xref: /freebsd/sys/kern/kern_descrip.c (revision a3b4dcfd682d2f0dcad7e9e15dc00df2d7a283f2)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98     "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 
102 MALLOC_DECLARE(M_FADVISE);
103 
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
107 VFS_SMR_DECLARE;
108 
109 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
110 		    struct thread *td, int holdleaders);
111 static int	fd_first_free(struct filedesc *fdp, int low, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the last used fd.
217  *
218  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219  * Otherwise use fdlastfile.
220  */
221 int
222 fdlastfile_single(struct filedesc *fdp)
223 {
224 	NDSLOTTYPE *map = fdp->fd_map;
225 	int off, minoff;
226 
227 	off = NDSLOT(fdp->fd_nfiles - 1);
228 	for (minoff = NDSLOT(0); off >= minoff; --off)
229 		if (map[off] != 0)
230 			return (off * NDENTRIES + flsl(map[off]) - 1);
231 	return (-1);
232 }
233 
234 int
235 fdlastfile(struct filedesc *fdp)
236 {
237 
238 	FILEDESC_LOCK_ASSERT(fdp);
239 	return (fdlastfile_single(fdp));
240 }
241 
242 static int
243 fdisused(struct filedesc *fdp, int fd)
244 {
245 
246 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 
249 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
250 }
251 
252 /*
253  * Mark a file descriptor as used.
254  */
255 static void
256 fdused_init(struct filedesc *fdp, int fd)
257 {
258 
259 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 
261 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
262 }
263 
264 static void
265 fdused(struct filedesc *fdp, int fd)
266 {
267 
268 	FILEDESC_XLOCK_ASSERT(fdp);
269 
270 	fdused_init(fdp, fd);
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile++;
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	fde = &fdp->fd_ofiles[fd];
311 #ifdef CAPABILITIES
312 	seqc_write_begin(&fde->fde_seqc);
313 #endif
314 	fde->fde_file = NULL;
315 #ifdef CAPABILITIES
316 	seqc_write_end(&fde->fde_seqc);
317 #endif
318 	fdefree_last(fde);
319 	fdunused(fdp, fd);
320 }
321 
322 /*
323  * System calls on descriptors.
324  */
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
327 	int	dummy;
328 };
329 #endif
330 /* ARGSUSED */
331 int
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
333 {
334 #ifdef	RACCT
335 	uint64_t lim;
336 #endif
337 
338 	td->td_retval[0] = getmaxfd(td);
339 #ifdef	RACCT
340 	PROC_LOCK(td->td_proc);
341 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
342 	PROC_UNLOCK(td->td_proc);
343 	if (lim < td->td_retval[0])
344 		td->td_retval[0] = lim;
345 #endif
346 	return (0);
347 }
348 
349 /*
350  * Duplicate a file descriptor to a particular value.
351  *
352  * Note: keep in mind that a potential race condition exists when closing
353  * descriptors from a shared descriptor table (via rfork).
354  */
355 #ifndef _SYS_SYSPROTO_H_
356 struct dup2_args {
357 	u_int	from;
358 	u_int	to;
359 };
360 #endif
361 /* ARGSUSED */
362 int
363 sys_dup2(struct thread *td, struct dup2_args *uap)
364 {
365 
366 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
367 }
368 
369 /*
370  * Duplicate a file descriptor.
371  */
372 #ifndef _SYS_SYSPROTO_H_
373 struct dup_args {
374 	u_int	fd;
375 };
376 #endif
377 /* ARGSUSED */
378 int
379 sys_dup(struct thread *td, struct dup_args *uap)
380 {
381 
382 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
383 }
384 
385 /*
386  * The file control system call.
387  */
388 #ifndef _SYS_SYSPROTO_H_
389 struct fcntl_args {
390 	int	fd;
391 	int	cmd;
392 	long	arg;
393 };
394 #endif
395 /* ARGSUSED */
396 int
397 sys_fcntl(struct thread *td, struct fcntl_args *uap)
398 {
399 
400 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
401 }
402 
403 int
404 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
405 {
406 	struct flock fl;
407 	struct __oflock ofl;
408 	intptr_t arg1;
409 	int error, newcmd;
410 
411 	error = 0;
412 	newcmd = cmd;
413 	switch (cmd) {
414 	case F_OGETLK:
415 	case F_OSETLK:
416 	case F_OSETLKW:
417 		/*
418 		 * Convert old flock structure to new.
419 		 */
420 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
421 		fl.l_start = ofl.l_start;
422 		fl.l_len = ofl.l_len;
423 		fl.l_pid = ofl.l_pid;
424 		fl.l_type = ofl.l_type;
425 		fl.l_whence = ofl.l_whence;
426 		fl.l_sysid = 0;
427 
428 		switch (cmd) {
429 		case F_OGETLK:
430 			newcmd = F_GETLK;
431 			break;
432 		case F_OSETLK:
433 			newcmd = F_SETLK;
434 			break;
435 		case F_OSETLKW:
436 			newcmd = F_SETLKW;
437 			break;
438 		}
439 		arg1 = (intptr_t)&fl;
440 		break;
441 	case F_GETLK:
442 	case F_SETLK:
443 	case F_SETLKW:
444 	case F_SETLK_REMOTE:
445 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
446 		arg1 = (intptr_t)&fl;
447 		break;
448 	default:
449 		arg1 = arg;
450 		break;
451 	}
452 	if (error)
453 		return (error);
454 	error = kern_fcntl(td, fd, newcmd, arg1);
455 	if (error)
456 		return (error);
457 	if (cmd == F_OGETLK) {
458 		ofl.l_start = fl.l_start;
459 		ofl.l_len = fl.l_len;
460 		ofl.l_pid = fl.l_pid;
461 		ofl.l_type = fl.l_type;
462 		ofl.l_whence = fl.l_whence;
463 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
464 	} else if (cmd == F_GETLK) {
465 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
466 	}
467 	return (error);
468 }
469 
470 int
471 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
472 {
473 	struct filedesc *fdp;
474 	struct flock *flp;
475 	struct file *fp, *fp2;
476 	struct filedescent *fde;
477 	struct proc *p;
478 	struct vnode *vp;
479 	struct mount *mp;
480 	int error, flg, seals, tmp;
481 	uint64_t bsize;
482 	off_t foffset;
483 
484 	error = 0;
485 	flg = F_POSIX;
486 	p = td->td_proc;
487 	fdp = p->p_fd;
488 
489 	AUDIT_ARG_FD(cmd);
490 	AUDIT_ARG_CMD(cmd);
491 	switch (cmd) {
492 	case F_DUPFD:
493 		tmp = arg;
494 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
495 		break;
496 
497 	case F_DUPFD_CLOEXEC:
498 		tmp = arg;
499 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
500 		break;
501 
502 	case F_DUP2FD:
503 		tmp = arg;
504 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
505 		break;
506 
507 	case F_DUP2FD_CLOEXEC:
508 		tmp = arg;
509 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
510 		break;
511 
512 	case F_GETFD:
513 		error = EBADF;
514 		FILEDESC_SLOCK(fdp);
515 		fde = fdeget_locked(fdp, fd);
516 		if (fde != NULL) {
517 			td->td_retval[0] =
518 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
519 			error = 0;
520 		}
521 		FILEDESC_SUNLOCK(fdp);
522 		break;
523 
524 	case F_SETFD:
525 		error = EBADF;
526 		FILEDESC_XLOCK(fdp);
527 		fde = fdeget_locked(fdp, fd);
528 		if (fde != NULL) {
529 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
530 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
531 			error = 0;
532 		}
533 		FILEDESC_XUNLOCK(fdp);
534 		break;
535 
536 	case F_GETFL:
537 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
538 		if (error != 0)
539 			break;
540 		td->td_retval[0] = OFLAGS(fp->f_flag);
541 		fdrop(fp, td);
542 		break;
543 
544 	case F_SETFL:
545 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
546 		if (error != 0)
547 			break;
548 		do {
549 			tmp = flg = fp->f_flag;
550 			tmp &= ~FCNTLFLAGS;
551 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
552 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
553 		tmp = fp->f_flag & FNONBLOCK;
554 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
555 		if (error != 0) {
556 			fdrop(fp, td);
557 			break;
558 		}
559 		tmp = fp->f_flag & FASYNC;
560 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
561 		if (error == 0) {
562 			fdrop(fp, td);
563 			break;
564 		}
565 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
566 		tmp = 0;
567 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
568 		fdrop(fp, td);
569 		break;
570 
571 	case F_GETOWN:
572 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
573 		if (error != 0)
574 			break;
575 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
576 		if (error == 0)
577 			td->td_retval[0] = tmp;
578 		fdrop(fp, td);
579 		break;
580 
581 	case F_SETOWN:
582 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
583 		if (error != 0)
584 			break;
585 		tmp = arg;
586 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
587 		fdrop(fp, td);
588 		break;
589 
590 	case F_SETLK_REMOTE:
591 		error = priv_check(td, PRIV_NFS_LOCKD);
592 		if (error != 0)
593 			return (error);
594 		flg = F_REMOTE;
595 		goto do_setlk;
596 
597 	case F_SETLKW:
598 		flg |= F_WAIT;
599 		/* FALLTHROUGH F_SETLK */
600 
601 	case F_SETLK:
602 	do_setlk:
603 		flp = (struct flock *)arg;
604 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
605 			error = EINVAL;
606 			break;
607 		}
608 
609 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
610 		if (error != 0)
611 			break;
612 		if (fp->f_type != DTYPE_VNODE) {
613 			error = EBADF;
614 			fdrop(fp, td);
615 			break;
616 		}
617 
618 		if (flp->l_whence == SEEK_CUR) {
619 			foffset = foffset_get(fp);
620 			if (foffset < 0 ||
621 			    (flp->l_start > 0 &&
622 			     foffset > OFF_MAX - flp->l_start)) {
623 				error = EOVERFLOW;
624 				fdrop(fp, td);
625 				break;
626 			}
627 			flp->l_start += foffset;
628 		}
629 
630 		vp = fp->f_vnode;
631 		switch (flp->l_type) {
632 		case F_RDLCK:
633 			if ((fp->f_flag & FREAD) == 0) {
634 				error = EBADF;
635 				break;
636 			}
637 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
638 				PROC_LOCK(p->p_leader);
639 				p->p_leader->p_flag |= P_ADVLOCK;
640 				PROC_UNLOCK(p->p_leader);
641 			}
642 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
643 			    flp, flg);
644 			break;
645 		case F_WRLCK:
646 			if ((fp->f_flag & FWRITE) == 0) {
647 				error = EBADF;
648 				break;
649 			}
650 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 				PROC_LOCK(p->p_leader);
652 				p->p_leader->p_flag |= P_ADVLOCK;
653 				PROC_UNLOCK(p->p_leader);
654 			}
655 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
656 			    flp, flg);
657 			break;
658 		case F_UNLCK:
659 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
660 			    flp, flg);
661 			break;
662 		case F_UNLCKSYS:
663 			if (flg != F_REMOTE) {
664 				error = EINVAL;
665 				break;
666 			}
667 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
668 			    F_UNLCKSYS, flp, flg);
669 			break;
670 		default:
671 			error = EINVAL;
672 			break;
673 		}
674 		if (error != 0 || flp->l_type == F_UNLCK ||
675 		    flp->l_type == F_UNLCKSYS) {
676 			fdrop(fp, td);
677 			break;
678 		}
679 
680 		/*
681 		 * Check for a race with close.
682 		 *
683 		 * The vnode is now advisory locked (or unlocked, but this case
684 		 * is not really important) as the caller requested.
685 		 * We had to drop the filedesc lock, so we need to recheck if
686 		 * the descriptor is still valid, because if it was closed
687 		 * in the meantime we need to remove advisory lock from the
688 		 * vnode - close on any descriptor leading to an advisory
689 		 * locked vnode, removes that lock.
690 		 * We will return 0 on purpose in that case, as the result of
691 		 * successful advisory lock might have been externally visible
692 		 * already. This is fine - effectively we pretend to the caller
693 		 * that the closing thread was a bit slower and that the
694 		 * advisory lock succeeded before the close.
695 		 */
696 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
697 		if (error != 0) {
698 			fdrop(fp, td);
699 			break;
700 		}
701 		if (fp != fp2) {
702 			flp->l_whence = SEEK_SET;
703 			flp->l_start = 0;
704 			flp->l_len = 0;
705 			flp->l_type = F_UNLCK;
706 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
707 			    F_UNLCK, flp, F_POSIX);
708 		}
709 		fdrop(fp, td);
710 		fdrop(fp2, td);
711 		break;
712 
713 	case F_GETLK:
714 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
715 		if (error != 0)
716 			break;
717 		if (fp->f_type != DTYPE_VNODE) {
718 			error = EBADF;
719 			fdrop(fp, td);
720 			break;
721 		}
722 		flp = (struct flock *)arg;
723 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
724 		    flp->l_type != F_UNLCK) {
725 			error = EINVAL;
726 			fdrop(fp, td);
727 			break;
728 		}
729 		if (flp->l_whence == SEEK_CUR) {
730 			foffset = foffset_get(fp);
731 			if ((flp->l_start > 0 &&
732 			    foffset > OFF_MAX - flp->l_start) ||
733 			    (flp->l_start < 0 &&
734 			    foffset < OFF_MIN - flp->l_start)) {
735 				error = EOVERFLOW;
736 				fdrop(fp, td);
737 				break;
738 			}
739 			flp->l_start += foffset;
740 		}
741 		vp = fp->f_vnode;
742 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
743 		    F_POSIX);
744 		fdrop(fp, td);
745 		break;
746 
747 	case F_ADD_SEALS:
748 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
749 		if (error != 0)
750 			break;
751 		error = fo_add_seals(fp, arg);
752 		fdrop(fp, td);
753 		break;
754 
755 	case F_GET_SEALS:
756 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
757 		if (error != 0)
758 			break;
759 		if (fo_get_seals(fp, &seals) == 0)
760 			td->td_retval[0] = seals;
761 		else
762 			error = EINVAL;
763 		fdrop(fp, td);
764 		break;
765 
766 	case F_RDAHEAD:
767 		arg = arg ? 128 * 1024: 0;
768 		/* FALLTHROUGH */
769 	case F_READAHEAD:
770 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
771 		if (error != 0)
772 			break;
773 		if (fp->f_type != DTYPE_VNODE) {
774 			fdrop(fp, td);
775 			error = EBADF;
776 			break;
777 		}
778 		vp = fp->f_vnode;
779 		if (vp->v_type != VREG) {
780 			fdrop(fp, td);
781 			error = ENOTTY;
782 			break;
783 		}
784 
785 		/*
786 		 * Exclusive lock synchronizes against f_seqcount reads and
787 		 * writes in sequential_heuristic().
788 		 */
789 		error = vn_lock(vp, LK_EXCLUSIVE);
790 		if (error != 0) {
791 			fdrop(fp, td);
792 			break;
793 		}
794 		if (arg >= 0) {
795 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
796 			arg = MIN(arg, INT_MAX - bsize + 1);
797 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
798 			    (arg + bsize - 1) / bsize);
799 			atomic_set_int(&fp->f_flag, FRDAHEAD);
800 		} else {
801 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
802 		}
803 		VOP_UNLOCK(vp);
804 		fdrop(fp, td);
805 		break;
806 
807 	case F_ISUNIONSTACK:
808 		/*
809 		 * Check if the vnode is part of a union stack (either the
810 		 * "union" flag from mount(2) or unionfs).
811 		 *
812 		 * Prior to introduction of this op libc's readdir would call
813 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
814 		 * data just to check fs name and a mount flag.
815 		 *
816 		 * Fixing the code to handle everything in the kernel instead
817 		 * is a non-trivial endeavor and has low priority, thus this
818 		 * horrible kludge facilitates the current behavior in a much
819 		 * cheaper manner until someone(tm) sorts this out.
820 		 */
821 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
822 		if (error != 0)
823 			break;
824 		if (fp->f_type != DTYPE_VNODE) {
825 			fdrop(fp, td);
826 			error = EBADF;
827 			break;
828 		}
829 		vp = fp->f_vnode;
830 		/*
831 		 * Since we don't prevent dooming the vnode even non-null mp
832 		 * found can become immediately stale. This is tolerable since
833 		 * mount points are type-stable (providing safe memory access)
834 		 * and any vfs op on this vnode going forward will return an
835 		 * error (meaning return value in this case is meaningless).
836 		 */
837 		mp = atomic_load_ptr(&vp->v_mount);
838 		if (__predict_false(mp == NULL)) {
839 			fdrop(fp, td);
840 			error = EBADF;
841 			break;
842 		}
843 		td->td_retval[0] = 0;
844 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
845 		    mp->mnt_flag & MNT_UNION)
846 			td->td_retval[0] = 1;
847 		fdrop(fp, td);
848 		break;
849 
850 	default:
851 		error = EINVAL;
852 		break;
853 	}
854 	return (error);
855 }
856 
857 static int
858 getmaxfd(struct thread *td)
859 {
860 
861 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
862 }
863 
864 /*
865  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
866  */
867 int
868 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
869 {
870 	struct filedesc *fdp;
871 	struct filedescent *oldfde, *newfde;
872 	struct proc *p;
873 	struct file *delfp, *oldfp;
874 	u_long *oioctls, *nioctls;
875 	int error, maxfd;
876 
877 	p = td->td_proc;
878 	fdp = p->p_fd;
879 	oioctls = NULL;
880 
881 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
882 	MPASS(mode < FDDUP_LASTMODE);
883 
884 	AUDIT_ARG_FD(old);
885 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
886 
887 	/*
888 	 * Verify we have a valid descriptor to dup from and possibly to
889 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
890 	 * return EINVAL when the new descriptor is out of bounds.
891 	 */
892 	if (old < 0)
893 		return (EBADF);
894 	if (new < 0)
895 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
896 	maxfd = getmaxfd(td);
897 	if (new >= maxfd)
898 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
899 
900 	error = EBADF;
901 	FILEDESC_XLOCK(fdp);
902 	if (fget_locked(fdp, old) == NULL)
903 		goto unlock;
904 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
905 		td->td_retval[0] = new;
906 		if (flags & FDDUP_FLAG_CLOEXEC)
907 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
908 		error = 0;
909 		goto unlock;
910 	}
911 
912 	oldfde = &fdp->fd_ofiles[old];
913 	oldfp = oldfde->fde_file;
914 	if (!fhold(oldfp))
915 		goto unlock;
916 
917 	/*
918 	 * If the caller specified a file descriptor, make sure the file
919 	 * table is large enough to hold it, and grab it.  Otherwise, just
920 	 * allocate a new descriptor the usual way.
921 	 */
922 	switch (mode) {
923 	case FDDUP_NORMAL:
924 	case FDDUP_FCNTL:
925 		if ((error = fdalloc(td, new, &new)) != 0) {
926 			fdrop(oldfp, td);
927 			goto unlock;
928 		}
929 		break;
930 	case FDDUP_MUSTREPLACE:
931 		/* Target file descriptor must exist. */
932 		if (fget_locked(fdp, new) == NULL) {
933 			fdrop(oldfp, td);
934 			goto unlock;
935 		}
936 		break;
937 	case FDDUP_FIXED:
938 		if (new >= fdp->fd_nfiles) {
939 			/*
940 			 * The resource limits are here instead of e.g.
941 			 * fdalloc(), because the file descriptor table may be
942 			 * shared between processes, so we can't really use
943 			 * racct_add()/racct_sub().  Instead of counting the
944 			 * number of actually allocated descriptors, just put
945 			 * the limit on the size of the file descriptor table.
946 			 */
947 #ifdef RACCT
948 			if (RACCT_ENABLED()) {
949 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
950 				if (error != 0) {
951 					error = EMFILE;
952 					fdrop(oldfp, td);
953 					goto unlock;
954 				}
955 			}
956 #endif
957 			fdgrowtable_exp(fdp, new + 1);
958 		}
959 		if (!fdisused(fdp, new))
960 			fdused(fdp, new);
961 		break;
962 	default:
963 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
964 	}
965 
966 	KASSERT(old != new, ("new fd is same as old"));
967 
968 	/* Refetch oldfde because the table may have grown and old one freed. */
969 	oldfde = &fdp->fd_ofiles[old];
970 	KASSERT(oldfp == oldfde->fde_file,
971 	    ("fdt_ofiles shift from growth observed at fd %d",
972 	    old));
973 
974 	newfde = &fdp->fd_ofiles[new];
975 	delfp = newfde->fde_file;
976 
977 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
978 
979 	/*
980 	 * Duplicate the source descriptor.
981 	 */
982 #ifdef CAPABILITIES
983 	seqc_write_begin(&newfde->fde_seqc);
984 #endif
985 	oioctls = filecaps_free_prep(&newfde->fde_caps);
986 	memcpy(newfde, oldfde, fde_change_size);
987 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
988 	    nioctls);
989 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
990 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
991 	else
992 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
993 #ifdef CAPABILITIES
994 	seqc_write_end(&newfde->fde_seqc);
995 #endif
996 	td->td_retval[0] = new;
997 
998 	error = 0;
999 
1000 	if (delfp != NULL) {
1001 		(void) closefp(fdp, new, delfp, td, 1);
1002 		FILEDESC_UNLOCK_ASSERT(fdp);
1003 	} else {
1004 unlock:
1005 		FILEDESC_XUNLOCK(fdp);
1006 	}
1007 
1008 	filecaps_free_finish(oioctls);
1009 	return (error);
1010 }
1011 
1012 static void
1013 sigiofree(struct sigio *sigio)
1014 {
1015 	crfree(sigio->sio_ucred);
1016 	free(sigio, M_SIGIO);
1017 }
1018 
1019 static struct sigio *
1020 funsetown_locked(struct sigio *sigio)
1021 {
1022 	struct proc *p;
1023 	struct pgrp *pg;
1024 
1025 	SIGIO_ASSERT_LOCKED();
1026 
1027 	if (sigio == NULL)
1028 		return (NULL);
1029 	*(sigio->sio_myref) = NULL;
1030 	if (sigio->sio_pgid < 0) {
1031 		pg = sigio->sio_pgrp;
1032 		PGRP_LOCK(pg);
1033 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1034 		    sigio, sio_pgsigio);
1035 		PGRP_UNLOCK(pg);
1036 	} else {
1037 		p = sigio->sio_proc;
1038 		PROC_LOCK(p);
1039 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1040 		    sigio, sio_pgsigio);
1041 		PROC_UNLOCK(p);
1042 	}
1043 	return (sigio);
1044 }
1045 
1046 /*
1047  * If sigio is on the list associated with a process or process group,
1048  * disable signalling from the device, remove sigio from the list and
1049  * free sigio.
1050  */
1051 void
1052 funsetown(struct sigio **sigiop)
1053 {
1054 	struct sigio *sigio;
1055 
1056 	/* Racy check, consumers must provide synchronization. */
1057 	if (*sigiop == NULL)
1058 		return;
1059 
1060 	SIGIO_LOCK();
1061 	sigio = funsetown_locked(*sigiop);
1062 	SIGIO_UNLOCK();
1063 	if (sigio != NULL)
1064 		sigiofree(sigio);
1065 }
1066 
1067 /*
1068  * Free a list of sigio structures.  The caller must ensure that new sigio
1069  * structures cannot be added after this point.  For process groups this is
1070  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1071  * as an interlock.
1072  */
1073 void
1074 funsetownlst(struct sigiolst *sigiolst)
1075 {
1076 	struct proc *p;
1077 	struct pgrp *pg;
1078 	struct sigio *sigio, *tmp;
1079 
1080 	/* Racy check. */
1081 	sigio = SLIST_FIRST(sigiolst);
1082 	if (sigio == NULL)
1083 		return;
1084 
1085 	p = NULL;
1086 	pg = NULL;
1087 
1088 	SIGIO_LOCK();
1089 	sigio = SLIST_FIRST(sigiolst);
1090 	if (sigio == NULL) {
1091 		SIGIO_UNLOCK();
1092 		return;
1093 	}
1094 
1095 	/*
1096 	 * Every entry of the list should belong to a single proc or pgrp.
1097 	 */
1098 	if (sigio->sio_pgid < 0) {
1099 		pg = sigio->sio_pgrp;
1100 		sx_assert(&proctree_lock, SX_XLOCKED);
1101 		PGRP_LOCK(pg);
1102 	} else /* if (sigio->sio_pgid > 0) */ {
1103 		p = sigio->sio_proc;
1104 		PROC_LOCK(p);
1105 		KASSERT((p->p_flag & P_WEXIT) != 0,
1106 		    ("%s: process %p is not exiting", __func__, p));
1107 	}
1108 
1109 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1110 		*sigio->sio_myref = NULL;
1111 		if (pg != NULL) {
1112 			KASSERT(sigio->sio_pgid < 0,
1113 			    ("Proc sigio in pgrp sigio list"));
1114 			KASSERT(sigio->sio_pgrp == pg,
1115 			    ("Bogus pgrp in sigio list"));
1116 		} else /* if (p != NULL) */ {
1117 			KASSERT(sigio->sio_pgid > 0,
1118 			    ("Pgrp sigio in proc sigio list"));
1119 			KASSERT(sigio->sio_proc == p,
1120 			    ("Bogus proc in sigio list"));
1121 		}
1122 	}
1123 
1124 	if (pg != NULL)
1125 		PGRP_UNLOCK(pg);
1126 	else
1127 		PROC_UNLOCK(p);
1128 	SIGIO_UNLOCK();
1129 
1130 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1131 		sigiofree(sigio);
1132 }
1133 
1134 /*
1135  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1136  *
1137  * After permission checking, add a sigio structure to the sigio list for
1138  * the process or process group.
1139  */
1140 int
1141 fsetown(pid_t pgid, struct sigio **sigiop)
1142 {
1143 	struct proc *proc;
1144 	struct pgrp *pgrp;
1145 	struct sigio *osigio, *sigio;
1146 	int ret;
1147 
1148 	if (pgid == 0) {
1149 		funsetown(sigiop);
1150 		return (0);
1151 	}
1152 
1153 	ret = 0;
1154 
1155 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1156 	sigio->sio_pgid = pgid;
1157 	sigio->sio_ucred = crhold(curthread->td_ucred);
1158 	sigio->sio_myref = sigiop;
1159 
1160 	sx_slock(&proctree_lock);
1161 	SIGIO_LOCK();
1162 	osigio = funsetown_locked(*sigiop);
1163 	if (pgid > 0) {
1164 		proc = pfind(pgid);
1165 		if (proc == NULL) {
1166 			ret = ESRCH;
1167 			goto fail;
1168 		}
1169 
1170 		/*
1171 		 * Policy - Don't allow a process to FSETOWN a process
1172 		 * in another session.
1173 		 *
1174 		 * Remove this test to allow maximum flexibility or
1175 		 * restrict FSETOWN to the current process or process
1176 		 * group for maximum safety.
1177 		 */
1178 		if (proc->p_session != curthread->td_proc->p_session) {
1179 			PROC_UNLOCK(proc);
1180 			ret = EPERM;
1181 			goto fail;
1182 		}
1183 
1184 		sigio->sio_proc = proc;
1185 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1186 		PROC_UNLOCK(proc);
1187 	} else /* if (pgid < 0) */ {
1188 		pgrp = pgfind(-pgid);
1189 		if (pgrp == NULL) {
1190 			ret = ESRCH;
1191 			goto fail;
1192 		}
1193 
1194 		/*
1195 		 * Policy - Don't allow a process to FSETOWN a process
1196 		 * in another session.
1197 		 *
1198 		 * Remove this test to allow maximum flexibility or
1199 		 * restrict FSETOWN to the current process or process
1200 		 * group for maximum safety.
1201 		 */
1202 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1203 			PGRP_UNLOCK(pgrp);
1204 			ret = EPERM;
1205 			goto fail;
1206 		}
1207 
1208 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1209 		sigio->sio_pgrp = pgrp;
1210 		PGRP_UNLOCK(pgrp);
1211 	}
1212 	sx_sunlock(&proctree_lock);
1213 	*sigiop = sigio;
1214 	SIGIO_UNLOCK();
1215 	if (osigio != NULL)
1216 		sigiofree(osigio);
1217 	return (0);
1218 
1219 fail:
1220 	SIGIO_UNLOCK();
1221 	sx_sunlock(&proctree_lock);
1222 	sigiofree(sigio);
1223 	if (osigio != NULL)
1224 		sigiofree(osigio);
1225 	return (ret);
1226 }
1227 
1228 /*
1229  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1230  */
1231 pid_t
1232 fgetown(struct sigio **sigiop)
1233 {
1234 	pid_t pgid;
1235 
1236 	SIGIO_LOCK();
1237 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1238 	SIGIO_UNLOCK();
1239 	return (pgid);
1240 }
1241 
1242 /*
1243  * Function drops the filedesc lock on return.
1244  */
1245 static int
1246 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1247     int holdleaders)
1248 {
1249 	int error;
1250 
1251 	FILEDESC_XLOCK_ASSERT(fdp);
1252 
1253 	if (holdleaders) {
1254 		if (td->td_proc->p_fdtol != NULL) {
1255 			/*
1256 			 * Ask fdfree() to sleep to ensure that all relevant
1257 			 * process leaders can be traversed in closef().
1258 			 */
1259 			fdp->fd_holdleaderscount++;
1260 		} else {
1261 			holdleaders = 0;
1262 		}
1263 	}
1264 
1265 	/*
1266 	 * We now hold the fp reference that used to be owned by the
1267 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1268 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1269 	 * added, and deleteing a knote for the new fd.
1270 	 */
1271 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1272 		knote_fdclose(td, fd);
1273 
1274 	/*
1275 	 * We need to notify mqueue if the object is of type mqueue.
1276 	 */
1277 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1278 		mq_fdclose(td, fd, fp);
1279 	FILEDESC_XUNLOCK(fdp);
1280 
1281 	error = closef(fp, td);
1282 	if (holdleaders) {
1283 		FILEDESC_XLOCK(fdp);
1284 		fdp->fd_holdleaderscount--;
1285 		if (fdp->fd_holdleaderscount == 0 &&
1286 		    fdp->fd_holdleaderswakeup != 0) {
1287 			fdp->fd_holdleaderswakeup = 0;
1288 			wakeup(&fdp->fd_holdleaderscount);
1289 		}
1290 		FILEDESC_XUNLOCK(fdp);
1291 	}
1292 	return (error);
1293 }
1294 
1295 /*
1296  * Close a file descriptor.
1297  */
1298 #ifndef _SYS_SYSPROTO_H_
1299 struct close_args {
1300 	int     fd;
1301 };
1302 #endif
1303 /* ARGSUSED */
1304 int
1305 sys_close(struct thread *td, struct close_args *uap)
1306 {
1307 
1308 	return (kern_close(td, uap->fd));
1309 }
1310 
1311 int
1312 kern_close(struct thread *td, int fd)
1313 {
1314 	struct filedesc *fdp;
1315 	struct file *fp;
1316 
1317 	fdp = td->td_proc->p_fd;
1318 
1319 	AUDIT_SYSCLOSE(td, fd);
1320 
1321 	FILEDESC_XLOCK(fdp);
1322 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1323 		FILEDESC_XUNLOCK(fdp);
1324 		return (EBADF);
1325 	}
1326 	fdfree(fdp, fd);
1327 
1328 	/* closefp() drops the FILEDESC lock for us. */
1329 	return (closefp(fdp, fd, fp, td, 1));
1330 }
1331 
1332 int
1333 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1334 {
1335 	struct filedesc *fdp;
1336 	int fd, ret, lastfile;
1337 
1338 	ret = 0;
1339 	fdp = td->td_proc->p_fd;
1340 	FILEDESC_SLOCK(fdp);
1341 
1342 	/*
1343 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1344 	 * open should not be a usage error.  From a close_range() perspective,
1345 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1346 	 * be a usage error as all fd above 3 are in-fact already closed.
1347 	 */
1348 	if (highfd < lowfd) {
1349 		ret = EINVAL;
1350 		goto out;
1351 	}
1352 
1353 	/*
1354 	 * If lastfile == -1, we're dealing with either a fresh file
1355 	 * table or one in which every fd has been closed.  Just return
1356 	 * successful; there's nothing left to do.
1357 	 */
1358 	lastfile = fdlastfile(fdp);
1359 	if (lastfile == -1)
1360 		goto out;
1361 	/* Clamped to [lowfd, lastfile] */
1362 	highfd = MIN(highfd, lastfile);
1363 	for (fd = lowfd; fd <= highfd; fd++) {
1364 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1365 			FILEDESC_SUNLOCK(fdp);
1366 			(void)kern_close(td, fd);
1367 			FILEDESC_SLOCK(fdp);
1368 		}
1369 	}
1370 out:
1371 	FILEDESC_SUNLOCK(fdp);
1372 	return (ret);
1373 }
1374 
1375 #ifndef _SYS_SYSPROTO_H_
1376 struct close_range_args {
1377 	u_int	lowfd;
1378 	u_int	highfd;
1379 	int	flags;
1380 };
1381 #endif
1382 int
1383 sys_close_range(struct thread *td, struct close_range_args *uap)
1384 {
1385 
1386 	/* No flags currently defined */
1387 	if (uap->flags != 0)
1388 		return (EINVAL);
1389 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1390 }
1391 
1392 #ifdef COMPAT_FREEBSD12
1393 /*
1394  * Close open file descriptors.
1395  */
1396 #ifndef _SYS_SYSPROTO_H_
1397 struct freebsd12_closefrom_args {
1398 	int	lowfd;
1399 };
1400 #endif
1401 /* ARGSUSED */
1402 int
1403 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1404 {
1405 	u_int lowfd;
1406 
1407 	AUDIT_ARG_FD(uap->lowfd);
1408 
1409 	/*
1410 	 * Treat negative starting file descriptor values identical to
1411 	 * closefrom(0) which closes all files.
1412 	 */
1413 	lowfd = MAX(0, uap->lowfd);
1414 	return (kern_close_range(td, lowfd, ~0U));
1415 }
1416 #endif	/* COMPAT_FREEBSD12 */
1417 
1418 #if defined(COMPAT_43)
1419 /*
1420  * Return status information about a file descriptor.
1421  */
1422 #ifndef _SYS_SYSPROTO_H_
1423 struct ofstat_args {
1424 	int	fd;
1425 	struct	ostat *sb;
1426 };
1427 #endif
1428 /* ARGSUSED */
1429 int
1430 ofstat(struct thread *td, struct ofstat_args *uap)
1431 {
1432 	struct ostat oub;
1433 	struct stat ub;
1434 	int error;
1435 
1436 	error = kern_fstat(td, uap->fd, &ub);
1437 	if (error == 0) {
1438 		cvtstat(&ub, &oub);
1439 		error = copyout(&oub, uap->sb, sizeof(oub));
1440 	}
1441 	return (error);
1442 }
1443 #endif /* COMPAT_43 */
1444 
1445 #if defined(COMPAT_FREEBSD11)
1446 int
1447 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1448 {
1449 	struct stat sb;
1450 	struct freebsd11_stat osb;
1451 	int error;
1452 
1453 	error = kern_fstat(td, uap->fd, &sb);
1454 	if (error != 0)
1455 		return (error);
1456 	error = freebsd11_cvtstat(&sb, &osb);
1457 	if (error == 0)
1458 		error = copyout(&osb, uap->sb, sizeof(osb));
1459 	return (error);
1460 }
1461 #endif	/* COMPAT_FREEBSD11 */
1462 
1463 /*
1464  * Return status information about a file descriptor.
1465  */
1466 #ifndef _SYS_SYSPROTO_H_
1467 struct fstat_args {
1468 	int	fd;
1469 	struct	stat *sb;
1470 };
1471 #endif
1472 /* ARGSUSED */
1473 int
1474 sys_fstat(struct thread *td, struct fstat_args *uap)
1475 {
1476 	struct stat ub;
1477 	int error;
1478 
1479 	error = kern_fstat(td, uap->fd, &ub);
1480 	if (error == 0)
1481 		error = copyout(&ub, uap->sb, sizeof(ub));
1482 	return (error);
1483 }
1484 
1485 int
1486 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1487 {
1488 	struct file *fp;
1489 	int error;
1490 
1491 	AUDIT_ARG_FD(fd);
1492 
1493 	error = fget(td, fd, &cap_fstat_rights, &fp);
1494 	if (__predict_false(error != 0))
1495 		return (error);
1496 
1497 	AUDIT_ARG_FILE(td->td_proc, fp);
1498 
1499 	error = fo_stat(fp, sbp, td->td_ucred, td);
1500 	fdrop(fp, td);
1501 #ifdef __STAT_TIME_T_EXT
1502 	sbp->st_atim_ext = 0;
1503 	sbp->st_mtim_ext = 0;
1504 	sbp->st_ctim_ext = 0;
1505 	sbp->st_btim_ext = 0;
1506 #endif
1507 #ifdef KTRACE
1508 	if (KTRPOINT(td, KTR_STRUCT))
1509 		ktrstat_error(sbp, error);
1510 #endif
1511 	return (error);
1512 }
1513 
1514 #if defined(COMPAT_FREEBSD11)
1515 /*
1516  * Return status information about a file descriptor.
1517  */
1518 #ifndef _SYS_SYSPROTO_H_
1519 struct freebsd11_nfstat_args {
1520 	int	fd;
1521 	struct	nstat *sb;
1522 };
1523 #endif
1524 /* ARGSUSED */
1525 int
1526 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1527 {
1528 	struct nstat nub;
1529 	struct stat ub;
1530 	int error;
1531 
1532 	error = kern_fstat(td, uap->fd, &ub);
1533 	if (error == 0) {
1534 		freebsd11_cvtnstat(&ub, &nub);
1535 		error = copyout(&nub, uap->sb, sizeof(nub));
1536 	}
1537 	return (error);
1538 }
1539 #endif /* COMPAT_FREEBSD11 */
1540 
1541 /*
1542  * Return pathconf information about a file descriptor.
1543  */
1544 #ifndef _SYS_SYSPROTO_H_
1545 struct fpathconf_args {
1546 	int	fd;
1547 	int	name;
1548 };
1549 #endif
1550 /* ARGSUSED */
1551 int
1552 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1553 {
1554 	long value;
1555 	int error;
1556 
1557 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1558 	if (error == 0)
1559 		td->td_retval[0] = value;
1560 	return (error);
1561 }
1562 
1563 int
1564 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1565 {
1566 	struct file *fp;
1567 	struct vnode *vp;
1568 	int error;
1569 
1570 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1571 	if (error != 0)
1572 		return (error);
1573 
1574 	if (name == _PC_ASYNC_IO) {
1575 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1576 		goto out;
1577 	}
1578 	vp = fp->f_vnode;
1579 	if (vp != NULL) {
1580 		vn_lock(vp, LK_SHARED | LK_RETRY);
1581 		error = VOP_PATHCONF(vp, name, valuep);
1582 		VOP_UNLOCK(vp);
1583 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1584 		if (name != _PC_PIPE_BUF) {
1585 			error = EINVAL;
1586 		} else {
1587 			*valuep = PIPE_BUF;
1588 			error = 0;
1589 		}
1590 	} else {
1591 		error = EOPNOTSUPP;
1592 	}
1593 out:
1594 	fdrop(fp, td);
1595 	return (error);
1596 }
1597 
1598 /*
1599  * Copy filecaps structure allocating memory for ioctls array if needed.
1600  *
1601  * The last parameter indicates whether the fdtable is locked. If it is not and
1602  * ioctls are encountered, copying fails and the caller must lock the table.
1603  *
1604  * Note that if the table was not locked, the caller has to check the relevant
1605  * sequence counter to determine whether the operation was successful.
1606  */
1607 bool
1608 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1609 {
1610 	size_t size;
1611 
1612 	if (src->fc_ioctls != NULL && !locked)
1613 		return (false);
1614 	memcpy(dst, src, sizeof(*src));
1615 	if (src->fc_ioctls == NULL)
1616 		return (true);
1617 
1618 	KASSERT(src->fc_nioctls > 0,
1619 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1620 
1621 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1622 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1623 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1624 	return (true);
1625 }
1626 
1627 static u_long *
1628 filecaps_copy_prep(const struct filecaps *src)
1629 {
1630 	u_long *ioctls;
1631 	size_t size;
1632 
1633 	if (__predict_true(src->fc_ioctls == NULL))
1634 		return (NULL);
1635 
1636 	KASSERT(src->fc_nioctls > 0,
1637 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1638 
1639 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1640 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1641 	return (ioctls);
1642 }
1643 
1644 static void
1645 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1646     u_long *ioctls)
1647 {
1648 	size_t size;
1649 
1650 	*dst = *src;
1651 	if (__predict_true(src->fc_ioctls == NULL)) {
1652 		MPASS(ioctls == NULL);
1653 		return;
1654 	}
1655 
1656 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1657 	dst->fc_ioctls = ioctls;
1658 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1659 }
1660 
1661 /*
1662  * Move filecaps structure to the new place and clear the old place.
1663  */
1664 void
1665 filecaps_move(struct filecaps *src, struct filecaps *dst)
1666 {
1667 
1668 	*dst = *src;
1669 	bzero(src, sizeof(*src));
1670 }
1671 
1672 /*
1673  * Fill the given filecaps structure with full rights.
1674  */
1675 static void
1676 filecaps_fill(struct filecaps *fcaps)
1677 {
1678 
1679 	CAP_ALL(&fcaps->fc_rights);
1680 	fcaps->fc_ioctls = NULL;
1681 	fcaps->fc_nioctls = -1;
1682 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1683 }
1684 
1685 /*
1686  * Free memory allocated within filecaps structure.
1687  */
1688 void
1689 filecaps_free(struct filecaps *fcaps)
1690 {
1691 
1692 	free(fcaps->fc_ioctls, M_FILECAPS);
1693 	bzero(fcaps, sizeof(*fcaps));
1694 }
1695 
1696 static u_long *
1697 filecaps_free_prep(struct filecaps *fcaps)
1698 {
1699 	u_long *ioctls;
1700 
1701 	ioctls = fcaps->fc_ioctls;
1702 	bzero(fcaps, sizeof(*fcaps));
1703 	return (ioctls);
1704 }
1705 
1706 static void
1707 filecaps_free_finish(u_long *ioctls)
1708 {
1709 
1710 	free(ioctls, M_FILECAPS);
1711 }
1712 
1713 /*
1714  * Validate the given filecaps structure.
1715  */
1716 static void
1717 filecaps_validate(const struct filecaps *fcaps, const char *func)
1718 {
1719 
1720 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1721 	    ("%s: invalid rights", func));
1722 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1723 	    ("%s: invalid fcntls", func));
1724 	KASSERT(fcaps->fc_fcntls == 0 ||
1725 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1726 	    ("%s: fcntls without CAP_FCNTL", func));
1727 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1728 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1729 	    ("%s: invalid ioctls", func));
1730 	KASSERT(fcaps->fc_nioctls == 0 ||
1731 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1732 	    ("%s: ioctls without CAP_IOCTL", func));
1733 }
1734 
1735 static void
1736 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1737 {
1738 	int nfd1;
1739 
1740 	FILEDESC_XLOCK_ASSERT(fdp);
1741 
1742 	nfd1 = fdp->fd_nfiles * 2;
1743 	if (nfd1 < nfd)
1744 		nfd1 = nfd;
1745 	fdgrowtable(fdp, nfd1);
1746 }
1747 
1748 /*
1749  * Grow the file table to accommodate (at least) nfd descriptors.
1750  */
1751 static void
1752 fdgrowtable(struct filedesc *fdp, int nfd)
1753 {
1754 	struct filedesc0 *fdp0;
1755 	struct freetable *ft;
1756 	struct fdescenttbl *ntable;
1757 	struct fdescenttbl *otable;
1758 	int nnfiles, onfiles;
1759 	NDSLOTTYPE *nmap, *omap;
1760 
1761 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1762 
1763 	/* save old values */
1764 	onfiles = fdp->fd_nfiles;
1765 	otable = fdp->fd_files;
1766 	omap = fdp->fd_map;
1767 
1768 	/* compute the size of the new table */
1769 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1770 	if (nnfiles <= onfiles)
1771 		/* the table is already large enough */
1772 		return;
1773 
1774 	/*
1775 	 * Allocate a new table.  We need enough space for the number of
1776 	 * entries, file entries themselves and the struct freetable we will use
1777 	 * when we decommission the table and place it on the freelist.
1778 	 * We place the struct freetable in the middle so we don't have
1779 	 * to worry about padding.
1780 	 */
1781 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1782 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1783 	    sizeof(struct freetable),
1784 	    M_FILEDESC, M_ZERO | M_WAITOK);
1785 	/* copy the old data */
1786 	ntable->fdt_nfiles = nnfiles;
1787 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1788 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1789 
1790 	/*
1791 	 * Allocate a new map only if the old is not large enough.  It will
1792 	 * grow at a slower rate than the table as it can map more
1793 	 * entries than the table can hold.
1794 	 */
1795 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1796 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1797 		    M_ZERO | M_WAITOK);
1798 		/* copy over the old data and update the pointer */
1799 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1800 		fdp->fd_map = nmap;
1801 	}
1802 
1803 	/*
1804 	 * Make sure that ntable is correctly initialized before we replace
1805 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1806 	 * data.
1807 	 */
1808 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1809 
1810 	/*
1811 	 * Free the old file table when not shared by other threads or processes.
1812 	 * The old file table is considered to be shared when either are true:
1813 	 * - The process has more than one thread.
1814 	 * - The file descriptor table has been shared via fdshare().
1815 	 *
1816 	 * When shared, the old file table will be placed on a freelist
1817 	 * which will be processed when the struct filedesc is released.
1818 	 *
1819 	 * Note that if onfiles == NDFILE, we're dealing with the original
1820 	 * static allocation contained within (struct filedesc0 *)fdp,
1821 	 * which must not be freed.
1822 	 */
1823 	if (onfiles > NDFILE) {
1824 		if (curproc->p_numthreads == 1 && fdp->fd_refcnt == 1)
1825 			free(otable, M_FILEDESC);
1826 		else {
1827 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1828 			fdp0 = (struct filedesc0 *)fdp;
1829 			ft->ft_table = otable;
1830 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1831 		}
1832 	}
1833 	/*
1834 	 * The map does not have the same possibility of threads still
1835 	 * holding references to it.  So always free it as long as it
1836 	 * does not reference the original static allocation.
1837 	 */
1838 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1839 		free(omap, M_FILEDESC);
1840 }
1841 
1842 /*
1843  * Allocate a file descriptor for the process.
1844  */
1845 int
1846 fdalloc(struct thread *td, int minfd, int *result)
1847 {
1848 	struct proc *p = td->td_proc;
1849 	struct filedesc *fdp = p->p_fd;
1850 	int fd, maxfd, allocfd;
1851 #ifdef RACCT
1852 	int error;
1853 #endif
1854 
1855 	FILEDESC_XLOCK_ASSERT(fdp);
1856 
1857 	if (fdp->fd_freefile > minfd)
1858 		minfd = fdp->fd_freefile;
1859 
1860 	maxfd = getmaxfd(td);
1861 
1862 	/*
1863 	 * Search the bitmap for a free descriptor starting at minfd.
1864 	 * If none is found, grow the file table.
1865 	 */
1866 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1867 	if (__predict_false(fd >= maxfd))
1868 		return (EMFILE);
1869 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1870 		allocfd = min(fd * 2, maxfd);
1871 #ifdef RACCT
1872 		if (RACCT_ENABLED()) {
1873 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1874 			if (error != 0)
1875 				return (EMFILE);
1876 		}
1877 #endif
1878 		/*
1879 		 * fd is already equal to first free descriptor >= minfd, so
1880 		 * we only need to grow the table and we are done.
1881 		 */
1882 		fdgrowtable_exp(fdp, allocfd);
1883 	}
1884 
1885 	/*
1886 	 * Perform some sanity checks, then mark the file descriptor as
1887 	 * used and return it to the caller.
1888 	 */
1889 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1890 	    ("invalid descriptor %d", fd));
1891 	KASSERT(!fdisused(fdp, fd),
1892 	    ("fd_first_free() returned non-free descriptor"));
1893 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1894 	    ("file descriptor isn't free"));
1895 	fdused(fdp, fd);
1896 	*result = fd;
1897 	return (0);
1898 }
1899 
1900 /*
1901  * Allocate n file descriptors for the process.
1902  */
1903 int
1904 fdallocn(struct thread *td, int minfd, int *fds, int n)
1905 {
1906 	struct proc *p = td->td_proc;
1907 	struct filedesc *fdp = p->p_fd;
1908 	int i;
1909 
1910 	FILEDESC_XLOCK_ASSERT(fdp);
1911 
1912 	for (i = 0; i < n; i++)
1913 		if (fdalloc(td, 0, &fds[i]) != 0)
1914 			break;
1915 
1916 	if (i < n) {
1917 		for (i--; i >= 0; i--)
1918 			fdunused(fdp, fds[i]);
1919 		return (EMFILE);
1920 	}
1921 
1922 	return (0);
1923 }
1924 
1925 /*
1926  * Create a new open file structure and allocate a file descriptor for the
1927  * process that refers to it.  We add one reference to the file for the
1928  * descriptor table and one reference for resultfp. This is to prevent us
1929  * being preempted and the entry in the descriptor table closed after we
1930  * release the FILEDESC lock.
1931  */
1932 int
1933 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1934     struct filecaps *fcaps)
1935 {
1936 	struct file *fp;
1937 	int error, fd;
1938 
1939 	error = falloc_noinstall(td, &fp);
1940 	if (error)
1941 		return (error);		/* no reference held on error */
1942 
1943 	error = finstall(td, fp, &fd, flags, fcaps);
1944 	if (error) {
1945 		fdrop(fp, td);		/* one reference (fp only) */
1946 		return (error);
1947 	}
1948 
1949 	if (resultfp != NULL)
1950 		*resultfp = fp;		/* copy out result */
1951 	else
1952 		fdrop(fp, td);		/* release local reference */
1953 
1954 	if (resultfd != NULL)
1955 		*resultfd = fd;
1956 
1957 	return (0);
1958 }
1959 
1960 /*
1961  * Create a new open file structure without allocating a file descriptor.
1962  */
1963 int
1964 falloc_noinstall(struct thread *td, struct file **resultfp)
1965 {
1966 	struct file *fp;
1967 	int maxuserfiles = maxfiles - (maxfiles / 20);
1968 	int openfiles_new;
1969 	static struct timeval lastfail;
1970 	static int curfail;
1971 
1972 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1973 
1974 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1975 	if ((openfiles_new >= maxuserfiles &&
1976 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1977 	    openfiles_new >= maxfiles) {
1978 		atomic_subtract_int(&openfiles, 1);
1979 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1980 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1981 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1982 		}
1983 		return (ENFILE);
1984 	}
1985 	fp = uma_zalloc(file_zone, M_WAITOK);
1986 	bzero(fp, sizeof(*fp));
1987 	refcount_init(&fp->f_count, 1);
1988 	fp->f_cred = crhold(td->td_ucred);
1989 	fp->f_ops = &badfileops;
1990 	*resultfp = fp;
1991 	return (0);
1992 }
1993 
1994 /*
1995  * Install a file in a file descriptor table.
1996  */
1997 void
1998 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1999     struct filecaps *fcaps)
2000 {
2001 	struct filedescent *fde;
2002 
2003 	MPASS(fp != NULL);
2004 	if (fcaps != NULL)
2005 		filecaps_validate(fcaps, __func__);
2006 	FILEDESC_XLOCK_ASSERT(fdp);
2007 
2008 	fde = &fdp->fd_ofiles[fd];
2009 #ifdef CAPABILITIES
2010 	seqc_write_begin(&fde->fde_seqc);
2011 #endif
2012 	fde->fde_file = fp;
2013 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2014 	if (fcaps != NULL)
2015 		filecaps_move(fcaps, &fde->fde_caps);
2016 	else
2017 		filecaps_fill(&fde->fde_caps);
2018 #ifdef CAPABILITIES
2019 	seqc_write_end(&fde->fde_seqc);
2020 #endif
2021 }
2022 
2023 int
2024 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2025     struct filecaps *fcaps)
2026 {
2027 	struct filedesc *fdp = td->td_proc->p_fd;
2028 	int error;
2029 
2030 	MPASS(fd != NULL);
2031 
2032 	if (!fhold(fp))
2033 		return (EBADF);
2034 	FILEDESC_XLOCK(fdp);
2035 	error = fdalloc(td, 0, fd);
2036 	if (__predict_false(error != 0)) {
2037 		FILEDESC_XUNLOCK(fdp);
2038 		fdrop(fp, td);
2039 		return (error);
2040 	}
2041 	_finstall(fdp, fp, *fd, flags, fcaps);
2042 	FILEDESC_XUNLOCK(fdp);
2043 	return (0);
2044 }
2045 
2046 /*
2047  * Build a new filedesc structure from another.
2048  *
2049  * If fdp is not NULL, return with it shared locked.
2050  */
2051 struct filedesc *
2052 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2053 {
2054 	struct filedesc0 *newfdp0;
2055 	struct filedesc *newfdp;
2056 
2057 	if (prepfiles)
2058 		MPASS(lastfile != NULL);
2059 	else
2060 		MPASS(lastfile == NULL);
2061 
2062 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2063 	newfdp = &newfdp0->fd_fd;
2064 
2065 	/* Create the file descriptor table. */
2066 	FILEDESC_LOCK_INIT(newfdp);
2067 	refcount_init(&newfdp->fd_refcnt, 1);
2068 	refcount_init(&newfdp->fd_holdcnt, 1);
2069 	newfdp->fd_map = newfdp0->fd_dmap;
2070 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2071 	newfdp->fd_files->fdt_nfiles = NDFILE;
2072 
2073 	if (fdp == NULL)
2074 		return (newfdp);
2075 
2076 	FILEDESC_SLOCK(fdp);
2077 	if (!prepfiles) {
2078 		FILEDESC_SUNLOCK(fdp);
2079 		return (newfdp);
2080 	}
2081 
2082 	for (;;) {
2083 		*lastfile = fdlastfile(fdp);
2084 		if (*lastfile < newfdp->fd_nfiles)
2085 			break;
2086 		FILEDESC_SUNLOCK(fdp);
2087 		fdgrowtable(newfdp, *lastfile + 1);
2088 		FILEDESC_SLOCK(fdp);
2089 	}
2090 
2091 	return (newfdp);
2092 }
2093 
2094 /*
2095  * Build a pwddesc structure from another.
2096  * Copy the current, root, and jail root vnode references.
2097  *
2098  * If pdp is not NULL, return with it shared locked.
2099  */
2100 struct pwddesc *
2101 pdinit(struct pwddesc *pdp, bool keeplock)
2102 {
2103 	struct pwddesc *newpdp;
2104 	struct pwd *newpwd;
2105 
2106 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2107 
2108 	PWDDESC_LOCK_INIT(newpdp);
2109 	refcount_init(&newpdp->pd_refcount, 1);
2110 	newpdp->pd_cmask = CMASK;
2111 
2112 	if (pdp == NULL) {
2113 		newpwd = pwd_alloc();
2114 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2115 		return (newpdp);
2116 	}
2117 
2118 	PWDDESC_XLOCK(pdp);
2119 	newpwd = pwd_hold_pwddesc(pdp);
2120 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2121 	if (!keeplock)
2122 		PWDDESC_XUNLOCK(pdp);
2123 	return (newpdp);
2124 }
2125 
2126 static struct filedesc *
2127 fdhold(struct proc *p)
2128 {
2129 	struct filedesc *fdp;
2130 
2131 	PROC_LOCK_ASSERT(p, MA_OWNED);
2132 	fdp = p->p_fd;
2133 	if (fdp != NULL)
2134 		refcount_acquire(&fdp->fd_holdcnt);
2135 	return (fdp);
2136 }
2137 
2138 static struct pwddesc *
2139 pdhold(struct proc *p)
2140 {
2141 	struct pwddesc *pdp;
2142 
2143 	PROC_LOCK_ASSERT(p, MA_OWNED);
2144 	pdp = p->p_pd;
2145 	if (pdp != NULL)
2146 		refcount_acquire(&pdp->pd_refcount);
2147 	return (pdp);
2148 }
2149 
2150 static void
2151 fddrop(struct filedesc *fdp)
2152 {
2153 
2154 	if (fdp->fd_holdcnt > 1) {
2155 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2156 			return;
2157 	}
2158 
2159 	FILEDESC_LOCK_DESTROY(fdp);
2160 	uma_zfree(filedesc0_zone, fdp);
2161 }
2162 
2163 static void
2164 pddrop(struct pwddesc *pdp)
2165 {
2166 	struct pwd *pwd;
2167 
2168 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2169 		return;
2170 
2171 	PWDDESC_XLOCK(pdp);
2172 	if (refcount_release(&pdp->pd_refcount) == 0) {
2173 		PWDDESC_XUNLOCK(pdp);
2174 		return;
2175 	}
2176 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2177 	pwd_set(pdp, NULL);
2178 	PWDDESC_XUNLOCK(pdp);
2179 	pwd_drop(pwd);
2180 
2181 	PWDDESC_LOCK_DESTROY(pdp);
2182 	free(pdp, M_PWDDESC);
2183 }
2184 
2185 /*
2186  * Share a filedesc structure.
2187  */
2188 struct filedesc *
2189 fdshare(struct filedesc *fdp)
2190 {
2191 
2192 	refcount_acquire(&fdp->fd_refcnt);
2193 	return (fdp);
2194 }
2195 
2196 /*
2197  * Share a pwddesc structure.
2198  */
2199 struct pwddesc *
2200 pdshare(struct pwddesc *pdp)
2201 {
2202 	refcount_acquire(&pdp->pd_refcount);
2203 	return (pdp);
2204 }
2205 
2206 /*
2207  * Unshare a filedesc structure, if necessary by making a copy
2208  */
2209 void
2210 fdunshare(struct thread *td)
2211 {
2212 	struct filedesc *tmp;
2213 	struct proc *p = td->td_proc;
2214 
2215 	if (p->p_fd->fd_refcnt == 1)
2216 		return;
2217 
2218 	tmp = fdcopy(p->p_fd);
2219 	fdescfree(td);
2220 	p->p_fd = tmp;
2221 }
2222 
2223 /*
2224  * Unshare a pwddesc structure.
2225  */
2226 void
2227 pdunshare(struct thread *td)
2228 {
2229 	struct pwddesc *pdp;
2230 	struct proc *p;
2231 
2232 	p = td->td_proc;
2233 	/* Not shared. */
2234 	if (p->p_pd->pd_refcount == 1)
2235 		return;
2236 
2237 	pdp = pdcopy(p->p_pd);
2238 	pdescfree(td);
2239 	p->p_pd = pdp;
2240 }
2241 
2242 void
2243 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2244 {
2245 
2246 	fdescfree(td);
2247 	td->td_proc->p_fd = fdp;
2248 }
2249 
2250 /*
2251  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2252  * this is to ease callers, not catch errors.
2253  */
2254 struct filedesc *
2255 fdcopy(struct filedesc *fdp)
2256 {
2257 	struct filedesc *newfdp;
2258 	struct filedescent *nfde, *ofde;
2259 	int i, lastfile;
2260 
2261 	MPASS(fdp != NULL);
2262 
2263 	newfdp = fdinit(fdp, true, &lastfile);
2264 	/* copy all passable descriptors (i.e. not kqueue) */
2265 	newfdp->fd_freefile = -1;
2266 	for (i = 0; i <= lastfile; ++i) {
2267 		ofde = &fdp->fd_ofiles[i];
2268 		if (ofde->fde_file == NULL ||
2269 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2270 		    !fhold(ofde->fde_file)) {
2271 			if (newfdp->fd_freefile == -1)
2272 				newfdp->fd_freefile = i;
2273 			continue;
2274 		}
2275 		nfde = &newfdp->fd_ofiles[i];
2276 		*nfde = *ofde;
2277 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2278 		fdused_init(newfdp, i);
2279 	}
2280 	if (newfdp->fd_freefile == -1)
2281 		newfdp->fd_freefile = i;
2282 	FILEDESC_SUNLOCK(fdp);
2283 	return (newfdp);
2284 }
2285 
2286 /*
2287  * Copy a pwddesc structure.
2288  */
2289 struct pwddesc *
2290 pdcopy(struct pwddesc *pdp)
2291 {
2292 	struct pwddesc *newpdp;
2293 
2294 	MPASS(pdp != NULL);
2295 
2296 	newpdp = pdinit(pdp, true);
2297 	newpdp->pd_cmask = pdp->pd_cmask;
2298 	PWDDESC_XUNLOCK(pdp);
2299 	return (newpdp);
2300 }
2301 
2302 /*
2303  * Copies a filedesc structure, while remapping all file descriptors
2304  * stored inside using a translation table.
2305  *
2306  * File descriptors are copied over to the new file descriptor table,
2307  * regardless of whether the close-on-exec flag is set.
2308  */
2309 int
2310 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2311     struct filedesc **ret)
2312 {
2313 	struct filedesc *newfdp;
2314 	struct filedescent *nfde, *ofde;
2315 	int error, i, lastfile;
2316 
2317 	MPASS(fdp != NULL);
2318 
2319 	newfdp = fdinit(fdp, true, &lastfile);
2320 	if (nfds > lastfile + 1) {
2321 		/* New table cannot be larger than the old one. */
2322 		error = E2BIG;
2323 		goto bad;
2324 	}
2325 	/* Copy all passable descriptors (i.e. not kqueue). */
2326 	newfdp->fd_freefile = nfds;
2327 	for (i = 0; i < nfds; ++i) {
2328 		if (fds[i] < 0 || fds[i] > lastfile) {
2329 			/* File descriptor out of bounds. */
2330 			error = EBADF;
2331 			goto bad;
2332 		}
2333 		ofde = &fdp->fd_ofiles[fds[i]];
2334 		if (ofde->fde_file == NULL) {
2335 			/* Unused file descriptor. */
2336 			error = EBADF;
2337 			goto bad;
2338 		}
2339 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2340 			/* File descriptor cannot be passed. */
2341 			error = EINVAL;
2342 			goto bad;
2343 		}
2344 		if (!fhold(ofde->fde_file)) {
2345 			error = EBADF;
2346 			goto bad;
2347 		}
2348 		nfde = &newfdp->fd_ofiles[i];
2349 		*nfde = *ofde;
2350 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2351 		fdused_init(newfdp, i);
2352 	}
2353 	FILEDESC_SUNLOCK(fdp);
2354 	*ret = newfdp;
2355 	return (0);
2356 bad:
2357 	FILEDESC_SUNLOCK(fdp);
2358 	fdescfree_remapped(newfdp);
2359 	return (error);
2360 }
2361 
2362 /*
2363  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2364  * one of processes using it exits) and the table used to be shared.
2365  */
2366 static void
2367 fdclearlocks(struct thread *td)
2368 {
2369 	struct filedesc *fdp;
2370 	struct filedesc_to_leader *fdtol;
2371 	struct flock lf;
2372 	struct file *fp;
2373 	struct proc *p;
2374 	struct vnode *vp;
2375 	int i, lastfile;
2376 
2377 	p = td->td_proc;
2378 	fdp = p->p_fd;
2379 	fdtol = p->p_fdtol;
2380 	MPASS(fdtol != NULL);
2381 
2382 	FILEDESC_XLOCK(fdp);
2383 	KASSERT(fdtol->fdl_refcount > 0,
2384 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2385 	    fdtol->fdl_refcount));
2386 	if (fdtol->fdl_refcount == 1 &&
2387 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2388 		lastfile = fdlastfile(fdp);
2389 		for (i = 0; i <= lastfile; i++) {
2390 			fp = fdp->fd_ofiles[i].fde_file;
2391 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2392 			    !fhold(fp))
2393 				continue;
2394 			FILEDESC_XUNLOCK(fdp);
2395 			lf.l_whence = SEEK_SET;
2396 			lf.l_start = 0;
2397 			lf.l_len = 0;
2398 			lf.l_type = F_UNLCK;
2399 			vp = fp->f_vnode;
2400 			(void) VOP_ADVLOCK(vp,
2401 			    (caddr_t)p->p_leader, F_UNLCK,
2402 			    &lf, F_POSIX);
2403 			FILEDESC_XLOCK(fdp);
2404 			fdrop(fp, td);
2405 		}
2406 	}
2407 retry:
2408 	if (fdtol->fdl_refcount == 1) {
2409 		if (fdp->fd_holdleaderscount > 0 &&
2410 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2411 			/*
2412 			 * close() or kern_dup() has cleared a reference
2413 			 * in a shared file descriptor table.
2414 			 */
2415 			fdp->fd_holdleaderswakeup = 1;
2416 			sx_sleep(&fdp->fd_holdleaderscount,
2417 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2418 			goto retry;
2419 		}
2420 		if (fdtol->fdl_holdcount > 0) {
2421 			/*
2422 			 * Ensure that fdtol->fdl_leader remains
2423 			 * valid in closef().
2424 			 */
2425 			fdtol->fdl_wakeup = 1;
2426 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2427 			    "fdlhold", 0);
2428 			goto retry;
2429 		}
2430 	}
2431 	fdtol->fdl_refcount--;
2432 	if (fdtol->fdl_refcount == 0 &&
2433 	    fdtol->fdl_holdcount == 0) {
2434 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2435 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2436 	} else
2437 		fdtol = NULL;
2438 	p->p_fdtol = NULL;
2439 	FILEDESC_XUNLOCK(fdp);
2440 	if (fdtol != NULL)
2441 		free(fdtol, M_FILEDESC_TO_LEADER);
2442 }
2443 
2444 /*
2445  * Release a filedesc structure.
2446  */
2447 static void
2448 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2449 {
2450 	struct filedesc0 *fdp0;
2451 	struct freetable *ft, *tft;
2452 	struct filedescent *fde;
2453 	struct file *fp;
2454 	int i, lastfile;
2455 
2456 	lastfile = fdlastfile_single(fdp);
2457 	for (i = 0; i <= lastfile; i++) {
2458 		fde = &fdp->fd_ofiles[i];
2459 		fp = fde->fde_file;
2460 		if (fp != NULL) {
2461 			fdefree_last(fde);
2462 			if (needclose)
2463 				(void) closef(fp, td);
2464 			else
2465 				fdrop(fp, td);
2466 		}
2467 	}
2468 
2469 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2470 		free(fdp->fd_map, M_FILEDESC);
2471 	if (fdp->fd_nfiles > NDFILE)
2472 		free(fdp->fd_files, M_FILEDESC);
2473 
2474 	fdp0 = (struct filedesc0 *)fdp;
2475 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2476 		free(ft->ft_table, M_FILEDESC);
2477 
2478 	fddrop(fdp);
2479 }
2480 
2481 void
2482 fdescfree(struct thread *td)
2483 {
2484 	struct proc *p;
2485 	struct filedesc *fdp;
2486 
2487 	p = td->td_proc;
2488 	fdp = p->p_fd;
2489 	MPASS(fdp != NULL);
2490 
2491 #ifdef RACCT
2492 	if (RACCT_ENABLED())
2493 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2494 #endif
2495 
2496 	if (p->p_fdtol != NULL)
2497 		fdclearlocks(td);
2498 
2499 	PROC_LOCK(p);
2500 	p->p_fd = NULL;
2501 	PROC_UNLOCK(p);
2502 
2503 	if (refcount_release(&fdp->fd_refcnt) == 0)
2504 		return;
2505 
2506 	fdescfree_fds(td, fdp, 1);
2507 }
2508 
2509 void
2510 pdescfree(struct thread *td)
2511 {
2512 	struct proc *p;
2513 	struct pwddesc *pdp;
2514 
2515 	p = td->td_proc;
2516 	pdp = p->p_pd;
2517 	MPASS(pdp != NULL);
2518 
2519 	PROC_LOCK(p);
2520 	p->p_pd = NULL;
2521 	PROC_UNLOCK(p);
2522 
2523 	pddrop(pdp);
2524 }
2525 
2526 void
2527 fdescfree_remapped(struct filedesc *fdp)
2528 {
2529 	fdescfree_fds(curthread, fdp, 0);
2530 }
2531 
2532 /*
2533  * For setugid programs, we don't want to people to use that setugidness
2534  * to generate error messages which write to a file which otherwise would
2535  * otherwise be off-limits to the process.  We check for filesystems where
2536  * the vnode can change out from under us after execve (like [lin]procfs).
2537  *
2538  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2539  * sufficient.  We also don't check for setugidness since we know we are.
2540  */
2541 static bool
2542 is_unsafe(struct file *fp)
2543 {
2544 	struct vnode *vp;
2545 
2546 	if (fp->f_type != DTYPE_VNODE)
2547 		return (false);
2548 
2549 	vp = fp->f_vnode;
2550 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2551 }
2552 
2553 /*
2554  * Make this setguid thing safe, if at all possible.
2555  */
2556 void
2557 fdsetugidsafety(struct thread *td)
2558 {
2559 	struct filedesc *fdp;
2560 	struct file *fp;
2561 	int i;
2562 
2563 	fdp = td->td_proc->p_fd;
2564 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2565 	MPASS(fdp->fd_nfiles >= 3);
2566 	for (i = 0; i <= 2; i++) {
2567 		fp = fdp->fd_ofiles[i].fde_file;
2568 		if (fp != NULL && is_unsafe(fp)) {
2569 			FILEDESC_XLOCK(fdp);
2570 			knote_fdclose(td, i);
2571 			/*
2572 			 * NULL-out descriptor prior to close to avoid
2573 			 * a race while close blocks.
2574 			 */
2575 			fdfree(fdp, i);
2576 			FILEDESC_XUNLOCK(fdp);
2577 			(void) closef(fp, td);
2578 		}
2579 	}
2580 }
2581 
2582 /*
2583  * If a specific file object occupies a specific file descriptor, close the
2584  * file descriptor entry and drop a reference on the file object.  This is a
2585  * convenience function to handle a subsequent error in a function that calls
2586  * falloc() that handles the race that another thread might have closed the
2587  * file descriptor out from under the thread creating the file object.
2588  */
2589 void
2590 fdclose(struct thread *td, struct file *fp, int idx)
2591 {
2592 	struct filedesc *fdp = td->td_proc->p_fd;
2593 
2594 	FILEDESC_XLOCK(fdp);
2595 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2596 		fdfree(fdp, idx);
2597 		FILEDESC_XUNLOCK(fdp);
2598 		fdrop(fp, td);
2599 	} else
2600 		FILEDESC_XUNLOCK(fdp);
2601 }
2602 
2603 /*
2604  * Close any files on exec?
2605  */
2606 void
2607 fdcloseexec(struct thread *td)
2608 {
2609 	struct filedesc *fdp;
2610 	struct filedescent *fde;
2611 	struct file *fp;
2612 	int i, lastfile;
2613 
2614 	fdp = td->td_proc->p_fd;
2615 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2616 	lastfile = fdlastfile_single(fdp);
2617 	for (i = 0; i <= lastfile; i++) {
2618 		fde = &fdp->fd_ofiles[i];
2619 		fp = fde->fde_file;
2620 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2621 		    (fde->fde_flags & UF_EXCLOSE))) {
2622 			FILEDESC_XLOCK(fdp);
2623 			fdfree(fdp, i);
2624 			(void) closefp(fdp, i, fp, td, 0);
2625 			FILEDESC_UNLOCK_ASSERT(fdp);
2626 		}
2627 	}
2628 }
2629 
2630 /*
2631  * It is unsafe for set[ug]id processes to be started with file
2632  * descriptors 0..2 closed, as these descriptors are given implicit
2633  * significance in the Standard C library.  fdcheckstd() will create a
2634  * descriptor referencing /dev/null for each of stdin, stdout, and
2635  * stderr that is not already open.
2636  */
2637 int
2638 fdcheckstd(struct thread *td)
2639 {
2640 	struct filedesc *fdp;
2641 	register_t save;
2642 	int i, error, devnull;
2643 
2644 	fdp = td->td_proc->p_fd;
2645 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2646 	MPASS(fdp->fd_nfiles >= 3);
2647 	devnull = -1;
2648 	for (i = 0; i <= 2; i++) {
2649 		if (fdp->fd_ofiles[i].fde_file != NULL)
2650 			continue;
2651 
2652 		save = td->td_retval[0];
2653 		if (devnull != -1) {
2654 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2655 		} else {
2656 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2657 			    UIO_SYSSPACE, O_RDWR, 0);
2658 			if (error == 0) {
2659 				devnull = td->td_retval[0];
2660 				KASSERT(devnull == i, ("we didn't get our fd"));
2661 			}
2662 		}
2663 		td->td_retval[0] = save;
2664 		if (error != 0)
2665 			return (error);
2666 	}
2667 	return (0);
2668 }
2669 
2670 /*
2671  * Internal form of close.  Decrement reference count on file structure.
2672  * Note: td may be NULL when closing a file that was being passed in a
2673  * message.
2674  */
2675 int
2676 closef(struct file *fp, struct thread *td)
2677 {
2678 	struct vnode *vp;
2679 	struct flock lf;
2680 	struct filedesc_to_leader *fdtol;
2681 	struct filedesc *fdp;
2682 
2683 	/*
2684 	 * POSIX record locking dictates that any close releases ALL
2685 	 * locks owned by this process.  This is handled by setting
2686 	 * a flag in the unlock to free ONLY locks obeying POSIX
2687 	 * semantics, and not to free BSD-style file locks.
2688 	 * If the descriptor was in a message, POSIX-style locks
2689 	 * aren't passed with the descriptor, and the thread pointer
2690 	 * will be NULL.  Callers should be careful only to pass a
2691 	 * NULL thread pointer when there really is no owning
2692 	 * context that might have locks, or the locks will be
2693 	 * leaked.
2694 	 */
2695 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2696 		vp = fp->f_vnode;
2697 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2698 			lf.l_whence = SEEK_SET;
2699 			lf.l_start = 0;
2700 			lf.l_len = 0;
2701 			lf.l_type = F_UNLCK;
2702 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2703 			    F_UNLCK, &lf, F_POSIX);
2704 		}
2705 		fdtol = td->td_proc->p_fdtol;
2706 		if (fdtol != NULL) {
2707 			/*
2708 			 * Handle special case where file descriptor table is
2709 			 * shared between multiple process leaders.
2710 			 */
2711 			fdp = td->td_proc->p_fd;
2712 			FILEDESC_XLOCK(fdp);
2713 			for (fdtol = fdtol->fdl_next;
2714 			    fdtol != td->td_proc->p_fdtol;
2715 			    fdtol = fdtol->fdl_next) {
2716 				if ((fdtol->fdl_leader->p_flag &
2717 				    P_ADVLOCK) == 0)
2718 					continue;
2719 				fdtol->fdl_holdcount++;
2720 				FILEDESC_XUNLOCK(fdp);
2721 				lf.l_whence = SEEK_SET;
2722 				lf.l_start = 0;
2723 				lf.l_len = 0;
2724 				lf.l_type = F_UNLCK;
2725 				vp = fp->f_vnode;
2726 				(void) VOP_ADVLOCK(vp,
2727 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2728 				    F_POSIX);
2729 				FILEDESC_XLOCK(fdp);
2730 				fdtol->fdl_holdcount--;
2731 				if (fdtol->fdl_holdcount == 0 &&
2732 				    fdtol->fdl_wakeup != 0) {
2733 					fdtol->fdl_wakeup = 0;
2734 					wakeup(fdtol);
2735 				}
2736 			}
2737 			FILEDESC_XUNLOCK(fdp);
2738 		}
2739 	}
2740 	return (fdrop(fp, td));
2741 }
2742 
2743 /*
2744  * Initialize the file pointer with the specified properties.
2745  *
2746  * The ops are set with release semantics to be certain that the flags, type,
2747  * and data are visible when ops is.  This is to prevent ops methods from being
2748  * called with bad data.
2749  */
2750 void
2751 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2752 {
2753 	fp->f_data = data;
2754 	fp->f_flag = flag;
2755 	fp->f_type = type;
2756 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2757 }
2758 
2759 void
2760 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2761 {
2762 	fp->f_seqcount[UIO_READ] = 1;
2763 	fp->f_seqcount[UIO_WRITE] = 1;
2764 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2765 	    data, ops);
2766 }
2767 
2768 int
2769 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2770     struct file **fpp, struct filecaps *havecapsp)
2771 {
2772 	struct filedescent *fde;
2773 	int error;
2774 
2775 	FILEDESC_LOCK_ASSERT(fdp);
2776 
2777 	fde = fdeget_locked(fdp, fd);
2778 	if (fde == NULL) {
2779 		error = EBADF;
2780 		goto out;
2781 	}
2782 
2783 #ifdef CAPABILITIES
2784 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2785 	if (error != 0)
2786 		goto out;
2787 #endif
2788 
2789 	if (havecapsp != NULL)
2790 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2791 
2792 	*fpp = fde->fde_file;
2793 
2794 	error = 0;
2795 out:
2796 	return (error);
2797 }
2798 
2799 int
2800 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2801     struct file **fpp, struct filecaps *havecapsp)
2802 {
2803 	struct filedesc *fdp = td->td_proc->p_fd;
2804 	int error;
2805 #ifndef CAPABILITIES
2806 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2807 	if (havecapsp != NULL && error == 0)
2808 		filecaps_fill(havecapsp);
2809 #else
2810 	struct file *fp;
2811 	seqc_t seq;
2812 
2813 	*fpp = NULL;
2814 	for (;;) {
2815 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2816 		if (error != 0)
2817 			return (error);
2818 
2819 		if (havecapsp != NULL) {
2820 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2821 			    havecapsp, false)) {
2822 				fdrop(fp, td);
2823 				goto get_locked;
2824 			}
2825 		}
2826 
2827 		if (!fd_modified(fdp, fd, seq))
2828 			break;
2829 		fdrop(fp, td);
2830 	}
2831 
2832 	*fpp = fp;
2833 	return (0);
2834 
2835 get_locked:
2836 	FILEDESC_SLOCK(fdp);
2837 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2838 	if (error == 0 && !fhold(*fpp))
2839 		error = EBADF;
2840 	FILEDESC_SUNLOCK(fdp);
2841 #endif
2842 	return (error);
2843 }
2844 
2845 #ifdef CAPABILITIES
2846 int
2847 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2848 {
2849 	const struct filedescent *fde;
2850 	const struct fdescenttbl *fdt;
2851 	struct filedesc *fdp;
2852 	struct file *fp;
2853 	struct vnode *vp;
2854 	const cap_rights_t *haverights;
2855 	cap_rights_t rights;
2856 	seqc_t seq;
2857 
2858 	VFS_SMR_ASSERT_ENTERED();
2859 
2860 	rights = *ndp->ni_rightsneeded;
2861 	cap_rights_set_one(&rights, CAP_LOOKUP);
2862 
2863 	fdp = curproc->p_fd;
2864 	fdt = fdp->fd_files;
2865 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2866 		return (EBADF);
2867 	seq = seqc_read_any(fd_seqc(fdt, fd));
2868 	if (__predict_false(seqc_in_modify(seq)))
2869 		return (EAGAIN);
2870 	fde = &fdt->fdt_ofiles[fd];
2871 	haverights = cap_rights_fde_inline(fde);
2872 	fp = fde->fde_file;
2873 	if (__predict_false(fp == NULL))
2874 		return (EAGAIN);
2875 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2876 		return (EAGAIN);
2877 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2878 	vp = fp->f_vnode;
2879 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2880 		return (EAGAIN);
2881 	}
2882 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2883 		return (EAGAIN);
2884 	}
2885 	/*
2886 	 * Use an acquire barrier to force re-reading of fdt so it is
2887 	 * refreshed for verification.
2888 	 */
2889 	atomic_thread_fence_acq();
2890 	fdt = fdp->fd_files;
2891 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2892 		return (EAGAIN);
2893 	/*
2894 	 * If file descriptor doesn't have all rights,
2895 	 * all lookups relative to it must also be
2896 	 * strictly relative.
2897 	 *
2898 	 * Not yet supported by fast path.
2899 	 */
2900 	CAP_ALL(&rights);
2901 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2902 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2903 	    ndp->ni_filecaps.fc_nioctls != -1) {
2904 #ifdef notyet
2905 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2906 #else
2907 		return (EAGAIN);
2908 #endif
2909 	}
2910 	*vpp = vp;
2911 	return (0);
2912 }
2913 #else
2914 int
2915 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2916 {
2917 	const struct fdescenttbl *fdt;
2918 	struct filedesc *fdp;
2919 	struct file *fp;
2920 	struct vnode *vp;
2921 
2922 	VFS_SMR_ASSERT_ENTERED();
2923 
2924 	fdp = curproc->p_fd;
2925 	fdt = fdp->fd_files;
2926 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2927 		return (EBADF);
2928 	fp = fdt->fdt_ofiles[fd].fde_file;
2929 	if (__predict_false(fp == NULL))
2930 		return (EAGAIN);
2931 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2932 	vp = fp->f_vnode;
2933 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2934 		return (EAGAIN);
2935 	}
2936 	/*
2937 	 * Use an acquire barrier to force re-reading of fdt so it is
2938 	 * refreshed for verification.
2939 	 */
2940 	atomic_thread_fence_acq();
2941 	fdt = fdp->fd_files;
2942 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2943 		return (EAGAIN);
2944 	filecaps_fill(&ndp->ni_filecaps);
2945 	*vpp = vp;
2946 	return (0);
2947 }
2948 #endif
2949 
2950 int
2951 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2952     struct file **fpp, seqc_t *seqp)
2953 {
2954 #ifdef CAPABILITIES
2955 	const struct filedescent *fde;
2956 #endif
2957 	const struct fdescenttbl *fdt;
2958 	struct file *fp;
2959 #ifdef CAPABILITIES
2960 	seqc_t seq;
2961 	cap_rights_t haverights;
2962 	int error;
2963 #endif
2964 
2965 	fdt = fdp->fd_files;
2966 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2967 		return (EBADF);
2968 	/*
2969 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2970 	 * never raising a refcount above 0.  To accomplish this we have
2971 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2972 	 * must be re-verified once we acquire a reference to be certain
2973 	 * that the identity is still correct and we did not lose a race
2974 	 * due to preemption.
2975 	 */
2976 	for (;;) {
2977 #ifdef CAPABILITIES
2978 		seq = seqc_read(fd_seqc(fdt, fd));
2979 		fde = &fdt->fdt_ofiles[fd];
2980 		haverights = *cap_rights_fde_inline(fde);
2981 		fp = fde->fde_file;
2982 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2983 			continue;
2984 #else
2985 		fp = fdt->fdt_ofiles[fd].fde_file;
2986 #endif
2987 		if (fp == NULL)
2988 			return (EBADF);
2989 #ifdef CAPABILITIES
2990 		error = cap_check_inline(&haverights, needrightsp);
2991 		if (error != 0)
2992 			return (error);
2993 #endif
2994 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
2995 			/*
2996 			 * The count was found either saturated or zero.
2997 			 * This re-read is not any more racy than using the
2998 			 * return value from fcmpset.
2999 			 */
3000 			if (refcount_load(&fp->f_count) != 0)
3001 				return (EBADF);
3002 			/*
3003 			 * Force a reload. Other thread could reallocate the
3004 			 * table before this fd was closed, so it is possible
3005 			 * that there is a stale fp pointer in cached version.
3006 			 */
3007 			fdt = atomic_load_ptr(&fdp->fd_files);
3008 			continue;
3009 		}
3010 		/*
3011 		 * Use an acquire barrier to force re-reading of fdt so it is
3012 		 * refreshed for verification.
3013 		 */
3014 		atomic_thread_fence_acq();
3015 		fdt = fdp->fd_files;
3016 #ifdef	CAPABILITIES
3017 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3018 #else
3019 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3020 #endif
3021 			break;
3022 		fdrop(fp, curthread);
3023 	}
3024 	*fpp = fp;
3025 	if (seqp != NULL) {
3026 #ifdef CAPABILITIES
3027 		*seqp = seq;
3028 #endif
3029 	}
3030 	return (0);
3031 }
3032 
3033 /*
3034  * See the comments in fget_unlocked_seq for an explanation of how this works.
3035  *
3036  * This is a simplified variant which bails out to the aforementioned routine
3037  * if anything goes wrong. In practice this only happens when userspace is
3038  * racing with itself.
3039  */
3040 int
3041 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3042     struct file **fpp)
3043 {
3044 #ifdef CAPABILITIES
3045 	const struct filedescent *fde;
3046 #endif
3047 	const struct fdescenttbl *fdt;
3048 	struct file *fp;
3049 #ifdef CAPABILITIES
3050 	seqc_t seq;
3051 	const cap_rights_t *haverights;
3052 #endif
3053 
3054 	fdt = fdp->fd_files;
3055 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3056 		return (EBADF);
3057 #ifdef CAPABILITIES
3058 	seq = seqc_read_any(fd_seqc(fdt, fd));
3059 	if (__predict_false(seqc_in_modify(seq)))
3060 		goto out_fallback;
3061 	fde = &fdt->fdt_ofiles[fd];
3062 	haverights = cap_rights_fde_inline(fde);
3063 	fp = fde->fde_file;
3064 #else
3065 	fp = fdt->fdt_ofiles[fd].fde_file;
3066 #endif
3067 	if (__predict_false(fp == NULL))
3068 		goto out_fallback;
3069 #ifdef CAPABILITIES
3070 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3071 		goto out_fallback;
3072 #endif
3073 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3074 		goto out_fallback;
3075 
3076 	/*
3077 	 * Use an acquire barrier to force re-reading of fdt so it is
3078 	 * refreshed for verification.
3079 	 */
3080 	atomic_thread_fence_acq();
3081 	fdt = fdp->fd_files;
3082 #ifdef	CAPABILITIES
3083 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3084 #else
3085 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3086 #endif
3087 		goto out_fdrop;
3088 	*fpp = fp;
3089 	return (0);
3090 out_fdrop:
3091 	fdrop(fp, curthread);
3092 out_fallback:
3093 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3094 }
3095 
3096 /*
3097  * Extract the file pointer associated with the specified descriptor for the
3098  * current user process.
3099  *
3100  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3101  * returned.
3102  *
3103  * File's rights will be checked against the capability rights mask.
3104  *
3105  * If an error occurred the non-zero error is returned and *fpp is set to
3106  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3107  * responsible for fdrop().
3108  */
3109 static __inline int
3110 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3111     cap_rights_t *needrightsp)
3112 {
3113 	struct filedesc *fdp;
3114 	struct file *fp;
3115 	int error;
3116 
3117 	*fpp = NULL;
3118 	fdp = td->td_proc->p_fd;
3119 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3120 	if (__predict_false(error != 0))
3121 		return (error);
3122 	if (__predict_false(fp->f_ops == &badfileops)) {
3123 		fdrop(fp, td);
3124 		return (EBADF);
3125 	}
3126 
3127 	/*
3128 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3129 	 */
3130 	error = 0;
3131 	switch (flags) {
3132 	case FREAD:
3133 	case FWRITE:
3134 		if ((fp->f_flag & flags) == 0)
3135 			error = EBADF;
3136 		break;
3137 	case FEXEC:
3138 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3139 		    ((fp->f_flag & FWRITE) != 0))
3140 			error = EBADF;
3141 		break;
3142 	case 0:
3143 		break;
3144 	default:
3145 		KASSERT(0, ("wrong flags"));
3146 	}
3147 
3148 	if (error != 0) {
3149 		fdrop(fp, td);
3150 		return (error);
3151 	}
3152 
3153 	*fpp = fp;
3154 	return (0);
3155 }
3156 
3157 int
3158 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3159 {
3160 
3161 	return (_fget(td, fd, fpp, 0, rightsp));
3162 }
3163 
3164 int
3165 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3166     struct file **fpp)
3167 {
3168 	int error;
3169 #ifndef CAPABILITIES
3170 	error = _fget(td, fd, fpp, 0, rightsp);
3171 	if (maxprotp != NULL)
3172 		*maxprotp = VM_PROT_ALL;
3173 	return (error);
3174 #else
3175 	cap_rights_t fdrights;
3176 	struct filedesc *fdp;
3177 	struct file *fp;
3178 	seqc_t seq;
3179 
3180 	*fpp = NULL;
3181 	fdp = td->td_proc->p_fd;
3182 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3183 	for (;;) {
3184 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3185 		if (__predict_false(error != 0))
3186 			return (error);
3187 		if (__predict_false(fp->f_ops == &badfileops)) {
3188 			fdrop(fp, td);
3189 			return (EBADF);
3190 		}
3191 		if (maxprotp != NULL)
3192 			fdrights = *cap_rights(fdp, fd);
3193 		if (!fd_modified(fdp, fd, seq))
3194 			break;
3195 		fdrop(fp, td);
3196 	}
3197 
3198 	/*
3199 	 * If requested, convert capability rights to access flags.
3200 	 */
3201 	if (maxprotp != NULL)
3202 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3203 	*fpp = fp;
3204 	return (0);
3205 #endif
3206 }
3207 
3208 int
3209 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3210 {
3211 
3212 	return (_fget(td, fd, fpp, FREAD, rightsp));
3213 }
3214 
3215 int
3216 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3217 {
3218 
3219 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3220 }
3221 
3222 int
3223 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3224     struct file **fpp)
3225 {
3226 	struct filedesc *fdp = td->td_proc->p_fd;
3227 #ifndef CAPABILITIES
3228 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3229 #else
3230 	struct file *fp;
3231 	int error;
3232 	seqc_t seq;
3233 
3234 	*fpp = NULL;
3235 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3236 	for (;;) {
3237 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3238 		if (error != 0)
3239 			return (error);
3240 		error = cap_fcntl_check(fdp, fd, needfcntl);
3241 		if (!fd_modified(fdp, fd, seq))
3242 			break;
3243 		fdrop(fp, td);
3244 	}
3245 	if (error != 0) {
3246 		fdrop(fp, td);
3247 		return (error);
3248 	}
3249 	*fpp = fp;
3250 	return (0);
3251 #endif
3252 }
3253 
3254 /*
3255  * Like fget() but loads the underlying vnode, or returns an error if the
3256  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3257  * never have VM objects.  The returned vnode will be vref()'d.
3258  *
3259  * XXX: what about the unused flags ?
3260  */
3261 static __inline int
3262 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3263     struct vnode **vpp)
3264 {
3265 	struct file *fp;
3266 	int error;
3267 
3268 	*vpp = NULL;
3269 	error = _fget(td, fd, &fp, flags, needrightsp);
3270 	if (error != 0)
3271 		return (error);
3272 	if (fp->f_vnode == NULL) {
3273 		error = EINVAL;
3274 	} else {
3275 		*vpp = fp->f_vnode;
3276 		vrefact(*vpp);
3277 	}
3278 	fdrop(fp, td);
3279 
3280 	return (error);
3281 }
3282 
3283 int
3284 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3285 {
3286 
3287 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3288 }
3289 
3290 int
3291 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3292     struct filecaps *havecaps, struct vnode **vpp)
3293 {
3294 	struct filecaps caps;
3295 	struct file *fp;
3296 	int error;
3297 
3298 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3299 	if (error != 0)
3300 		return (error);
3301 	if (fp->f_ops == &badfileops) {
3302 		error = EBADF;
3303 		goto out;
3304 	}
3305 	if (fp->f_vnode == NULL) {
3306 		error = EINVAL;
3307 		goto out;
3308 	}
3309 
3310 	*havecaps = caps;
3311 	*vpp = fp->f_vnode;
3312 	vrefact(*vpp);
3313 	fdrop(fp, td);
3314 
3315 	return (0);
3316 out:
3317 	filecaps_free(&caps);
3318 	fdrop(fp, td);
3319 	return (error);
3320 }
3321 
3322 int
3323 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3324 {
3325 
3326 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3327 }
3328 
3329 int
3330 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3331 {
3332 
3333 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3334 }
3335 
3336 #ifdef notyet
3337 int
3338 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3339     struct vnode **vpp)
3340 {
3341 
3342 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3343 }
3344 #endif
3345 
3346 /*
3347  * Handle the last reference to a file being closed.
3348  *
3349  * Without the noinline attribute clang keeps inlining the func thorough this
3350  * file when fdrop is used.
3351  */
3352 int __noinline
3353 _fdrop(struct file *fp, struct thread *td)
3354 {
3355 	int error;
3356 #ifdef INVARIANTS
3357 	int count;
3358 
3359 	count = refcount_load(&fp->f_count);
3360 	if (count != 0)
3361 		panic("fdrop: fp %p count %d", fp, count);
3362 #endif
3363 	error = fo_close(fp, td);
3364 	atomic_subtract_int(&openfiles, 1);
3365 	crfree(fp->f_cred);
3366 	free(fp->f_advice, M_FADVISE);
3367 	uma_zfree(file_zone, fp);
3368 
3369 	return (error);
3370 }
3371 
3372 /*
3373  * Apply an advisory lock on a file descriptor.
3374  *
3375  * Just attempt to get a record lock of the requested type on the entire file
3376  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3377  */
3378 #ifndef _SYS_SYSPROTO_H_
3379 struct flock_args {
3380 	int	fd;
3381 	int	how;
3382 };
3383 #endif
3384 /* ARGSUSED */
3385 int
3386 sys_flock(struct thread *td, struct flock_args *uap)
3387 {
3388 	struct file *fp;
3389 	struct vnode *vp;
3390 	struct flock lf;
3391 	int error;
3392 
3393 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3394 	if (error != 0)
3395 		return (error);
3396 	if (fp->f_type != DTYPE_VNODE) {
3397 		fdrop(fp, td);
3398 		return (EOPNOTSUPP);
3399 	}
3400 
3401 	vp = fp->f_vnode;
3402 	lf.l_whence = SEEK_SET;
3403 	lf.l_start = 0;
3404 	lf.l_len = 0;
3405 	if (uap->how & LOCK_UN) {
3406 		lf.l_type = F_UNLCK;
3407 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3408 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3409 		goto done2;
3410 	}
3411 	if (uap->how & LOCK_EX)
3412 		lf.l_type = F_WRLCK;
3413 	else if (uap->how & LOCK_SH)
3414 		lf.l_type = F_RDLCK;
3415 	else {
3416 		error = EBADF;
3417 		goto done2;
3418 	}
3419 	atomic_set_int(&fp->f_flag, FHASLOCK);
3420 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3421 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3422 done2:
3423 	fdrop(fp, td);
3424 	return (error);
3425 }
3426 /*
3427  * Duplicate the specified descriptor to a free descriptor.
3428  */
3429 int
3430 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3431     int openerror, int *indxp)
3432 {
3433 	struct filedescent *newfde, *oldfde;
3434 	struct file *fp;
3435 	u_long *ioctls;
3436 	int error, indx;
3437 
3438 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3439 	    ("unexpected error %d in %s", openerror, __func__));
3440 
3441 	/*
3442 	 * If the to-be-dup'd fd number is greater than the allowed number
3443 	 * of file descriptors, or the fd to be dup'd has already been
3444 	 * closed, then reject.
3445 	 */
3446 	FILEDESC_XLOCK(fdp);
3447 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3448 		FILEDESC_XUNLOCK(fdp);
3449 		return (EBADF);
3450 	}
3451 
3452 	error = fdalloc(td, 0, &indx);
3453 	if (error != 0) {
3454 		FILEDESC_XUNLOCK(fdp);
3455 		return (error);
3456 	}
3457 
3458 	/*
3459 	 * There are two cases of interest here.
3460 	 *
3461 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3462 	 *
3463 	 * For ENXIO steal away the file structure from (dfd) and store it in
3464 	 * (indx).  (dfd) is effectively closed by this operation.
3465 	 */
3466 	switch (openerror) {
3467 	case ENODEV:
3468 		/*
3469 		 * Check that the mode the file is being opened for is a
3470 		 * subset of the mode of the existing descriptor.
3471 		 */
3472 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3473 			fdunused(fdp, indx);
3474 			FILEDESC_XUNLOCK(fdp);
3475 			return (EACCES);
3476 		}
3477 		if (!fhold(fp)) {
3478 			fdunused(fdp, indx);
3479 			FILEDESC_XUNLOCK(fdp);
3480 			return (EBADF);
3481 		}
3482 		newfde = &fdp->fd_ofiles[indx];
3483 		oldfde = &fdp->fd_ofiles[dfd];
3484 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3485 #ifdef CAPABILITIES
3486 		seqc_write_begin(&newfde->fde_seqc);
3487 #endif
3488 		memcpy(newfde, oldfde, fde_change_size);
3489 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3490 		    ioctls);
3491 #ifdef CAPABILITIES
3492 		seqc_write_end(&newfde->fde_seqc);
3493 #endif
3494 		break;
3495 	case ENXIO:
3496 		/*
3497 		 * Steal away the file pointer from dfd and stuff it into indx.
3498 		 */
3499 		newfde = &fdp->fd_ofiles[indx];
3500 		oldfde = &fdp->fd_ofiles[dfd];
3501 #ifdef CAPABILITIES
3502 		seqc_write_begin(&newfde->fde_seqc);
3503 #endif
3504 		memcpy(newfde, oldfde, fde_change_size);
3505 		oldfde->fde_file = NULL;
3506 		fdunused(fdp, dfd);
3507 #ifdef CAPABILITIES
3508 		seqc_write_end(&newfde->fde_seqc);
3509 #endif
3510 		break;
3511 	}
3512 	FILEDESC_XUNLOCK(fdp);
3513 	*indxp = indx;
3514 	return (0);
3515 }
3516 
3517 /*
3518  * This sysctl determines if we will allow a process to chroot(2) if it
3519  * has a directory open:
3520  *	0: disallowed for all processes.
3521  *	1: allowed for processes that were not already chroot(2)'ed.
3522  *	2: allowed for all processes.
3523  */
3524 
3525 static int chroot_allow_open_directories = 1;
3526 
3527 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3528     &chroot_allow_open_directories, 0,
3529     "Allow a process to chroot(2) if it has a directory open");
3530 
3531 /*
3532  * Helper function for raised chroot(2) security function:  Refuse if
3533  * any filedescriptors are open directories.
3534  */
3535 static int
3536 chroot_refuse_vdir_fds(struct filedesc *fdp)
3537 {
3538 	struct vnode *vp;
3539 	struct file *fp;
3540 	int fd, lastfile;
3541 
3542 	FILEDESC_LOCK_ASSERT(fdp);
3543 
3544 	lastfile = fdlastfile(fdp);
3545 	for (fd = 0; fd <= lastfile; fd++) {
3546 		fp = fget_locked(fdp, fd);
3547 		if (fp == NULL)
3548 			continue;
3549 		if (fp->f_type == DTYPE_VNODE) {
3550 			vp = fp->f_vnode;
3551 			if (vp->v_type == VDIR)
3552 				return (EPERM);
3553 		}
3554 	}
3555 	return (0);
3556 }
3557 
3558 static void
3559 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3560 {
3561 
3562 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3563 		vrefact(oldpwd->pwd_cdir);
3564 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3565 	}
3566 
3567 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3568 		vrefact(oldpwd->pwd_rdir);
3569 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3570 	}
3571 
3572 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3573 		vrefact(oldpwd->pwd_jdir);
3574 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3575 	}
3576 }
3577 
3578 struct pwd *
3579 pwd_hold_pwddesc(struct pwddesc *pdp)
3580 {
3581 	struct pwd *pwd;
3582 
3583 	PWDDESC_ASSERT_XLOCKED(pdp);
3584 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3585 	if (pwd != NULL)
3586 		refcount_acquire(&pwd->pwd_refcount);
3587 	return (pwd);
3588 }
3589 
3590 bool
3591 pwd_hold_smr(struct pwd *pwd)
3592 {
3593 
3594 	MPASS(pwd != NULL);
3595 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3596 		return (true);
3597 	}
3598 	return (false);
3599 }
3600 
3601 struct pwd *
3602 pwd_hold(struct thread *td)
3603 {
3604 	struct pwddesc *pdp;
3605 	struct pwd *pwd;
3606 
3607 	pdp = td->td_proc->p_pd;
3608 
3609 	vfs_smr_enter();
3610 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3611 	if (pwd_hold_smr(pwd)) {
3612 		vfs_smr_exit();
3613 		return (pwd);
3614 	}
3615 	vfs_smr_exit();
3616 	PWDDESC_XLOCK(pdp);
3617 	pwd = pwd_hold_pwddesc(pdp);
3618 	MPASS(pwd != NULL);
3619 	PWDDESC_XUNLOCK(pdp);
3620 	return (pwd);
3621 }
3622 
3623 struct pwd *
3624 pwd_get_smr(void)
3625 {
3626 	struct pwd *pwd;
3627 
3628 	pwd = vfs_smr_entered_load(&curproc->p_pd->pd_pwd);
3629 	MPASS(pwd != NULL);
3630 	return (pwd);
3631 }
3632 
3633 static struct pwd *
3634 pwd_alloc(void)
3635 {
3636 	struct pwd *pwd;
3637 
3638 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3639 	bzero(pwd, sizeof(*pwd));
3640 	refcount_init(&pwd->pwd_refcount, 1);
3641 	return (pwd);
3642 }
3643 
3644 void
3645 pwd_drop(struct pwd *pwd)
3646 {
3647 
3648 	if (!refcount_release(&pwd->pwd_refcount))
3649 		return;
3650 
3651 	if (pwd->pwd_cdir != NULL)
3652 		vrele(pwd->pwd_cdir);
3653 	if (pwd->pwd_rdir != NULL)
3654 		vrele(pwd->pwd_rdir);
3655 	if (pwd->pwd_jdir != NULL)
3656 		vrele(pwd->pwd_jdir);
3657 	uma_zfree_smr(pwd_zone, pwd);
3658 }
3659 
3660 /*
3661 * Common routine for kern_chroot() and jail_attach().  The caller is
3662 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3663 * authorize this operation.
3664 */
3665 int
3666 pwd_chroot(struct thread *td, struct vnode *vp)
3667 {
3668 	struct pwddesc *pdp;
3669 	struct filedesc *fdp;
3670 	struct pwd *newpwd, *oldpwd;
3671 	int error;
3672 
3673 	fdp = td->td_proc->p_fd;
3674 	pdp = td->td_proc->p_pd;
3675 	newpwd = pwd_alloc();
3676 	FILEDESC_SLOCK(fdp);
3677 	PWDDESC_XLOCK(pdp);
3678 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3679 	if (chroot_allow_open_directories == 0 ||
3680 	    (chroot_allow_open_directories == 1 &&
3681 	    oldpwd->pwd_rdir != rootvnode)) {
3682 		error = chroot_refuse_vdir_fds(fdp);
3683 		FILEDESC_SUNLOCK(fdp);
3684 		if (error != 0) {
3685 			PWDDESC_XUNLOCK(pdp);
3686 			pwd_drop(newpwd);
3687 			return (error);
3688 		}
3689 	} else {
3690 		FILEDESC_SUNLOCK(fdp);
3691 	}
3692 
3693 	vrefact(vp);
3694 	newpwd->pwd_rdir = vp;
3695 	if (oldpwd->pwd_jdir == NULL) {
3696 		vrefact(vp);
3697 		newpwd->pwd_jdir = vp;
3698 	}
3699 	pwd_fill(oldpwd, newpwd);
3700 	pwd_set(pdp, newpwd);
3701 	PWDDESC_XUNLOCK(pdp);
3702 	pwd_drop(oldpwd);
3703 	return (0);
3704 }
3705 
3706 void
3707 pwd_chdir(struct thread *td, struct vnode *vp)
3708 {
3709 	struct pwddesc *pdp;
3710 	struct pwd *newpwd, *oldpwd;
3711 
3712 	VNPASS(vp->v_usecount > 0, vp);
3713 
3714 	newpwd = pwd_alloc();
3715 	pdp = td->td_proc->p_pd;
3716 	PWDDESC_XLOCK(pdp);
3717 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3718 	newpwd->pwd_cdir = vp;
3719 	pwd_fill(oldpwd, newpwd);
3720 	pwd_set(pdp, newpwd);
3721 	PWDDESC_XUNLOCK(pdp);
3722 	pwd_drop(oldpwd);
3723 }
3724 
3725 void
3726 pwd_ensure_dirs(void)
3727 {
3728 	struct pwddesc *pdp;
3729 	struct pwd *oldpwd, *newpwd;
3730 
3731 	pdp = curproc->p_pd;
3732 	PWDDESC_XLOCK(pdp);
3733 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3734 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3735 		PWDDESC_XUNLOCK(pdp);
3736 		return;
3737 	}
3738 	PWDDESC_XUNLOCK(pdp);
3739 
3740 	newpwd = pwd_alloc();
3741 	PWDDESC_XLOCK(pdp);
3742 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3743 	pwd_fill(oldpwd, newpwd);
3744 	if (newpwd->pwd_cdir == NULL) {
3745 		vrefact(rootvnode);
3746 		newpwd->pwd_cdir = rootvnode;
3747 	}
3748 	if (newpwd->pwd_rdir == NULL) {
3749 		vrefact(rootvnode);
3750 		newpwd->pwd_rdir = rootvnode;
3751 	}
3752 	pwd_set(pdp, newpwd);
3753 	PWDDESC_XUNLOCK(pdp);
3754 	pwd_drop(oldpwd);
3755 }
3756 
3757 void
3758 pwd_set_rootvnode(void)
3759 {
3760 	struct pwddesc *pdp;
3761 	struct pwd *oldpwd, *newpwd;
3762 
3763 	pdp = curproc->p_pd;
3764 
3765 	newpwd = pwd_alloc();
3766 	PWDDESC_XLOCK(pdp);
3767 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3768 	vrefact(rootvnode);
3769 	newpwd->pwd_cdir = rootvnode;
3770 	vrefact(rootvnode);
3771 	newpwd->pwd_rdir = rootvnode;
3772 	pwd_fill(oldpwd, newpwd);
3773 	pwd_set(pdp, newpwd);
3774 	PWDDESC_XUNLOCK(pdp);
3775 	pwd_drop(oldpwd);
3776 }
3777 
3778 /*
3779  * Scan all active processes and prisons to see if any of them have a current
3780  * or root directory of `olddp'. If so, replace them with the new mount point.
3781  */
3782 void
3783 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3784 {
3785 	struct pwddesc *pdp;
3786 	struct pwd *newpwd, *oldpwd;
3787 	struct prison *pr;
3788 	struct proc *p;
3789 	int nrele;
3790 
3791 	if (vrefcnt(olddp) == 1)
3792 		return;
3793 	nrele = 0;
3794 	newpwd = pwd_alloc();
3795 	sx_slock(&allproc_lock);
3796 	FOREACH_PROC_IN_SYSTEM(p) {
3797 		PROC_LOCK(p);
3798 		pdp = pdhold(p);
3799 		PROC_UNLOCK(p);
3800 		if (pdp == NULL)
3801 			continue;
3802 		PWDDESC_XLOCK(pdp);
3803 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3804 		if (oldpwd == NULL ||
3805 		    (oldpwd->pwd_cdir != olddp &&
3806 		    oldpwd->pwd_rdir != olddp &&
3807 		    oldpwd->pwd_jdir != olddp)) {
3808 			PWDDESC_XUNLOCK(pdp);
3809 			pddrop(pdp);
3810 			continue;
3811 		}
3812 		if (oldpwd->pwd_cdir == olddp) {
3813 			vrefact(newdp);
3814 			newpwd->pwd_cdir = newdp;
3815 		}
3816 		if (oldpwd->pwd_rdir == olddp) {
3817 			vrefact(newdp);
3818 			newpwd->pwd_rdir = newdp;
3819 		}
3820 		if (oldpwd->pwd_jdir == olddp) {
3821 			vrefact(newdp);
3822 			newpwd->pwd_jdir = newdp;
3823 		}
3824 		pwd_fill(oldpwd, newpwd);
3825 		pwd_set(pdp, newpwd);
3826 		PWDDESC_XUNLOCK(pdp);
3827 		pwd_drop(oldpwd);
3828 		pddrop(pdp);
3829 		newpwd = pwd_alloc();
3830 	}
3831 	sx_sunlock(&allproc_lock);
3832 	pwd_drop(newpwd);
3833 	if (rootvnode == olddp) {
3834 		vrefact(newdp);
3835 		rootvnode = newdp;
3836 		nrele++;
3837 	}
3838 	mtx_lock(&prison0.pr_mtx);
3839 	if (prison0.pr_root == olddp) {
3840 		vrefact(newdp);
3841 		prison0.pr_root = newdp;
3842 		nrele++;
3843 	}
3844 	mtx_unlock(&prison0.pr_mtx);
3845 	sx_slock(&allprison_lock);
3846 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3847 		mtx_lock(&pr->pr_mtx);
3848 		if (pr->pr_root == olddp) {
3849 			vrefact(newdp);
3850 			pr->pr_root = newdp;
3851 			nrele++;
3852 		}
3853 		mtx_unlock(&pr->pr_mtx);
3854 	}
3855 	sx_sunlock(&allprison_lock);
3856 	while (nrele--)
3857 		vrele(olddp);
3858 }
3859 
3860 struct filedesc_to_leader *
3861 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3862 {
3863 	struct filedesc_to_leader *fdtol;
3864 
3865 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3866 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3867 	fdtol->fdl_refcount = 1;
3868 	fdtol->fdl_holdcount = 0;
3869 	fdtol->fdl_wakeup = 0;
3870 	fdtol->fdl_leader = leader;
3871 	if (old != NULL) {
3872 		FILEDESC_XLOCK(fdp);
3873 		fdtol->fdl_next = old->fdl_next;
3874 		fdtol->fdl_prev = old;
3875 		old->fdl_next = fdtol;
3876 		fdtol->fdl_next->fdl_prev = fdtol;
3877 		FILEDESC_XUNLOCK(fdp);
3878 	} else {
3879 		fdtol->fdl_next = fdtol;
3880 		fdtol->fdl_prev = fdtol;
3881 	}
3882 	return (fdtol);
3883 }
3884 
3885 static int
3886 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3887 {
3888 	NDSLOTTYPE *map;
3889 	struct filedesc *fdp;
3890 	int count, off, minoff;
3891 
3892 	if (*(int *)arg1 != 0)
3893 		return (EINVAL);
3894 
3895 	fdp = curproc->p_fd;
3896 	count = 0;
3897 	FILEDESC_SLOCK(fdp);
3898 	map = fdp->fd_map;
3899 	off = NDSLOT(fdp->fd_nfiles - 1);
3900 	for (minoff = NDSLOT(0); off >= minoff; --off)
3901 		count += bitcountl(map[off]);
3902 	FILEDESC_SUNLOCK(fdp);
3903 
3904 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3905 }
3906 
3907 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3908     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3909     "Number of open file descriptors");
3910 
3911 /*
3912  * Get file structures globally.
3913  */
3914 static int
3915 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3916 {
3917 	struct xfile xf;
3918 	struct filedesc *fdp;
3919 	struct file *fp;
3920 	struct proc *p;
3921 	int error, n, lastfile;
3922 
3923 	error = sysctl_wire_old_buffer(req, 0);
3924 	if (error != 0)
3925 		return (error);
3926 	if (req->oldptr == NULL) {
3927 		n = 0;
3928 		sx_slock(&allproc_lock);
3929 		FOREACH_PROC_IN_SYSTEM(p) {
3930 			PROC_LOCK(p);
3931 			if (p->p_state == PRS_NEW) {
3932 				PROC_UNLOCK(p);
3933 				continue;
3934 			}
3935 			fdp = fdhold(p);
3936 			PROC_UNLOCK(p);
3937 			if (fdp == NULL)
3938 				continue;
3939 			/* overestimates sparse tables. */
3940 			n += fdp->fd_nfiles;
3941 			fddrop(fdp);
3942 		}
3943 		sx_sunlock(&allproc_lock);
3944 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3945 	}
3946 	error = 0;
3947 	bzero(&xf, sizeof(xf));
3948 	xf.xf_size = sizeof(xf);
3949 	sx_slock(&allproc_lock);
3950 	FOREACH_PROC_IN_SYSTEM(p) {
3951 		PROC_LOCK(p);
3952 		if (p->p_state == PRS_NEW) {
3953 			PROC_UNLOCK(p);
3954 			continue;
3955 		}
3956 		if (p_cansee(req->td, p) != 0) {
3957 			PROC_UNLOCK(p);
3958 			continue;
3959 		}
3960 		xf.xf_pid = p->p_pid;
3961 		xf.xf_uid = p->p_ucred->cr_uid;
3962 		fdp = fdhold(p);
3963 		PROC_UNLOCK(p);
3964 		if (fdp == NULL)
3965 			continue;
3966 		FILEDESC_SLOCK(fdp);
3967 		lastfile = fdlastfile(fdp);
3968 		for (n = 0; fdp->fd_refcnt > 0 && n <= lastfile; ++n) {
3969 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3970 				continue;
3971 			xf.xf_fd = n;
3972 			xf.xf_file = (uintptr_t)fp;
3973 			xf.xf_data = (uintptr_t)fp->f_data;
3974 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3975 			xf.xf_type = (uintptr_t)fp->f_type;
3976 			xf.xf_count = refcount_load(&fp->f_count);
3977 			xf.xf_msgcount = 0;
3978 			xf.xf_offset = foffset_get(fp);
3979 			xf.xf_flag = fp->f_flag;
3980 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3981 			if (error)
3982 				break;
3983 		}
3984 		FILEDESC_SUNLOCK(fdp);
3985 		fddrop(fdp);
3986 		if (error)
3987 			break;
3988 	}
3989 	sx_sunlock(&allproc_lock);
3990 	return (error);
3991 }
3992 
3993 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3994     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3995 
3996 #ifdef KINFO_FILE_SIZE
3997 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3998 #endif
3999 
4000 static int
4001 xlate_fflags(int fflags)
4002 {
4003 	static const struct {
4004 		int	fflag;
4005 		int	kf_fflag;
4006 	} fflags_table[] = {
4007 		{ FAPPEND, KF_FLAG_APPEND },
4008 		{ FASYNC, KF_FLAG_ASYNC },
4009 		{ FFSYNC, KF_FLAG_FSYNC },
4010 		{ FHASLOCK, KF_FLAG_HASLOCK },
4011 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4012 		{ FREAD, KF_FLAG_READ },
4013 		{ FWRITE, KF_FLAG_WRITE },
4014 		{ O_CREAT, KF_FLAG_CREAT },
4015 		{ O_DIRECT, KF_FLAG_DIRECT },
4016 		{ O_EXCL, KF_FLAG_EXCL },
4017 		{ O_EXEC, KF_FLAG_EXEC },
4018 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4019 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4020 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4021 		{ O_TRUNC, KF_FLAG_TRUNC }
4022 	};
4023 	unsigned int i;
4024 	int kflags;
4025 
4026 	kflags = 0;
4027 	for (i = 0; i < nitems(fflags_table); i++)
4028 		if (fflags & fflags_table[i].fflag)
4029 			kflags |=  fflags_table[i].kf_fflag;
4030 	return (kflags);
4031 }
4032 
4033 /* Trim unused data from kf_path by truncating the structure size. */
4034 void
4035 pack_kinfo(struct kinfo_file *kif)
4036 {
4037 
4038 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4039 	    strlen(kif->kf_path) + 1;
4040 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4041 }
4042 
4043 static void
4044 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4045     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4046 {
4047 	int error;
4048 
4049 	bzero(kif, sizeof(*kif));
4050 
4051 	/* Set a default type to allow for empty fill_kinfo() methods. */
4052 	kif->kf_type = KF_TYPE_UNKNOWN;
4053 	kif->kf_flags = xlate_fflags(fp->f_flag);
4054 	if (rightsp != NULL)
4055 		kif->kf_cap_rights = *rightsp;
4056 	else
4057 		cap_rights_init_zero(&kif->kf_cap_rights);
4058 	kif->kf_fd = fd;
4059 	kif->kf_ref_count = refcount_load(&fp->f_count);
4060 	kif->kf_offset = foffset_get(fp);
4061 
4062 	/*
4063 	 * This may drop the filedesc lock, so the 'fp' cannot be
4064 	 * accessed after this call.
4065 	 */
4066 	error = fo_fill_kinfo(fp, kif, fdp);
4067 	if (error == 0)
4068 		kif->kf_status |= KF_ATTR_VALID;
4069 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4070 		pack_kinfo(kif);
4071 	else
4072 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4073 }
4074 
4075 static void
4076 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4077     struct kinfo_file *kif, int flags)
4078 {
4079 	int error;
4080 
4081 	bzero(kif, sizeof(*kif));
4082 
4083 	kif->kf_type = KF_TYPE_VNODE;
4084 	error = vn_fill_kinfo_vnode(vp, kif);
4085 	if (error == 0)
4086 		kif->kf_status |= KF_ATTR_VALID;
4087 	kif->kf_flags = xlate_fflags(fflags);
4088 	cap_rights_init_zero(&kif->kf_cap_rights);
4089 	kif->kf_fd = fd;
4090 	kif->kf_ref_count = -1;
4091 	kif->kf_offset = -1;
4092 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4093 		pack_kinfo(kif);
4094 	else
4095 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4096 	vrele(vp);
4097 }
4098 
4099 struct export_fd_buf {
4100 	struct filedesc		*fdp;
4101 	struct pwddesc	*pdp;
4102 	struct sbuf 		*sb;
4103 	ssize_t			remainder;
4104 	struct kinfo_file	kif;
4105 	int			flags;
4106 };
4107 
4108 static int
4109 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4110 {
4111 	struct kinfo_file *kif;
4112 
4113 	kif = &efbuf->kif;
4114 	if (efbuf->remainder != -1) {
4115 		if (efbuf->remainder < kif->kf_structsize) {
4116 			/* Terminate export. */
4117 			efbuf->remainder = 0;
4118 			return (0);
4119 		}
4120 		efbuf->remainder -= kif->kf_structsize;
4121 	}
4122 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4123 }
4124 
4125 static int
4126 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4127     struct export_fd_buf *efbuf)
4128 {
4129 	int error;
4130 
4131 	if (efbuf->remainder == 0)
4132 		return (0);
4133 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4134 	    efbuf->flags);
4135 	FILEDESC_SUNLOCK(efbuf->fdp);
4136 	error = export_kinfo_to_sb(efbuf);
4137 	FILEDESC_SLOCK(efbuf->fdp);
4138 	return (error);
4139 }
4140 
4141 static int
4142 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4143     struct export_fd_buf *efbuf)
4144 {
4145 	int error;
4146 
4147 	if (efbuf->remainder == 0)
4148 		return (0);
4149 	if (efbuf->pdp != NULL)
4150 		PWDDESC_XUNLOCK(efbuf->pdp);
4151 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4152 	error = export_kinfo_to_sb(efbuf);
4153 	if (efbuf->pdp != NULL)
4154 		PWDDESC_XLOCK(efbuf->pdp);
4155 	return (error);
4156 }
4157 
4158 /*
4159  * Store a process file descriptor information to sbuf.
4160  *
4161  * Takes a locked proc as argument, and returns with the proc unlocked.
4162  */
4163 int
4164 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4165     int flags)
4166 {
4167 	struct file *fp;
4168 	struct filedesc *fdp;
4169 	struct pwddesc *pdp;
4170 	struct export_fd_buf *efbuf;
4171 	struct vnode *cttyvp, *textvp, *tracevp;
4172 	struct pwd *pwd;
4173 	int error, i, lastfile;
4174 	cap_rights_t rights;
4175 
4176 	PROC_LOCK_ASSERT(p, MA_OWNED);
4177 
4178 	/* ktrace vnode */
4179 	tracevp = p->p_tracevp;
4180 	if (tracevp != NULL)
4181 		vrefact(tracevp);
4182 	/* text vnode */
4183 	textvp = p->p_textvp;
4184 	if (textvp != NULL)
4185 		vrefact(textvp);
4186 	/* Controlling tty. */
4187 	cttyvp = NULL;
4188 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4189 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4190 		if (cttyvp != NULL)
4191 			vrefact(cttyvp);
4192 	}
4193 	fdp = fdhold(p);
4194 	pdp = pdhold(p);
4195 	PROC_UNLOCK(p);
4196 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4197 	efbuf->fdp = NULL;
4198 	efbuf->pdp = NULL;
4199 	efbuf->sb = sb;
4200 	efbuf->remainder = maxlen;
4201 	efbuf->flags = flags;
4202 	if (tracevp != NULL)
4203 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4204 		    efbuf);
4205 	if (textvp != NULL)
4206 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4207 	if (cttyvp != NULL)
4208 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4209 		    efbuf);
4210 	error = 0;
4211 	if (pdp == NULL || fdp == NULL)
4212 		goto fail;
4213 	efbuf->fdp = fdp;
4214 	efbuf->pdp = pdp;
4215 	PWDDESC_XLOCK(pdp);
4216 	pwd = pwd_hold_pwddesc(pdp);
4217 	if (pwd != NULL) {
4218 		/* working directory */
4219 		if (pwd->pwd_cdir != NULL) {
4220 			vrefact(pwd->pwd_cdir);
4221 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4222 		}
4223 		/* root directory */
4224 		if (pwd->pwd_rdir != NULL) {
4225 			vrefact(pwd->pwd_rdir);
4226 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4227 		}
4228 		/* jail directory */
4229 		if (pwd->pwd_jdir != NULL) {
4230 			vrefact(pwd->pwd_jdir);
4231 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4232 		}
4233 	}
4234 	PWDDESC_XUNLOCK(pdp);
4235 	if (pwd != NULL)
4236 		pwd_drop(pwd);
4237 	FILEDESC_SLOCK(fdp);
4238 	lastfile = fdlastfile(fdp);
4239 	for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4240 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4241 			continue;
4242 #ifdef CAPABILITIES
4243 		rights = *cap_rights(fdp, i);
4244 #else /* !CAPABILITIES */
4245 		rights = cap_no_rights;
4246 #endif
4247 		/*
4248 		 * Create sysctl entry.  It is OK to drop the filedesc
4249 		 * lock inside of export_file_to_sb() as we will
4250 		 * re-validate and re-evaluate its properties when the
4251 		 * loop continues.
4252 		 */
4253 		error = export_file_to_sb(fp, i, &rights, efbuf);
4254 		if (error != 0 || efbuf->remainder == 0)
4255 			break;
4256 	}
4257 	FILEDESC_SUNLOCK(fdp);
4258 fail:
4259 	if (fdp != NULL)
4260 		fddrop(fdp);
4261 	if (pdp != NULL)
4262 		pddrop(pdp);
4263 	free(efbuf, M_TEMP);
4264 	return (error);
4265 }
4266 
4267 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4268 
4269 /*
4270  * Get per-process file descriptors for use by procstat(1), et al.
4271  */
4272 static int
4273 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4274 {
4275 	struct sbuf sb;
4276 	struct proc *p;
4277 	ssize_t maxlen;
4278 	int error, error2, *name;
4279 
4280 	name = (int *)arg1;
4281 
4282 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4283 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4284 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4285 	if (error != 0) {
4286 		sbuf_delete(&sb);
4287 		return (error);
4288 	}
4289 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4290 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4291 	    KERN_FILEDESC_PACK_KINFO);
4292 	error2 = sbuf_finish(&sb);
4293 	sbuf_delete(&sb);
4294 	return (error != 0 ? error : error2);
4295 }
4296 
4297 #ifdef COMPAT_FREEBSD7
4298 #ifdef KINFO_OFILE_SIZE
4299 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4300 #endif
4301 
4302 static void
4303 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4304 {
4305 
4306 	okif->kf_structsize = sizeof(*okif);
4307 	okif->kf_type = kif->kf_type;
4308 	okif->kf_fd = kif->kf_fd;
4309 	okif->kf_ref_count = kif->kf_ref_count;
4310 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4311 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4312 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4313 	okif->kf_offset = kif->kf_offset;
4314 	if (kif->kf_type == KF_TYPE_VNODE)
4315 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4316 	else
4317 		okif->kf_vnode_type = KF_VTYPE_VNON;
4318 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4319 	if (kif->kf_type == KF_TYPE_SOCKET) {
4320 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4321 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4322 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4323 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4324 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4325 	} else {
4326 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4327 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4328 	}
4329 }
4330 
4331 static int
4332 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4333     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4334 {
4335 	int error;
4336 
4337 	vrefact(vp);
4338 	PWDDESC_XUNLOCK(pdp);
4339 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4340 	kinfo_to_okinfo(kif, okif);
4341 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4342 	PWDDESC_XLOCK(pdp);
4343 	return (error);
4344 }
4345 
4346 /*
4347  * Get per-process file descriptors for use by procstat(1), et al.
4348  */
4349 static int
4350 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4351 {
4352 	struct kinfo_ofile *okif;
4353 	struct kinfo_file *kif;
4354 	struct filedesc *fdp;
4355 	struct pwddesc *pdp;
4356 	struct pwd *pwd;
4357 	int error, i, lastfile, *name;
4358 	struct file *fp;
4359 	struct proc *p;
4360 
4361 	name = (int *)arg1;
4362 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4363 	if (error != 0)
4364 		return (error);
4365 	fdp = fdhold(p);
4366 	if (fdp != NULL)
4367 		pdp = pdhold(p);
4368 	PROC_UNLOCK(p);
4369 	if (fdp == NULL || pdp == NULL) {
4370 		if (fdp != NULL)
4371 			fddrop(fdp);
4372 		return (ENOENT);
4373 	}
4374 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4375 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4376 	PWDDESC_XLOCK(pdp);
4377 	pwd = pwd_hold_pwddesc(pdp);
4378 	if (pwd != NULL) {
4379 		if (pwd->pwd_cdir != NULL)
4380 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4381 			    okif, pdp, req);
4382 		if (pwd->pwd_rdir != NULL)
4383 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4384 			    okif, pdp, req);
4385 		if (pwd->pwd_jdir != NULL)
4386 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4387 			    okif, pdp, req);
4388 	}
4389 	PWDDESC_XUNLOCK(pdp);
4390 	if (pwd != NULL)
4391 		pwd_drop(pwd);
4392 	FILEDESC_SLOCK(fdp);
4393 	lastfile = fdlastfile(fdp);
4394 	for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4395 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4396 			continue;
4397 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4398 		    KERN_FILEDESC_PACK_KINFO);
4399 		FILEDESC_SUNLOCK(fdp);
4400 		kinfo_to_okinfo(kif, okif);
4401 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4402 		FILEDESC_SLOCK(fdp);
4403 		if (error)
4404 			break;
4405 	}
4406 	FILEDESC_SUNLOCK(fdp);
4407 	fddrop(fdp);
4408 	pddrop(pdp);
4409 	free(kif, M_TEMP);
4410 	free(okif, M_TEMP);
4411 	return (0);
4412 }
4413 
4414 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4415     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4416     "Process ofiledesc entries");
4417 #endif	/* COMPAT_FREEBSD7 */
4418 
4419 int
4420 vntype_to_kinfo(int vtype)
4421 {
4422 	struct {
4423 		int	vtype;
4424 		int	kf_vtype;
4425 	} vtypes_table[] = {
4426 		{ VBAD, KF_VTYPE_VBAD },
4427 		{ VBLK, KF_VTYPE_VBLK },
4428 		{ VCHR, KF_VTYPE_VCHR },
4429 		{ VDIR, KF_VTYPE_VDIR },
4430 		{ VFIFO, KF_VTYPE_VFIFO },
4431 		{ VLNK, KF_VTYPE_VLNK },
4432 		{ VNON, KF_VTYPE_VNON },
4433 		{ VREG, KF_VTYPE_VREG },
4434 		{ VSOCK, KF_VTYPE_VSOCK }
4435 	};
4436 	unsigned int i;
4437 
4438 	/*
4439 	 * Perform vtype translation.
4440 	 */
4441 	for (i = 0; i < nitems(vtypes_table); i++)
4442 		if (vtypes_table[i].vtype == vtype)
4443 			return (vtypes_table[i].kf_vtype);
4444 
4445 	return (KF_VTYPE_UNKNOWN);
4446 }
4447 
4448 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4449     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4450     "Process filedesc entries");
4451 
4452 /*
4453  * Store a process current working directory information to sbuf.
4454  *
4455  * Takes a locked proc as argument, and returns with the proc unlocked.
4456  */
4457 int
4458 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4459 {
4460 	struct pwddesc *pdp;
4461 	struct pwd *pwd;
4462 	struct export_fd_buf *efbuf;
4463 	struct vnode *cdir;
4464 	int error;
4465 
4466 	PROC_LOCK_ASSERT(p, MA_OWNED);
4467 
4468 	pdp = pdhold(p);
4469 	PROC_UNLOCK(p);
4470 	if (pdp == NULL)
4471 		return (EINVAL);
4472 
4473 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4474 	efbuf->pdp = pdp;
4475 	efbuf->sb = sb;
4476 	efbuf->remainder = maxlen;
4477 
4478 	PWDDESC_XLOCK(pdp);
4479 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4480 	cdir = pwd->pwd_cdir;
4481 	if (cdir == NULL) {
4482 		error = EINVAL;
4483 	} else {
4484 		vrefact(cdir);
4485 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4486 	}
4487 	PWDDESC_XUNLOCK(pdp);
4488 	pddrop(pdp);
4489 	free(efbuf, M_TEMP);
4490 	return (error);
4491 }
4492 
4493 /*
4494  * Get per-process current working directory.
4495  */
4496 static int
4497 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4498 {
4499 	struct sbuf sb;
4500 	struct proc *p;
4501 	ssize_t maxlen;
4502 	int error, error2, *name;
4503 
4504 	name = (int *)arg1;
4505 
4506 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4507 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4508 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4509 	if (error != 0) {
4510 		sbuf_delete(&sb);
4511 		return (error);
4512 	}
4513 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4514 	error = kern_proc_cwd_out(p, &sb, maxlen);
4515 	error2 = sbuf_finish(&sb);
4516 	sbuf_delete(&sb);
4517 	return (error != 0 ? error : error2);
4518 }
4519 
4520 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4521     sysctl_kern_proc_cwd, "Process current working directory");
4522 
4523 #ifdef DDB
4524 /*
4525  * For the purposes of debugging, generate a human-readable string for the
4526  * file type.
4527  */
4528 static const char *
4529 file_type_to_name(short type)
4530 {
4531 
4532 	switch (type) {
4533 	case 0:
4534 		return ("zero");
4535 	case DTYPE_VNODE:
4536 		return ("vnode");
4537 	case DTYPE_SOCKET:
4538 		return ("socket");
4539 	case DTYPE_PIPE:
4540 		return ("pipe");
4541 	case DTYPE_FIFO:
4542 		return ("fifo");
4543 	case DTYPE_KQUEUE:
4544 		return ("kqueue");
4545 	case DTYPE_CRYPTO:
4546 		return ("crypto");
4547 	case DTYPE_MQUEUE:
4548 		return ("mqueue");
4549 	case DTYPE_SHM:
4550 		return ("shm");
4551 	case DTYPE_SEM:
4552 		return ("ksem");
4553 	case DTYPE_PTS:
4554 		return ("pts");
4555 	case DTYPE_DEV:
4556 		return ("dev");
4557 	case DTYPE_PROCDESC:
4558 		return ("proc");
4559 	case DTYPE_LINUXEFD:
4560 		return ("levent");
4561 	case DTYPE_LINUXTFD:
4562 		return ("ltimer");
4563 	default:
4564 		return ("unkn");
4565 	}
4566 }
4567 
4568 /*
4569  * For the purposes of debugging, identify a process (if any, perhaps one of
4570  * many) that references the passed file in its file descriptor array. Return
4571  * NULL if none.
4572  */
4573 static struct proc *
4574 file_to_first_proc(struct file *fp)
4575 {
4576 	struct filedesc *fdp;
4577 	struct proc *p;
4578 	int n;
4579 
4580 	FOREACH_PROC_IN_SYSTEM(p) {
4581 		if (p->p_state == PRS_NEW)
4582 			continue;
4583 		fdp = p->p_fd;
4584 		if (fdp == NULL)
4585 			continue;
4586 		for (n = 0; n < fdp->fd_nfiles; n++) {
4587 			if (fp == fdp->fd_ofiles[n].fde_file)
4588 				return (p);
4589 		}
4590 	}
4591 	return (NULL);
4592 }
4593 
4594 static void
4595 db_print_file(struct file *fp, int header)
4596 {
4597 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4598 	struct proc *p;
4599 
4600 	if (header)
4601 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4602 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4603 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4604 		    "FCmd");
4605 	p = file_to_first_proc(fp);
4606 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4607 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4608 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4609 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4610 
4611 #undef XPTRWIDTH
4612 }
4613 
4614 DB_SHOW_COMMAND(file, db_show_file)
4615 {
4616 	struct file *fp;
4617 
4618 	if (!have_addr) {
4619 		db_printf("usage: show file <addr>\n");
4620 		return;
4621 	}
4622 	fp = (struct file *)addr;
4623 	db_print_file(fp, 1);
4624 }
4625 
4626 DB_SHOW_COMMAND(files, db_show_files)
4627 {
4628 	struct filedesc *fdp;
4629 	struct file *fp;
4630 	struct proc *p;
4631 	int header;
4632 	int n;
4633 
4634 	header = 1;
4635 	FOREACH_PROC_IN_SYSTEM(p) {
4636 		if (p->p_state == PRS_NEW)
4637 			continue;
4638 		if ((fdp = p->p_fd) == NULL)
4639 			continue;
4640 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4641 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4642 				continue;
4643 			db_print_file(fp, header);
4644 			header = 0;
4645 		}
4646 	}
4647 }
4648 #endif
4649 
4650 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4651     &maxfilesperproc, 0, "Maximum files allowed open per process");
4652 
4653 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4654     &maxfiles, 0, "Maximum number of files");
4655 
4656 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4657     &openfiles, 0, "System-wide number of open files");
4658 
4659 /* ARGSUSED*/
4660 static void
4661 filelistinit(void *dummy)
4662 {
4663 
4664 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4665 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4666 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4667 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4668 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4669 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4670 	/*
4671 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4672 	 * the vnode zone.
4673 	 */
4674 	vfs_smr = uma_zone_get_smr(pwd_zone);
4675 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4676 }
4677 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4678 
4679 /*-------------------------------------------------------------------*/
4680 
4681 static int
4682 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4683     int flags, struct thread *td)
4684 {
4685 
4686 	return (EBADF);
4687 }
4688 
4689 static int
4690 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4691     struct thread *td)
4692 {
4693 
4694 	return (EINVAL);
4695 }
4696 
4697 static int
4698 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4699     struct thread *td)
4700 {
4701 
4702 	return (EBADF);
4703 }
4704 
4705 static int
4706 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4707     struct thread *td)
4708 {
4709 
4710 	return (0);
4711 }
4712 
4713 static int
4714 badfo_kqfilter(struct file *fp, struct knote *kn)
4715 {
4716 
4717 	return (EBADF);
4718 }
4719 
4720 static int
4721 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4722     struct thread *td)
4723 {
4724 
4725 	return (EBADF);
4726 }
4727 
4728 static int
4729 badfo_close(struct file *fp, struct thread *td)
4730 {
4731 
4732 	return (0);
4733 }
4734 
4735 static int
4736 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4737     struct thread *td)
4738 {
4739 
4740 	return (EBADF);
4741 }
4742 
4743 static int
4744 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4745     struct thread *td)
4746 {
4747 
4748 	return (EBADF);
4749 }
4750 
4751 static int
4752 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4753     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4754     struct thread *td)
4755 {
4756 
4757 	return (EBADF);
4758 }
4759 
4760 static int
4761 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4762 {
4763 
4764 	return (0);
4765 }
4766 
4767 struct fileops badfileops = {
4768 	.fo_read = badfo_readwrite,
4769 	.fo_write = badfo_readwrite,
4770 	.fo_truncate = badfo_truncate,
4771 	.fo_ioctl = badfo_ioctl,
4772 	.fo_poll = badfo_poll,
4773 	.fo_kqfilter = badfo_kqfilter,
4774 	.fo_stat = badfo_stat,
4775 	.fo_close = badfo_close,
4776 	.fo_chmod = badfo_chmod,
4777 	.fo_chown = badfo_chown,
4778 	.fo_sendfile = badfo_sendfile,
4779 	.fo_fill_kinfo = badfo_fill_kinfo,
4780 };
4781 
4782 int
4783 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4784     int flags, struct thread *td)
4785 {
4786 
4787 	return (EOPNOTSUPP);
4788 }
4789 
4790 int
4791 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4792     struct thread *td)
4793 {
4794 
4795 	return (EINVAL);
4796 }
4797 
4798 int
4799 invfo_ioctl(struct file *fp, u_long com, void *data,
4800     struct ucred *active_cred, struct thread *td)
4801 {
4802 
4803 	return (ENOTTY);
4804 }
4805 
4806 int
4807 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4808     struct thread *td)
4809 {
4810 
4811 	return (poll_no_poll(events));
4812 }
4813 
4814 int
4815 invfo_kqfilter(struct file *fp, struct knote *kn)
4816 {
4817 
4818 	return (EINVAL);
4819 }
4820 
4821 int
4822 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4823     struct thread *td)
4824 {
4825 
4826 	return (EINVAL);
4827 }
4828 
4829 int
4830 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4831     struct thread *td)
4832 {
4833 
4834 	return (EINVAL);
4835 }
4836 
4837 int
4838 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4839     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4840     struct thread *td)
4841 {
4842 
4843 	return (EINVAL);
4844 }
4845 
4846 /*-------------------------------------------------------------------*/
4847 
4848 /*
4849  * File Descriptor pseudo-device driver (/dev/fd/).
4850  *
4851  * Opening minor device N dup()s the file (if any) connected to file
4852  * descriptor N belonging to the calling process.  Note that this driver
4853  * consists of only the ``open()'' routine, because all subsequent
4854  * references to this file will be direct to the other driver.
4855  *
4856  * XXX: we could give this one a cloning event handler if necessary.
4857  */
4858 
4859 /* ARGSUSED */
4860 static int
4861 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4862 {
4863 
4864 	/*
4865 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4866 	 * the file descriptor being sought for duplication. The error
4867 	 * return ensures that the vnode for this device will be released
4868 	 * by vn_open. Open will detect this special error and take the
4869 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4870 	 * will simply report the error.
4871 	 */
4872 	td->td_dupfd = dev2unit(dev);
4873 	return (ENODEV);
4874 }
4875 
4876 static struct cdevsw fildesc_cdevsw = {
4877 	.d_version =	D_VERSION,
4878 	.d_open =	fdopen,
4879 	.d_name =	"FD",
4880 };
4881 
4882 static void
4883 fildesc_drvinit(void *unused)
4884 {
4885 	struct cdev *dev;
4886 
4887 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4888 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4889 	make_dev_alias(dev, "stdin");
4890 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4891 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4892 	make_dev_alias(dev, "stdout");
4893 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4894 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4895 	make_dev_alias(dev, "stderr");
4896 }
4897 
4898 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4899