Return
4792  * NULL if none.
4793  */
4794 static struct proc *
4795 file_to_first_proc(struct file *fp)
4796 {
4797 	struct filedesc *fdp;
4798 	struct proc *p;
4799 	int n;
4800 
4801 	FOREACH_PROC_IN_SYSTEM(p) {
4802 		if (p->p_state == PRS_NEW)
4803 			continue;
4804 		fdp = p->p_fd;
4805 		if (fdp == NULL)
4806 			continue;
4807 		for (n = 0; n < fdp->fd_nfiles; n++) {
4808 			if (fp == fdp->fd_ofiles[n].fde_file)
4809 				return (p);
4810 		}
4811 	}
4812 	return (NULL);
4813 }
4814 
4815 static void
4816 db_print_file(struct file *fp, int header)
4817 {
4818 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4819 	struct proc *p;
4820 
4821 	if (header)
4822 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4823 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4824 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4825 		    "FCmd");
4826 	p = file_to_first_proc(fp);
4827 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4828 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4829 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4830 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4831 
4832 #undef XPTRWIDTH
4833 }
4834 
4835 DB_SHOW_COMMAND(file, db_show_file)
4836 {
4837 	struct file *fp;
4838 
4839 	if (!have_addr) {
4840 		db_printf("usage: show file <addr>\n");
4841 		return;
4842 	}
4843 	fp = (struct file *)addr;
4844 	db_print_file(fp, 1);
4845 }
4846 
4847 DB_SHOW_COMMAND(files, db_show_files)
4848 {
4849 	struct filedesc *fdp;
4850 	struct file *fp;
4851 	struct proc *p;
4852 	int header;
4853 	int n;
4854 
4855 	header = 1;
4856 	FOREACH_PROC_IN_SYSTEM(p) {
4857 		if (p->p_state == PRS_NEW)
4858 			continue;
4859 		if ((fdp = p->p_fd) == NULL)
4860 			continue;
4861 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4862 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4863 				continue;
4864 			db_print_file(fp, header);
4865 			header = 0;
4866 		}
4867 	}
4868 }
4869 #endif
4870 
4871 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4872     &maxfilesperproc, 0, "Maximum files allowed open per process");
4873 
4874 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4875     &maxfiles, 0, "Maximum number of files");
4876 
4877 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4878     &openfiles, 0, "System-wide number of open files");
4879 
4880 /* ARGSUSED*/
4881 static void
4882 filelistinit(void *dummy)
4883 {
4884 
4885 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4886 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4887 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4888 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4889 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4890 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4891 	/*
4892 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4893 	 * the vnode zone.
4894 	 */
4895 	vfs_smr = uma_zone_get_smr(pwd_zone);
4896 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4897 }
4898 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4899 
4900 /*-------------------------------------------------------------------*/
4901 
4902 static int
4903 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4904     int flags, struct thread *td)
4905 {
4906 
4907 	return (EBADF);
4908 }
4909 
4910 static int
4911 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4912     struct thread *td)
4913 {
4914 
4915 	return (EINVAL);
4916 }
4917 
4918 static int
4919 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4920     struct thread *td)
4921 {
4922 
4923 	return (EBADF);
4924 }
4925 
4926 static int
4927 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4928     struct thread *td)
4929 {
4930 
4931 	return (0);
4932 }
4933 
4934 static int
4935 badfo_kqfilter(struct file *fp, struct knote *kn)
4936 {
4937 
4938 	return (EBADF);
4939 }
4940 
4941 static int
4942 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4943     struct thread *td)
4944 {
4945 
4946 	return (EBADF);
4947 }
4948 
4949 static int
4950 badfo_close(struct file *fp, struct thread *td)
4951 {
4952 
4953 	return (0);
4954 }
4955 
4956 static int
4957 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4958     struct thread *td)
4959 {
4960 
4961 	return (EBADF);
4962 }
4963 
4964 static int
4965 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4966     struct thread *td)
4967 {
4968 
4969 	return (EBADF);
4970 }
4971 
4972 static int
4973 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4974     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4975     struct thread *td)
4976 {
4977 
4978 	return (EBADF);
4979 }
4980 
4981 static int
4982 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4983 {
4984 
4985 	return (0);
4986 }
4987 
4988 struct fileops badfileops = {
4989 	.fo_read = badfo_readwrite,
4990 	.fo_write = badfo_readwrite,
4991 	.fo_truncate = badfo_truncate,
4992 	.fo_ioctl = badfo_ioctl,
4993 	.fo_poll = badfo_poll,
4994 	.fo_kqfilter = badfo_kqfilter,
4995 	.fo_stat = badfo_stat,
4996 	.fo_close = badfo_close,
4997 	.fo_chmod = badfo_chmod,
4998 	.fo_chown = badfo_chown,
4999 	.fo_sendfile = badfo_sendfile,
5000 	.fo_fill_kinfo = badfo_fill_kinfo,
5001 };
5002 
5003 static int
5004 path_poll(struct file *fp, int events, struct ucred *active_cred,
5005     struct thread *td)
5006 {
5007 	return (POLLNVAL);
5008 }
5009 
5010 static int
5011 path_close(struct file *fp, struct thread *td)
5012 {
5013 	MPASS(fp->f_type == DTYPE_VNODE);
5014 	fp->f_ops = &badfileops;
5015 	vdrop(fp->f_vnode);
5016 	return (0);
5017 }
5018 
5019 struct fileops path_fileops = {
5020 	.fo_read = badfo_readwrite,
5021 	.fo_write = badfo_readwrite,
5022 	.fo_truncate = badfo_truncate,
5023 	.fo_ioctl = badfo_ioctl,
5024 	.fo_poll = path_poll,
5025 	.fo_kqfilter = vn_kqfilter_opath,
5026 	.fo_stat = vn_statfile,
5027 	.fo_close = path_close,
5028 	.fo_chmod = badfo_chmod,
5029 	.fo_chown = badfo_chown,
5030 	.fo_sendfile = badfo_sendfile,
5031 	.fo_fill_kinfo = vn_fill_kinfo,
5032 	.fo_flags = DFLAG_PASSABLE,
5033 };
5034 
5035 int
5036 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5037     int flags, struct thread *td)
5038 {
5039 
5040 	return (EOPNOTSUPP);
5041 }
5042 
5043 int
5044 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5045     struct thread *td)
5046 {
5047 
5048 	return (EINVAL);
5049 }
5050 
5051 int
5052 invfo_ioctl(struct file *fp, u_long com, void *data,
5053     struct ucred *active_cred, struct thread *td)
5054 {
5055 
5056 	return (ENOTTY);
5057 }
5058 
5059 int
5060 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5061     struct thread *td)
5062 {
5063 
5064 	return (poll_no_poll(events));
5065 }
5066 
5067 int
5068 invfo_kqfilter(struct file *fp, struct knote *kn)
5069 {
5070 
5071 	return (EINVAL);
5072 }
5073 
5074 int
5075 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5076     struct thread *td)
5077 {
5078 
5079 	return (EINVAL);
5080 }
5081 
5082 int
5083 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5084     struct thread *td)
5085 {
5086 
5087 	return (EINVAL);
5088 }
5089 
5090 int
5091 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5092     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5093     struct thread *td)
5094 {
5095 
5096 	return (EINVAL);
5097 }
5098 
5099 /*-------------------------------------------------------------------*/
5100 
5101 /*
5102  * File Descriptor pseudo-device driver (/dev/fd/).
5103  *
5104  * Opening minor device N dup()s the file (if any) connected to file
5105  * descriptor N belonging to the calling process.  Note that this driver
5106  * consists of only the ``open()'' routine, because all subsequent
5107  * references to this file will be direct to the other driver.
5108  *
5109  * XXX: we could give this one a cloning event handler if necessary.
5110  */
5111 
5112 /* ARGSUSED */
5113 static int
5114 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5115 {
5116 
5117 	/*
5118 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5119 	 * the file descriptor being sought for duplication. The error
5120 	 * return ensures that the vnode for this device will be released
5121 	 * by vn_open. Open will detect this special error and take the
5122 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5123 	 * will simply report the error.
5124 	 */
5125 	td->td_dupfd = dev2unit(dev);
5126 	return (ENODEV);
5127 }
5128 
5129 static struct cdevsw fildesc_cdevsw = {
5130 	.d_version =	D_VERSION,
5131 	.d_open =	fdopen,
5132 	.d_name =	"FD",
5133 };
5134 
5135 static void
5136 fildesc_drvinit(void *unused)
5137 {
5138 	struct cdev *dev;
5139 
5140 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5141 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5142 	make_dev_alias(dev, "stdin");
5143 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5144 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5145 	make_dev_alias(dev, "stdout");
5146 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5147 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5148 	make_dev_alias(dev, "stderr");
5149 }
5150 
5151 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5152