xref: /freebsd/sys/kern/kern_descrip.c (revision 4b50c451720d8b427757a6da1dd2bb4c52cd9e35)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/stat.h>
73 #include <sys/sx.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/unistd.h>
78 #include <sys/user.h>
79 #include <sys/vnode.h>
80 #ifdef KTRACE
81 #include <sys/ktrace.h>
82 #endif
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
95     "file desc to leader structures");
96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
98 
99 MALLOC_DECLARE(M_FADVISE);
100 
101 static __read_mostly uma_zone_t file_zone;
102 static __read_mostly uma_zone_t filedesc0_zone;
103 
104 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
105 		    struct thread *td, int holdleaders);
106 static int	fd_first_free(struct filedesc *fdp, int low, int size);
107 static int	fd_last_used(struct filedesc *fdp, int size);
108 static void	fdgrowtable(struct filedesc *fdp, int nfd);
109 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void	fdunused(struct filedesc *fdp, int fd);
111 static void	fdused(struct filedesc *fdp, int fd);
112 static int	getmaxfd(struct thread *td);
113 static u_long	*filecaps_copy_prep(const struct filecaps *src);
114 static void	filecaps_copy_finish(const struct filecaps *src,
115 		    struct filecaps *dst, u_long *ioctls);
116 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
117 static void	filecaps_free_finish(u_long *ioctls);
118 
119 /*
120  * Each process has:
121  *
122  * - An array of open file descriptors (fd_ofiles)
123  * - An array of file flags (fd_ofileflags)
124  * - A bitmap recording which descriptors are in use (fd_map)
125  *
126  * A process starts out with NDFILE descriptors.  The value of NDFILE has
127  * been selected based the historical limit of 20 open files, and an
128  * assumption that the majority of processes, especially short-lived
129  * processes like shells, will never need more.
130  *
131  * If this initial allocation is exhausted, a larger descriptor table and
132  * map are allocated dynamically, and the pointers in the process's struct
133  * filedesc are updated to point to those.  This is repeated every time
134  * the process runs out of file descriptors (provided it hasn't hit its
135  * resource limit).
136  *
137  * Since threads may hold references to individual descriptor table
138  * entries, the tables are never freed.  Instead, they are placed on a
139  * linked list and freed only when the struct filedesc is released.
140  */
141 #define NDFILE		20
142 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
143 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x)	((x) / NDENTRIES)
145 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
147 
148 /*
149  * SLIST entry used to keep track of ofiles which must be reclaimed when
150  * the process exits.
151  */
152 struct freetable {
153 	struct fdescenttbl *ft_table;
154 	SLIST_ENTRY(freetable) ft_next;
155 };
156 
157 /*
158  * Initial allocation: a filedesc structure + the head of SLIST used to
159  * keep track of old ofiles + enough space for NDFILE descriptors.
160  */
161 
162 struct fdescenttbl0 {
163 	int	fdt_nfiles;
164 	struct	filedescent fdt_ofiles[NDFILE];
165 };
166 
167 struct filedesc0 {
168 	struct filedesc fd_fd;
169 	SLIST_HEAD(, freetable) fd_free;
170 	struct	fdescenttbl0 fd_dfiles;
171 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
172 };
173 
174 /*
175  * Descriptor management.
176  */
177 static int __exclusive_cache_line openfiles; /* actual number of open files */
178 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
180 
181 /*
182  * If low >= size, just return low. Otherwise find the first zero bit in the
183  * given bitmap, starting at low and not exceeding size - 1. Return size if
184  * not found.
185  */
186 static int
187 fd_first_free(struct filedesc *fdp, int low, int size)
188 {
189 	NDSLOTTYPE *map = fdp->fd_map;
190 	NDSLOTTYPE mask;
191 	int off, maxoff;
192 
193 	if (low >= size)
194 		return (low);
195 
196 	off = NDSLOT(low);
197 	if (low % NDENTRIES) {
198 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
199 		if ((mask &= ~map[off]) != 0UL)
200 			return (off * NDENTRIES + ffsl(mask) - 1);
201 		++off;
202 	}
203 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
204 		if (map[off] != ~0UL)
205 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
206 	return (size);
207 }
208 
209 /*
210  * Find the highest non-zero bit in the given bitmap, starting at 0 and
211  * not exceeding size - 1. Return -1 if not found.
212  */
213 static int
214 fd_last_used(struct filedesc *fdp, int size)
215 {
216 	NDSLOTTYPE *map = fdp->fd_map;
217 	NDSLOTTYPE mask;
218 	int off, minoff;
219 
220 	off = NDSLOT(size);
221 	if (size % NDENTRIES) {
222 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
223 		if ((mask &= map[off]) != 0)
224 			return (off * NDENTRIES + flsl(mask) - 1);
225 		--off;
226 	}
227 	for (minoff = NDSLOT(0); off >= minoff; --off)
228 		if (map[off] != 0)
229 			return (off * NDENTRIES + flsl(map[off]) - 1);
230 	return (-1);
231 }
232 
233 static int
234 fdisused(struct filedesc *fdp, int fd)
235 {
236 
237 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
238 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
239 
240 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
241 }
242 
243 /*
244  * Mark a file descriptor as used.
245  */
246 static void
247 fdused_init(struct filedesc *fdp, int fd)
248 {
249 
250 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
251 
252 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
253 }
254 
255 static void
256 fdused(struct filedesc *fdp, int fd)
257 {
258 
259 	FILEDESC_XLOCK_ASSERT(fdp);
260 
261 	fdused_init(fdp, fd);
262 	if (fd > fdp->fd_lastfile)
263 		fdp->fd_lastfile = fd;
264 	if (fd == fdp->fd_freefile)
265 		fdp->fd_freefile++;
266 }
267 
268 /*
269  * Mark a file descriptor as unused.
270  */
271 static void
272 fdunused(struct filedesc *fdp, int fd)
273 {
274 
275 	FILEDESC_XLOCK_ASSERT(fdp);
276 
277 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
278 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
279 	    ("fd=%d is still in use", fd));
280 
281 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
282 	if (fd < fdp->fd_freefile)
283 		fdp->fd_freefile = fd;
284 	if (fd == fdp->fd_lastfile)
285 		fdp->fd_lastfile = fd_last_used(fdp, fd);
286 }
287 
288 /*
289  * Free a file descriptor.
290  *
291  * Avoid some work if fdp is about to be destroyed.
292  */
293 static inline void
294 fdefree_last(struct filedescent *fde)
295 {
296 
297 	filecaps_free(&fde->fde_caps);
298 }
299 
300 static inline void
301 fdfree(struct filedesc *fdp, int fd)
302 {
303 	struct filedescent *fde;
304 
305 	fde = &fdp->fd_ofiles[fd];
306 #ifdef CAPABILITIES
307 	seqc_write_begin(&fde->fde_seqc);
308 #endif
309 	fde->fde_file = NULL;
310 #ifdef CAPABILITIES
311 	seqc_write_end(&fde->fde_seqc);
312 #endif
313 	fdefree_last(fde);
314 	fdunused(fdp, fd);
315 }
316 
317 void
318 pwd_ensure_dirs(void)
319 {
320 	struct filedesc *fdp;
321 
322 	fdp = curproc->p_fd;
323 	FILEDESC_XLOCK(fdp);
324 	if (fdp->fd_cdir == NULL) {
325 		fdp->fd_cdir = rootvnode;
326 		vrefact(rootvnode);
327 	}
328 	if (fdp->fd_rdir == NULL) {
329 		fdp->fd_rdir = rootvnode;
330 		vrefact(rootvnode);
331 	}
332 	FILEDESC_XUNLOCK(fdp);
333 }
334 
335 /*
336  * System calls on descriptors.
337  */
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
340 	int	dummy;
341 };
342 #endif
343 /* ARGSUSED */
344 int
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
346 {
347 #ifdef	RACCT
348 	uint64_t lim;
349 #endif
350 
351 	td->td_retval[0] = getmaxfd(td);
352 #ifdef	RACCT
353 	PROC_LOCK(td->td_proc);
354 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
355 	PROC_UNLOCK(td->td_proc);
356 	if (lim < td->td_retval[0])
357 		td->td_retval[0] = lim;
358 #endif
359 	return (0);
360 }
361 
362 /*
363  * Duplicate a file descriptor to a particular value.
364  *
365  * Note: keep in mind that a potential race condition exists when closing
366  * descriptors from a shared descriptor table (via rfork).
367  */
368 #ifndef _SYS_SYSPROTO_H_
369 struct dup2_args {
370 	u_int	from;
371 	u_int	to;
372 };
373 #endif
374 /* ARGSUSED */
375 int
376 sys_dup2(struct thread *td, struct dup2_args *uap)
377 {
378 
379 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
380 }
381 
382 /*
383  * Duplicate a file descriptor.
384  */
385 #ifndef _SYS_SYSPROTO_H_
386 struct dup_args {
387 	u_int	fd;
388 };
389 #endif
390 /* ARGSUSED */
391 int
392 sys_dup(struct thread *td, struct dup_args *uap)
393 {
394 
395 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
396 }
397 
398 /*
399  * The file control system call.
400  */
401 #ifndef _SYS_SYSPROTO_H_
402 struct fcntl_args {
403 	int	fd;
404 	int	cmd;
405 	long	arg;
406 };
407 #endif
408 /* ARGSUSED */
409 int
410 sys_fcntl(struct thread *td, struct fcntl_args *uap)
411 {
412 
413 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
414 }
415 
416 int
417 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
418 {
419 	struct flock fl;
420 	struct __oflock ofl;
421 	intptr_t arg1;
422 	int error, newcmd;
423 
424 	error = 0;
425 	newcmd = cmd;
426 	switch (cmd) {
427 	case F_OGETLK:
428 	case F_OSETLK:
429 	case F_OSETLKW:
430 		/*
431 		 * Convert old flock structure to new.
432 		 */
433 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
434 		fl.l_start = ofl.l_start;
435 		fl.l_len = ofl.l_len;
436 		fl.l_pid = ofl.l_pid;
437 		fl.l_type = ofl.l_type;
438 		fl.l_whence = ofl.l_whence;
439 		fl.l_sysid = 0;
440 
441 		switch (cmd) {
442 		case F_OGETLK:
443 			newcmd = F_GETLK;
444 			break;
445 		case F_OSETLK:
446 			newcmd = F_SETLK;
447 			break;
448 		case F_OSETLKW:
449 			newcmd = F_SETLKW;
450 			break;
451 		}
452 		arg1 = (intptr_t)&fl;
453 		break;
454 	case F_GETLK:
455 	case F_SETLK:
456 	case F_SETLKW:
457 	case F_SETLK_REMOTE:
458 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
459 		arg1 = (intptr_t)&fl;
460 		break;
461 	default:
462 		arg1 = arg;
463 		break;
464 	}
465 	if (error)
466 		return (error);
467 	error = kern_fcntl(td, fd, newcmd, arg1);
468 	if (error)
469 		return (error);
470 	if (cmd == F_OGETLK) {
471 		ofl.l_start = fl.l_start;
472 		ofl.l_len = fl.l_len;
473 		ofl.l_pid = fl.l_pid;
474 		ofl.l_type = fl.l_type;
475 		ofl.l_whence = fl.l_whence;
476 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
477 	} else if (cmd == F_GETLK) {
478 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
479 	}
480 	return (error);
481 }
482 
483 int
484 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
485 {
486 	struct filedesc *fdp;
487 	struct flock *flp;
488 	struct file *fp, *fp2;
489 	struct filedescent *fde;
490 	struct proc *p;
491 	struct vnode *vp;
492 	int error, flg, seals, tmp;
493 	uint64_t bsize;
494 	off_t foffset;
495 
496 	error = 0;
497 	flg = F_POSIX;
498 	p = td->td_proc;
499 	fdp = p->p_fd;
500 
501 	AUDIT_ARG_FD(cmd);
502 	AUDIT_ARG_CMD(cmd);
503 	switch (cmd) {
504 	case F_DUPFD:
505 		tmp = arg;
506 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
507 		break;
508 
509 	case F_DUPFD_CLOEXEC:
510 		tmp = arg;
511 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
512 		break;
513 
514 	case F_DUP2FD:
515 		tmp = arg;
516 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
517 		break;
518 
519 	case F_DUP2FD_CLOEXEC:
520 		tmp = arg;
521 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
522 		break;
523 
524 	case F_GETFD:
525 		error = EBADF;
526 		FILEDESC_SLOCK(fdp);
527 		fde = fdeget_locked(fdp, fd);
528 		if (fde != NULL) {
529 			td->td_retval[0] =
530 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
531 			error = 0;
532 		}
533 		FILEDESC_SUNLOCK(fdp);
534 		break;
535 
536 	case F_SETFD:
537 		error = EBADF;
538 		FILEDESC_XLOCK(fdp);
539 		fde = fdeget_locked(fdp, fd);
540 		if (fde != NULL) {
541 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
542 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
543 			error = 0;
544 		}
545 		FILEDESC_XUNLOCK(fdp);
546 		break;
547 
548 	case F_GETFL:
549 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
550 		if (error != 0)
551 			break;
552 		td->td_retval[0] = OFLAGS(fp->f_flag);
553 		fdrop(fp, td);
554 		break;
555 
556 	case F_SETFL:
557 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
558 		if (error != 0)
559 			break;
560 		do {
561 			tmp = flg = fp->f_flag;
562 			tmp &= ~FCNTLFLAGS;
563 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
564 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
565 		tmp = fp->f_flag & FNONBLOCK;
566 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
567 		if (error != 0) {
568 			fdrop(fp, td);
569 			break;
570 		}
571 		tmp = fp->f_flag & FASYNC;
572 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
573 		if (error == 0) {
574 			fdrop(fp, td);
575 			break;
576 		}
577 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
578 		tmp = 0;
579 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
580 		fdrop(fp, td);
581 		break;
582 
583 	case F_GETOWN:
584 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
585 		if (error != 0)
586 			break;
587 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
588 		if (error == 0)
589 			td->td_retval[0] = tmp;
590 		fdrop(fp, td);
591 		break;
592 
593 	case F_SETOWN:
594 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
595 		if (error != 0)
596 			break;
597 		tmp = arg;
598 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
599 		fdrop(fp, td);
600 		break;
601 
602 	case F_SETLK_REMOTE:
603 		error = priv_check(td, PRIV_NFS_LOCKD);
604 		if (error != 0)
605 			return (error);
606 		flg = F_REMOTE;
607 		goto do_setlk;
608 
609 	case F_SETLKW:
610 		flg |= F_WAIT;
611 		/* FALLTHROUGH F_SETLK */
612 
613 	case F_SETLK:
614 	do_setlk:
615 		flp = (struct flock *)arg;
616 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
617 			error = EINVAL;
618 			break;
619 		}
620 
621 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
622 		if (error != 0)
623 			break;
624 		if (fp->f_type != DTYPE_VNODE) {
625 			error = EBADF;
626 			fdrop(fp, td);
627 			break;
628 		}
629 
630 		if (flp->l_whence == SEEK_CUR) {
631 			foffset = foffset_get(fp);
632 			if (foffset < 0 ||
633 			    (flp->l_start > 0 &&
634 			     foffset > OFF_MAX - flp->l_start)) {
635 				error = EOVERFLOW;
636 				fdrop(fp, td);
637 				break;
638 			}
639 			flp->l_start += foffset;
640 		}
641 
642 		vp = fp->f_vnode;
643 		switch (flp->l_type) {
644 		case F_RDLCK:
645 			if ((fp->f_flag & FREAD) == 0) {
646 				error = EBADF;
647 				break;
648 			}
649 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
650 				PROC_LOCK(p->p_leader);
651 				p->p_leader->p_flag |= P_ADVLOCK;
652 				PROC_UNLOCK(p->p_leader);
653 			}
654 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
655 			    flp, flg);
656 			break;
657 		case F_WRLCK:
658 			if ((fp->f_flag & FWRITE) == 0) {
659 				error = EBADF;
660 				break;
661 			}
662 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
663 				PROC_LOCK(p->p_leader);
664 				p->p_leader->p_flag |= P_ADVLOCK;
665 				PROC_UNLOCK(p->p_leader);
666 			}
667 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
668 			    flp, flg);
669 			break;
670 		case F_UNLCK:
671 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
672 			    flp, flg);
673 			break;
674 		case F_UNLCKSYS:
675 			if (flg != F_REMOTE) {
676 				error = EINVAL;
677 				break;
678 			}
679 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
680 			    F_UNLCKSYS, flp, flg);
681 			break;
682 		default:
683 			error = EINVAL;
684 			break;
685 		}
686 		if (error != 0 || flp->l_type == F_UNLCK ||
687 		    flp->l_type == F_UNLCKSYS) {
688 			fdrop(fp, td);
689 			break;
690 		}
691 
692 		/*
693 		 * Check for a race with close.
694 		 *
695 		 * The vnode is now advisory locked (or unlocked, but this case
696 		 * is not really important) as the caller requested.
697 		 * We had to drop the filedesc lock, so we need to recheck if
698 		 * the descriptor is still valid, because if it was closed
699 		 * in the meantime we need to remove advisory lock from the
700 		 * vnode - close on any descriptor leading to an advisory
701 		 * locked vnode, removes that lock.
702 		 * We will return 0 on purpose in that case, as the result of
703 		 * successful advisory lock might have been externally visible
704 		 * already. This is fine - effectively we pretend to the caller
705 		 * that the closing thread was a bit slower and that the
706 		 * advisory lock succeeded before the close.
707 		 */
708 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
709 		if (error != 0) {
710 			fdrop(fp, td);
711 			break;
712 		}
713 		if (fp != fp2) {
714 			flp->l_whence = SEEK_SET;
715 			flp->l_start = 0;
716 			flp->l_len = 0;
717 			flp->l_type = F_UNLCK;
718 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
719 			    F_UNLCK, flp, F_POSIX);
720 		}
721 		fdrop(fp, td);
722 		fdrop(fp2, td);
723 		break;
724 
725 	case F_GETLK:
726 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
727 		if (error != 0)
728 			break;
729 		if (fp->f_type != DTYPE_VNODE) {
730 			error = EBADF;
731 			fdrop(fp, td);
732 			break;
733 		}
734 		flp = (struct flock *)arg;
735 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
736 		    flp->l_type != F_UNLCK) {
737 			error = EINVAL;
738 			fdrop(fp, td);
739 			break;
740 		}
741 		if (flp->l_whence == SEEK_CUR) {
742 			foffset = foffset_get(fp);
743 			if ((flp->l_start > 0 &&
744 			    foffset > OFF_MAX - flp->l_start) ||
745 			    (flp->l_start < 0 &&
746 			    foffset < OFF_MIN - flp->l_start)) {
747 				error = EOVERFLOW;
748 				fdrop(fp, td);
749 				break;
750 			}
751 			flp->l_start += foffset;
752 		}
753 		vp = fp->f_vnode;
754 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
755 		    F_POSIX);
756 		fdrop(fp, td);
757 		break;
758 
759 	case F_ADD_SEALS:
760 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
761 		if (error != 0)
762 			break;
763 		error = fo_add_seals(fp, arg);
764 		fdrop(fp, td);
765 		break;
766 
767 	case F_GET_SEALS:
768 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
769 		if (error != 0)
770 			break;
771 		if (fo_get_seals(fp, &seals) == 0)
772 			td->td_retval[0] = seals;
773 		else
774 			error = EINVAL;
775 		fdrop(fp, td);
776 		break;
777 
778 	case F_RDAHEAD:
779 		arg = arg ? 128 * 1024: 0;
780 		/* FALLTHROUGH */
781 	case F_READAHEAD:
782 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
783 		if (error != 0)
784 			break;
785 		if (fp->f_type != DTYPE_VNODE) {
786 			fdrop(fp, td);
787 			error = EBADF;
788 			break;
789 		}
790 		vp = fp->f_vnode;
791 		if (vp->v_type != VREG) {
792 			fdrop(fp, td);
793 			error = ENOTTY;
794 			break;
795 		}
796 
797 		/*
798 		 * Exclusive lock synchronizes against f_seqcount reads and
799 		 * writes in sequential_heuristic().
800 		 */
801 		error = vn_lock(vp, LK_EXCLUSIVE);
802 		if (error != 0) {
803 			fdrop(fp, td);
804 			break;
805 		}
806 		if (arg >= 0) {
807 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
808 			arg = MIN(arg, INT_MAX - bsize + 1);
809 			fp->f_seqcount = MIN(IO_SEQMAX,
810 			    (arg + bsize - 1) / bsize);
811 			atomic_set_int(&fp->f_flag, FRDAHEAD);
812 		} else {
813 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
814 		}
815 		VOP_UNLOCK(vp);
816 		fdrop(fp, td);
817 		break;
818 
819 	default:
820 		error = EINVAL;
821 		break;
822 	}
823 	return (error);
824 }
825 
826 static int
827 getmaxfd(struct thread *td)
828 {
829 
830 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
831 }
832 
833 /*
834  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
835  */
836 int
837 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
838 {
839 	struct filedesc *fdp;
840 	struct filedescent *oldfde, *newfde;
841 	struct proc *p;
842 	struct file *delfp;
843 	u_long *oioctls, *nioctls;
844 	int error, maxfd;
845 
846 	p = td->td_proc;
847 	fdp = p->p_fd;
848 	oioctls = NULL;
849 
850 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
851 	MPASS(mode < FDDUP_LASTMODE);
852 
853 	AUDIT_ARG_FD(old);
854 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
855 
856 	/*
857 	 * Verify we have a valid descriptor to dup from and possibly to
858 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
859 	 * return EINVAL when the new descriptor is out of bounds.
860 	 */
861 	if (old < 0)
862 		return (EBADF);
863 	if (new < 0)
864 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
865 	maxfd = getmaxfd(td);
866 	if (new >= maxfd)
867 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
868 
869 	error = EBADF;
870 	FILEDESC_XLOCK(fdp);
871 	if (fget_locked(fdp, old) == NULL)
872 		goto unlock;
873 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
874 		td->td_retval[0] = new;
875 		if (flags & FDDUP_FLAG_CLOEXEC)
876 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
877 		error = 0;
878 		goto unlock;
879 	}
880 
881 	oldfde = &fdp->fd_ofiles[old];
882 	if (!fhold(oldfde->fde_file))
883 		goto unlock;
884 
885 	/*
886 	 * If the caller specified a file descriptor, make sure the file
887 	 * table is large enough to hold it, and grab it.  Otherwise, just
888 	 * allocate a new descriptor the usual way.
889 	 */
890 	switch (mode) {
891 	case FDDUP_NORMAL:
892 	case FDDUP_FCNTL:
893 		if ((error = fdalloc(td, new, &new)) != 0) {
894 			fdrop(oldfde->fde_file, td);
895 			goto unlock;
896 		}
897 		break;
898 	case FDDUP_MUSTREPLACE:
899 		/* Target file descriptor must exist. */
900 		if (fget_locked(fdp, new) == NULL) {
901 			fdrop(oldfde->fde_file, td);
902 			goto unlock;
903 		}
904 		break;
905 	case FDDUP_FIXED:
906 		if (new >= fdp->fd_nfiles) {
907 			/*
908 			 * The resource limits are here instead of e.g.
909 			 * fdalloc(), because the file descriptor table may be
910 			 * shared between processes, so we can't really use
911 			 * racct_add()/racct_sub().  Instead of counting the
912 			 * number of actually allocated descriptors, just put
913 			 * the limit on the size of the file descriptor table.
914 			 */
915 #ifdef RACCT
916 			if (RACCT_ENABLED()) {
917 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
918 				if (error != 0) {
919 					error = EMFILE;
920 					fdrop(oldfde->fde_file, td);
921 					goto unlock;
922 				}
923 			}
924 #endif
925 			fdgrowtable_exp(fdp, new + 1);
926 		}
927 		if (!fdisused(fdp, new))
928 			fdused(fdp, new);
929 		break;
930 	default:
931 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
932 	}
933 
934 	KASSERT(old != new, ("new fd is same as old"));
935 
936 	newfde = &fdp->fd_ofiles[new];
937 	delfp = newfde->fde_file;
938 
939 	oioctls = filecaps_free_prep(&newfde->fde_caps);
940 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
941 
942 	/*
943 	 * Duplicate the source descriptor.
944 	 */
945 #ifdef CAPABILITIES
946 	seqc_write_begin(&newfde->fde_seqc);
947 #endif
948 	memcpy(newfde, oldfde, fde_change_size);
949 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
950 	    nioctls);
951 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
952 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
953 	else
954 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
955 #ifdef CAPABILITIES
956 	seqc_write_end(&newfde->fde_seqc);
957 #endif
958 	td->td_retval[0] = new;
959 
960 	error = 0;
961 
962 	if (delfp != NULL) {
963 		(void) closefp(fdp, new, delfp, td, 1);
964 		FILEDESC_UNLOCK_ASSERT(fdp);
965 	} else {
966 unlock:
967 		FILEDESC_XUNLOCK(fdp);
968 	}
969 
970 	filecaps_free_finish(oioctls);
971 	return (error);
972 }
973 
974 /*
975  * If sigio is on the list associated with a process or process group,
976  * disable signalling from the device, remove sigio from the list and
977  * free sigio.
978  */
979 void
980 funsetown(struct sigio **sigiop)
981 {
982 	struct sigio *sigio;
983 
984 	if (*sigiop == NULL)
985 		return;
986 	SIGIO_LOCK();
987 	sigio = *sigiop;
988 	if (sigio == NULL) {
989 		SIGIO_UNLOCK();
990 		return;
991 	}
992 	*(sigio->sio_myref) = NULL;
993 	if ((sigio)->sio_pgid < 0) {
994 		struct pgrp *pg = (sigio)->sio_pgrp;
995 		PGRP_LOCK(pg);
996 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
997 			    sigio, sio_pgsigio);
998 		PGRP_UNLOCK(pg);
999 	} else {
1000 		struct proc *p = (sigio)->sio_proc;
1001 		PROC_LOCK(p);
1002 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1003 			    sigio, sio_pgsigio);
1004 		PROC_UNLOCK(p);
1005 	}
1006 	SIGIO_UNLOCK();
1007 	crfree(sigio->sio_ucred);
1008 	free(sigio, M_SIGIO);
1009 }
1010 
1011 /*
1012  * Free a list of sigio structures.
1013  * We only need to lock the SIGIO_LOCK because we have made ourselves
1014  * inaccessible to callers of fsetown and therefore do not need to lock
1015  * the proc or pgrp struct for the list manipulation.
1016  */
1017 void
1018 funsetownlst(struct sigiolst *sigiolst)
1019 {
1020 	struct proc *p;
1021 	struct pgrp *pg;
1022 	struct sigio *sigio;
1023 
1024 	sigio = SLIST_FIRST(sigiolst);
1025 	if (sigio == NULL)
1026 		return;
1027 	p = NULL;
1028 	pg = NULL;
1029 
1030 	/*
1031 	 * Every entry of the list should belong
1032 	 * to a single proc or pgrp.
1033 	 */
1034 	if (sigio->sio_pgid < 0) {
1035 		pg = sigio->sio_pgrp;
1036 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1037 	} else /* if (sigio->sio_pgid > 0) */ {
1038 		p = sigio->sio_proc;
1039 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1040 	}
1041 
1042 	SIGIO_LOCK();
1043 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1044 		*(sigio->sio_myref) = NULL;
1045 		if (pg != NULL) {
1046 			KASSERT(sigio->sio_pgid < 0,
1047 			    ("Proc sigio in pgrp sigio list"));
1048 			KASSERT(sigio->sio_pgrp == pg,
1049 			    ("Bogus pgrp in sigio list"));
1050 			PGRP_LOCK(pg);
1051 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1052 			    sio_pgsigio);
1053 			PGRP_UNLOCK(pg);
1054 		} else /* if (p != NULL) */ {
1055 			KASSERT(sigio->sio_pgid > 0,
1056 			    ("Pgrp sigio in proc sigio list"));
1057 			KASSERT(sigio->sio_proc == p,
1058 			    ("Bogus proc in sigio list"));
1059 			PROC_LOCK(p);
1060 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1061 			    sio_pgsigio);
1062 			PROC_UNLOCK(p);
1063 		}
1064 		SIGIO_UNLOCK();
1065 		crfree(sigio->sio_ucred);
1066 		free(sigio, M_SIGIO);
1067 		SIGIO_LOCK();
1068 	}
1069 	SIGIO_UNLOCK();
1070 }
1071 
1072 /*
1073  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1074  *
1075  * After permission checking, add a sigio structure to the sigio list for
1076  * the process or process group.
1077  */
1078 int
1079 fsetown(pid_t pgid, struct sigio **sigiop)
1080 {
1081 	struct proc *proc;
1082 	struct pgrp *pgrp;
1083 	struct sigio *sigio;
1084 	int ret;
1085 
1086 	if (pgid == 0) {
1087 		funsetown(sigiop);
1088 		return (0);
1089 	}
1090 
1091 	ret = 0;
1092 
1093 	/* Allocate and fill in the new sigio out of locks. */
1094 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1095 	sigio->sio_pgid = pgid;
1096 	sigio->sio_ucred = crhold(curthread->td_ucred);
1097 	sigio->sio_myref = sigiop;
1098 
1099 	sx_slock(&proctree_lock);
1100 	if (pgid > 0) {
1101 		proc = pfind(pgid);
1102 		if (proc == NULL) {
1103 			ret = ESRCH;
1104 			goto fail;
1105 		}
1106 
1107 		/*
1108 		 * Policy - Don't allow a process to FSETOWN a process
1109 		 * in another session.
1110 		 *
1111 		 * Remove this test to allow maximum flexibility or
1112 		 * restrict FSETOWN to the current process or process
1113 		 * group for maximum safety.
1114 		 */
1115 		PROC_UNLOCK(proc);
1116 		if (proc->p_session != curthread->td_proc->p_session) {
1117 			ret = EPERM;
1118 			goto fail;
1119 		}
1120 
1121 		pgrp = NULL;
1122 	} else /* if (pgid < 0) */ {
1123 		pgrp = pgfind(-pgid);
1124 		if (pgrp == NULL) {
1125 			ret = ESRCH;
1126 			goto fail;
1127 		}
1128 		PGRP_UNLOCK(pgrp);
1129 
1130 		/*
1131 		 * Policy - Don't allow a process to FSETOWN a process
1132 		 * in another session.
1133 		 *
1134 		 * Remove this test to allow maximum flexibility or
1135 		 * restrict FSETOWN to the current process or process
1136 		 * group for maximum safety.
1137 		 */
1138 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1139 			ret = EPERM;
1140 			goto fail;
1141 		}
1142 
1143 		proc = NULL;
1144 	}
1145 	funsetown(sigiop);
1146 	if (pgid > 0) {
1147 		PROC_LOCK(proc);
1148 		/*
1149 		 * Since funsetownlst() is called without the proctree
1150 		 * locked, we need to check for P_WEXIT.
1151 		 * XXX: is ESRCH correct?
1152 		 */
1153 		if ((proc->p_flag & P_WEXIT) != 0) {
1154 			PROC_UNLOCK(proc);
1155 			ret = ESRCH;
1156 			goto fail;
1157 		}
1158 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1159 		sigio->sio_proc = proc;
1160 		PROC_UNLOCK(proc);
1161 	} else {
1162 		PGRP_LOCK(pgrp);
1163 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1164 		sigio->sio_pgrp = pgrp;
1165 		PGRP_UNLOCK(pgrp);
1166 	}
1167 	sx_sunlock(&proctree_lock);
1168 	SIGIO_LOCK();
1169 	*sigiop = sigio;
1170 	SIGIO_UNLOCK();
1171 	return (0);
1172 
1173 fail:
1174 	sx_sunlock(&proctree_lock);
1175 	crfree(sigio->sio_ucred);
1176 	free(sigio, M_SIGIO);
1177 	return (ret);
1178 }
1179 
1180 /*
1181  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1182  */
1183 pid_t
1184 fgetown(struct sigio **sigiop)
1185 {
1186 	pid_t pgid;
1187 
1188 	SIGIO_LOCK();
1189 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1190 	SIGIO_UNLOCK();
1191 	return (pgid);
1192 }
1193 
1194 /*
1195  * Function drops the filedesc lock on return.
1196  */
1197 static int
1198 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1199     int holdleaders)
1200 {
1201 	int error;
1202 
1203 	FILEDESC_XLOCK_ASSERT(fdp);
1204 
1205 	if (holdleaders) {
1206 		if (td->td_proc->p_fdtol != NULL) {
1207 			/*
1208 			 * Ask fdfree() to sleep to ensure that all relevant
1209 			 * process leaders can be traversed in closef().
1210 			 */
1211 			fdp->fd_holdleaderscount++;
1212 		} else {
1213 			holdleaders = 0;
1214 		}
1215 	}
1216 
1217 	/*
1218 	 * We now hold the fp reference that used to be owned by the
1219 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1220 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1221 	 * added, and deleteing a knote for the new fd.
1222 	 */
1223 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1224 		knote_fdclose(td, fd);
1225 
1226 	/*
1227 	 * We need to notify mqueue if the object is of type mqueue.
1228 	 */
1229 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1230 		mq_fdclose(td, fd, fp);
1231 	FILEDESC_XUNLOCK(fdp);
1232 
1233 	error = closef(fp, td);
1234 	if (holdleaders) {
1235 		FILEDESC_XLOCK(fdp);
1236 		fdp->fd_holdleaderscount--;
1237 		if (fdp->fd_holdleaderscount == 0 &&
1238 		    fdp->fd_holdleaderswakeup != 0) {
1239 			fdp->fd_holdleaderswakeup = 0;
1240 			wakeup(&fdp->fd_holdleaderscount);
1241 		}
1242 		FILEDESC_XUNLOCK(fdp);
1243 	}
1244 	return (error);
1245 }
1246 
1247 /*
1248  * Close a file descriptor.
1249  */
1250 #ifndef _SYS_SYSPROTO_H_
1251 struct close_args {
1252 	int     fd;
1253 };
1254 #endif
1255 /* ARGSUSED */
1256 int
1257 sys_close(struct thread *td, struct close_args *uap)
1258 {
1259 
1260 	return (kern_close(td, uap->fd));
1261 }
1262 
1263 int
1264 kern_close(struct thread *td, int fd)
1265 {
1266 	struct filedesc *fdp;
1267 	struct file *fp;
1268 
1269 	fdp = td->td_proc->p_fd;
1270 
1271 	AUDIT_SYSCLOSE(td, fd);
1272 
1273 	FILEDESC_XLOCK(fdp);
1274 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1275 		FILEDESC_XUNLOCK(fdp);
1276 		return (EBADF);
1277 	}
1278 	fdfree(fdp, fd);
1279 
1280 	/* closefp() drops the FILEDESC lock for us. */
1281 	return (closefp(fdp, fd, fp, td, 1));
1282 }
1283 
1284 /*
1285  * Close open file descriptors.
1286  */
1287 #ifndef _SYS_SYSPROTO_H_
1288 struct closefrom_args {
1289 	int	lowfd;
1290 };
1291 #endif
1292 /* ARGSUSED */
1293 int
1294 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1295 {
1296 	struct filedesc *fdp;
1297 	int fd;
1298 
1299 	fdp = td->td_proc->p_fd;
1300 	AUDIT_ARG_FD(uap->lowfd);
1301 
1302 	/*
1303 	 * Treat negative starting file descriptor values identical to
1304 	 * closefrom(0) which closes all files.
1305 	 */
1306 	if (uap->lowfd < 0)
1307 		uap->lowfd = 0;
1308 	FILEDESC_SLOCK(fdp);
1309 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1310 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1311 			FILEDESC_SUNLOCK(fdp);
1312 			(void)kern_close(td, fd);
1313 			FILEDESC_SLOCK(fdp);
1314 		}
1315 	}
1316 	FILEDESC_SUNLOCK(fdp);
1317 	return (0);
1318 }
1319 
1320 #if defined(COMPAT_43)
1321 /*
1322  * Return status information about a file descriptor.
1323  */
1324 #ifndef _SYS_SYSPROTO_H_
1325 struct ofstat_args {
1326 	int	fd;
1327 	struct	ostat *sb;
1328 };
1329 #endif
1330 /* ARGSUSED */
1331 int
1332 ofstat(struct thread *td, struct ofstat_args *uap)
1333 {
1334 	struct ostat oub;
1335 	struct stat ub;
1336 	int error;
1337 
1338 	error = kern_fstat(td, uap->fd, &ub);
1339 	if (error == 0) {
1340 		cvtstat(&ub, &oub);
1341 		error = copyout(&oub, uap->sb, sizeof(oub));
1342 	}
1343 	return (error);
1344 }
1345 #endif /* COMPAT_43 */
1346 
1347 #if defined(COMPAT_FREEBSD11)
1348 int
1349 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1350 {
1351 	struct stat sb;
1352 	struct freebsd11_stat osb;
1353 	int error;
1354 
1355 	error = kern_fstat(td, uap->fd, &sb);
1356 	if (error != 0)
1357 		return (error);
1358 	error = freebsd11_cvtstat(&sb, &osb);
1359 	if (error == 0)
1360 		error = copyout(&osb, uap->sb, sizeof(osb));
1361 	return (error);
1362 }
1363 #endif	/* COMPAT_FREEBSD11 */
1364 
1365 /*
1366  * Return status information about a file descriptor.
1367  */
1368 #ifndef _SYS_SYSPROTO_H_
1369 struct fstat_args {
1370 	int	fd;
1371 	struct	stat *sb;
1372 };
1373 #endif
1374 /* ARGSUSED */
1375 int
1376 sys_fstat(struct thread *td, struct fstat_args *uap)
1377 {
1378 	struct stat ub;
1379 	int error;
1380 
1381 	error = kern_fstat(td, uap->fd, &ub);
1382 	if (error == 0)
1383 		error = copyout(&ub, uap->sb, sizeof(ub));
1384 	return (error);
1385 }
1386 
1387 int
1388 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1389 {
1390 	struct file *fp;
1391 	int error;
1392 
1393 	AUDIT_ARG_FD(fd);
1394 
1395 	error = fget(td, fd, &cap_fstat_rights, &fp);
1396 	if (error != 0)
1397 		return (error);
1398 
1399 	AUDIT_ARG_FILE(td->td_proc, fp);
1400 
1401 	error = fo_stat(fp, sbp, td->td_ucred, td);
1402 	fdrop(fp, td);
1403 #ifdef __STAT_TIME_T_EXT
1404 	if (error == 0) {
1405 		sbp->st_atim_ext = 0;
1406 		sbp->st_mtim_ext = 0;
1407 		sbp->st_ctim_ext = 0;
1408 		sbp->st_btim_ext = 0;
1409 	}
1410 #endif
1411 #ifdef KTRACE
1412 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1413 		ktrstat(sbp);
1414 #endif
1415 	return (error);
1416 }
1417 
1418 #if defined(COMPAT_FREEBSD11)
1419 /*
1420  * Return status information about a file descriptor.
1421  */
1422 #ifndef _SYS_SYSPROTO_H_
1423 struct freebsd11_nfstat_args {
1424 	int	fd;
1425 	struct	nstat *sb;
1426 };
1427 #endif
1428 /* ARGSUSED */
1429 int
1430 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1431 {
1432 	struct nstat nub;
1433 	struct stat ub;
1434 	int error;
1435 
1436 	error = kern_fstat(td, uap->fd, &ub);
1437 	if (error == 0) {
1438 		freebsd11_cvtnstat(&ub, &nub);
1439 		error = copyout(&nub, uap->sb, sizeof(nub));
1440 	}
1441 	return (error);
1442 }
1443 #endif /* COMPAT_FREEBSD11 */
1444 
1445 /*
1446  * Return pathconf information about a file descriptor.
1447  */
1448 #ifndef _SYS_SYSPROTO_H_
1449 struct fpathconf_args {
1450 	int	fd;
1451 	int	name;
1452 };
1453 #endif
1454 /* ARGSUSED */
1455 int
1456 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1457 {
1458 	long value;
1459 	int error;
1460 
1461 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1462 	if (error == 0)
1463 		td->td_retval[0] = value;
1464 	return (error);
1465 }
1466 
1467 int
1468 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1469 {
1470 	struct file *fp;
1471 	struct vnode *vp;
1472 	int error;
1473 
1474 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1475 	if (error != 0)
1476 		return (error);
1477 
1478 	if (name == _PC_ASYNC_IO) {
1479 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1480 		goto out;
1481 	}
1482 	vp = fp->f_vnode;
1483 	if (vp != NULL) {
1484 		vn_lock(vp, LK_SHARED | LK_RETRY);
1485 		error = VOP_PATHCONF(vp, name, valuep);
1486 		VOP_UNLOCK(vp);
1487 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1488 		if (name != _PC_PIPE_BUF) {
1489 			error = EINVAL;
1490 		} else {
1491 			*valuep = PIPE_BUF;
1492 			error = 0;
1493 		}
1494 	} else {
1495 		error = EOPNOTSUPP;
1496 	}
1497 out:
1498 	fdrop(fp, td);
1499 	return (error);
1500 }
1501 
1502 /*
1503  * Copy filecaps structure allocating memory for ioctls array if needed.
1504  *
1505  * The last parameter indicates whether the fdtable is locked. If it is not and
1506  * ioctls are encountered, copying fails and the caller must lock the table.
1507  *
1508  * Note that if the table was not locked, the caller has to check the relevant
1509  * sequence counter to determine whether the operation was successful.
1510  */
1511 bool
1512 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1513 {
1514 	size_t size;
1515 
1516 	if (src->fc_ioctls != NULL && !locked)
1517 		return (false);
1518 	memcpy(dst, src, sizeof(*src));
1519 	if (src->fc_ioctls == NULL)
1520 		return (true);
1521 
1522 	KASSERT(src->fc_nioctls > 0,
1523 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1524 
1525 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1526 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1527 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1528 	return (true);
1529 }
1530 
1531 static u_long *
1532 filecaps_copy_prep(const struct filecaps *src)
1533 {
1534 	u_long *ioctls;
1535 	size_t size;
1536 
1537 	if (__predict_true(src->fc_ioctls == NULL))
1538 		return (NULL);
1539 
1540 	KASSERT(src->fc_nioctls > 0,
1541 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1542 
1543 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1544 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1545 	return (ioctls);
1546 }
1547 
1548 static void
1549 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1550     u_long *ioctls)
1551 {
1552 	size_t size;
1553 
1554 	*dst = *src;
1555 	if (__predict_true(src->fc_ioctls == NULL)) {
1556 		MPASS(ioctls == NULL);
1557 		return;
1558 	}
1559 
1560 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1561 	dst->fc_ioctls = ioctls;
1562 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1563 }
1564 
1565 /*
1566  * Move filecaps structure to the new place and clear the old place.
1567  */
1568 void
1569 filecaps_move(struct filecaps *src, struct filecaps *dst)
1570 {
1571 
1572 	*dst = *src;
1573 	bzero(src, sizeof(*src));
1574 }
1575 
1576 /*
1577  * Fill the given filecaps structure with full rights.
1578  */
1579 static void
1580 filecaps_fill(struct filecaps *fcaps)
1581 {
1582 
1583 	CAP_ALL(&fcaps->fc_rights);
1584 	fcaps->fc_ioctls = NULL;
1585 	fcaps->fc_nioctls = -1;
1586 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1587 }
1588 
1589 /*
1590  * Free memory allocated within filecaps structure.
1591  */
1592 void
1593 filecaps_free(struct filecaps *fcaps)
1594 {
1595 
1596 	free(fcaps->fc_ioctls, M_FILECAPS);
1597 	bzero(fcaps, sizeof(*fcaps));
1598 }
1599 
1600 static u_long *
1601 filecaps_free_prep(struct filecaps *fcaps)
1602 {
1603 	u_long *ioctls;
1604 
1605 	ioctls = fcaps->fc_ioctls;
1606 	bzero(fcaps, sizeof(*fcaps));
1607 	return (ioctls);
1608 }
1609 
1610 static void
1611 filecaps_free_finish(u_long *ioctls)
1612 {
1613 
1614 	free(ioctls, M_FILECAPS);
1615 }
1616 
1617 /*
1618  * Validate the given filecaps structure.
1619  */
1620 static void
1621 filecaps_validate(const struct filecaps *fcaps, const char *func)
1622 {
1623 
1624 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1625 	    ("%s: invalid rights", func));
1626 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1627 	    ("%s: invalid fcntls", func));
1628 	KASSERT(fcaps->fc_fcntls == 0 ||
1629 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1630 	    ("%s: fcntls without CAP_FCNTL", func));
1631 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1632 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1633 	    ("%s: invalid ioctls", func));
1634 	KASSERT(fcaps->fc_nioctls == 0 ||
1635 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1636 	    ("%s: ioctls without CAP_IOCTL", func));
1637 }
1638 
1639 static void
1640 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1641 {
1642 	int nfd1;
1643 
1644 	FILEDESC_XLOCK_ASSERT(fdp);
1645 
1646 	nfd1 = fdp->fd_nfiles * 2;
1647 	if (nfd1 < nfd)
1648 		nfd1 = nfd;
1649 	fdgrowtable(fdp, nfd1);
1650 }
1651 
1652 /*
1653  * Grow the file table to accommodate (at least) nfd descriptors.
1654  */
1655 static void
1656 fdgrowtable(struct filedesc *fdp, int nfd)
1657 {
1658 	struct filedesc0 *fdp0;
1659 	struct freetable *ft;
1660 	struct fdescenttbl *ntable;
1661 	struct fdescenttbl *otable;
1662 	int nnfiles, onfiles;
1663 	NDSLOTTYPE *nmap, *omap;
1664 
1665 	/*
1666 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1667 	 * growing it to accommodate for the one we are going to copy from. There
1668 	 * is no need to have a lock on this one as it's not visible to anyone.
1669 	 */
1670 	if (fdp->fd_lastfile != -1)
1671 		FILEDESC_XLOCK_ASSERT(fdp);
1672 
1673 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1674 
1675 	/* save old values */
1676 	onfiles = fdp->fd_nfiles;
1677 	otable = fdp->fd_files;
1678 	omap = fdp->fd_map;
1679 
1680 	/* compute the size of the new table */
1681 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1682 	if (nnfiles <= onfiles)
1683 		/* the table is already large enough */
1684 		return;
1685 
1686 	/*
1687 	 * Allocate a new table.  We need enough space for the number of
1688 	 * entries, file entries themselves and the struct freetable we will use
1689 	 * when we decommission the table and place it on the freelist.
1690 	 * We place the struct freetable in the middle so we don't have
1691 	 * to worry about padding.
1692 	 */
1693 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1694 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1695 	    sizeof(struct freetable),
1696 	    M_FILEDESC, M_ZERO | M_WAITOK);
1697 	/* copy the old data */
1698 	ntable->fdt_nfiles = nnfiles;
1699 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1700 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1701 
1702 	/*
1703 	 * Allocate a new map only if the old is not large enough.  It will
1704 	 * grow at a slower rate than the table as it can map more
1705 	 * entries than the table can hold.
1706 	 */
1707 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1708 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1709 		    M_ZERO | M_WAITOK);
1710 		/* copy over the old data and update the pointer */
1711 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1712 		fdp->fd_map = nmap;
1713 	}
1714 
1715 	/*
1716 	 * Make sure that ntable is correctly initialized before we replace
1717 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1718 	 * data.
1719 	 */
1720 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1721 
1722 	/*
1723 	 * Do not free the old file table, as some threads may still
1724 	 * reference entries within it.  Instead, place it on a freelist
1725 	 * which will be processed when the struct filedesc is released.
1726 	 *
1727 	 * Note that if onfiles == NDFILE, we're dealing with the original
1728 	 * static allocation contained within (struct filedesc0 *)fdp,
1729 	 * which must not be freed.
1730 	 */
1731 	if (onfiles > NDFILE) {
1732 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1733 		fdp0 = (struct filedesc0 *)fdp;
1734 		ft->ft_table = otable;
1735 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1736 	}
1737 	/*
1738 	 * The map does not have the same possibility of threads still
1739 	 * holding references to it.  So always free it as long as it
1740 	 * does not reference the original static allocation.
1741 	 */
1742 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1743 		free(omap, M_FILEDESC);
1744 }
1745 
1746 /*
1747  * Allocate a file descriptor for the process.
1748  */
1749 int
1750 fdalloc(struct thread *td, int minfd, int *result)
1751 {
1752 	struct proc *p = td->td_proc;
1753 	struct filedesc *fdp = p->p_fd;
1754 	int fd, maxfd, allocfd;
1755 #ifdef RACCT
1756 	int error;
1757 #endif
1758 
1759 	FILEDESC_XLOCK_ASSERT(fdp);
1760 
1761 	if (fdp->fd_freefile > minfd)
1762 		minfd = fdp->fd_freefile;
1763 
1764 	maxfd = getmaxfd(td);
1765 
1766 	/*
1767 	 * Search the bitmap for a free descriptor starting at minfd.
1768 	 * If none is found, grow the file table.
1769 	 */
1770 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1771 	if (fd >= maxfd)
1772 		return (EMFILE);
1773 	if (fd >= fdp->fd_nfiles) {
1774 		allocfd = min(fd * 2, maxfd);
1775 #ifdef RACCT
1776 		if (RACCT_ENABLED()) {
1777 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1778 			if (error != 0)
1779 				return (EMFILE);
1780 		}
1781 #endif
1782 		/*
1783 		 * fd is already equal to first free descriptor >= minfd, so
1784 		 * we only need to grow the table and we are done.
1785 		 */
1786 		fdgrowtable_exp(fdp, allocfd);
1787 	}
1788 
1789 	/*
1790 	 * Perform some sanity checks, then mark the file descriptor as
1791 	 * used and return it to the caller.
1792 	 */
1793 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1794 	    ("invalid descriptor %d", fd));
1795 	KASSERT(!fdisused(fdp, fd),
1796 	    ("fd_first_free() returned non-free descriptor"));
1797 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1798 	    ("file descriptor isn't free"));
1799 	fdused(fdp, fd);
1800 	*result = fd;
1801 	return (0);
1802 }
1803 
1804 /*
1805  * Allocate n file descriptors for the process.
1806  */
1807 int
1808 fdallocn(struct thread *td, int minfd, int *fds, int n)
1809 {
1810 	struct proc *p = td->td_proc;
1811 	struct filedesc *fdp = p->p_fd;
1812 	int i;
1813 
1814 	FILEDESC_XLOCK_ASSERT(fdp);
1815 
1816 	for (i = 0; i < n; i++)
1817 		if (fdalloc(td, 0, &fds[i]) != 0)
1818 			break;
1819 
1820 	if (i < n) {
1821 		for (i--; i >= 0; i--)
1822 			fdunused(fdp, fds[i]);
1823 		return (EMFILE);
1824 	}
1825 
1826 	return (0);
1827 }
1828 
1829 /*
1830  * Create a new open file structure and allocate a file descriptor for the
1831  * process that refers to it.  We add one reference to the file for the
1832  * descriptor table and one reference for resultfp. This is to prevent us
1833  * being preempted and the entry in the descriptor table closed after we
1834  * release the FILEDESC lock.
1835  */
1836 int
1837 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1838     struct filecaps *fcaps)
1839 {
1840 	struct file *fp;
1841 	int error, fd;
1842 
1843 	error = falloc_noinstall(td, &fp);
1844 	if (error)
1845 		return (error);		/* no reference held on error */
1846 
1847 	error = finstall(td, fp, &fd, flags, fcaps);
1848 	if (error) {
1849 		fdrop(fp, td);		/* one reference (fp only) */
1850 		return (error);
1851 	}
1852 
1853 	if (resultfp != NULL)
1854 		*resultfp = fp;		/* copy out result */
1855 	else
1856 		fdrop(fp, td);		/* release local reference */
1857 
1858 	if (resultfd != NULL)
1859 		*resultfd = fd;
1860 
1861 	return (0);
1862 }
1863 
1864 /*
1865  * Create a new open file structure without allocating a file descriptor.
1866  */
1867 int
1868 falloc_noinstall(struct thread *td, struct file **resultfp)
1869 {
1870 	struct file *fp;
1871 	int maxuserfiles = maxfiles - (maxfiles / 20);
1872 	int openfiles_new;
1873 	static struct timeval lastfail;
1874 	static int curfail;
1875 
1876 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1877 
1878 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1879 	if ((openfiles_new >= maxuserfiles &&
1880 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1881 	    openfiles_new >= maxfiles) {
1882 		atomic_subtract_int(&openfiles, 1);
1883 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1884 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1885 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1886 		}
1887 		return (ENFILE);
1888 	}
1889 	fp = uma_zalloc(file_zone, M_WAITOK);
1890 	bzero(fp, sizeof(*fp));
1891 	refcount_init(&fp->f_count, 1);
1892 	fp->f_cred = crhold(td->td_ucred);
1893 	fp->f_ops = &badfileops;
1894 	*resultfp = fp;
1895 	return (0);
1896 }
1897 
1898 /*
1899  * Install a file in a file descriptor table.
1900  */
1901 void
1902 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1903     struct filecaps *fcaps)
1904 {
1905 	struct filedescent *fde;
1906 
1907 	MPASS(fp != NULL);
1908 	if (fcaps != NULL)
1909 		filecaps_validate(fcaps, __func__);
1910 	FILEDESC_XLOCK_ASSERT(fdp);
1911 
1912 	fde = &fdp->fd_ofiles[fd];
1913 #ifdef CAPABILITIES
1914 	seqc_write_begin(&fde->fde_seqc);
1915 #endif
1916 	fde->fde_file = fp;
1917 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1918 	if (fcaps != NULL)
1919 		filecaps_move(fcaps, &fde->fde_caps);
1920 	else
1921 		filecaps_fill(&fde->fde_caps);
1922 #ifdef CAPABILITIES
1923 	seqc_write_end(&fde->fde_seqc);
1924 #endif
1925 }
1926 
1927 int
1928 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1929     struct filecaps *fcaps)
1930 {
1931 	struct filedesc *fdp = td->td_proc->p_fd;
1932 	int error;
1933 
1934 	MPASS(fd != NULL);
1935 
1936 	if (!fhold(fp))
1937 		return (EBADF);
1938 	FILEDESC_XLOCK(fdp);
1939 	if ((error = fdalloc(td, 0, fd))) {
1940 		FILEDESC_XUNLOCK(fdp);
1941 		fdrop(fp, td);
1942 		return (error);
1943 	}
1944 	_finstall(fdp, fp, *fd, flags, fcaps);
1945 	FILEDESC_XUNLOCK(fdp);
1946 	return (0);
1947 }
1948 
1949 /*
1950  * Build a new filedesc structure from another.
1951  * Copy the current, root, and jail root vnode references.
1952  *
1953  * If fdp is not NULL, return with it shared locked.
1954  */
1955 struct filedesc *
1956 fdinit(struct filedesc *fdp, bool prepfiles)
1957 {
1958 	struct filedesc0 *newfdp0;
1959 	struct filedesc *newfdp;
1960 
1961 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1962 	newfdp = &newfdp0->fd_fd;
1963 
1964 	/* Create the file descriptor table. */
1965 	FILEDESC_LOCK_INIT(newfdp);
1966 	refcount_init(&newfdp->fd_refcnt, 1);
1967 	refcount_init(&newfdp->fd_holdcnt, 1);
1968 	newfdp->fd_cmask = CMASK;
1969 	newfdp->fd_map = newfdp0->fd_dmap;
1970 	newfdp->fd_lastfile = -1;
1971 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1972 	newfdp->fd_files->fdt_nfiles = NDFILE;
1973 
1974 	if (fdp == NULL)
1975 		return (newfdp);
1976 
1977 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1978 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1979 
1980 	FILEDESC_SLOCK(fdp);
1981 	newfdp->fd_cdir = fdp->fd_cdir;
1982 	if (newfdp->fd_cdir)
1983 		vrefact(newfdp->fd_cdir);
1984 	newfdp->fd_rdir = fdp->fd_rdir;
1985 	if (newfdp->fd_rdir)
1986 		vrefact(newfdp->fd_rdir);
1987 	newfdp->fd_jdir = fdp->fd_jdir;
1988 	if (newfdp->fd_jdir)
1989 		vrefact(newfdp->fd_jdir);
1990 
1991 	if (!prepfiles) {
1992 		FILEDESC_SUNLOCK(fdp);
1993 	} else {
1994 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1995 			FILEDESC_SUNLOCK(fdp);
1996 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1997 			FILEDESC_SLOCK(fdp);
1998 		}
1999 	}
2000 
2001 	return (newfdp);
2002 }
2003 
2004 static struct filedesc *
2005 fdhold(struct proc *p)
2006 {
2007 	struct filedesc *fdp;
2008 
2009 	PROC_LOCK_ASSERT(p, MA_OWNED);
2010 	fdp = p->p_fd;
2011 	if (fdp != NULL)
2012 		refcount_acquire(&fdp->fd_holdcnt);
2013 	return (fdp);
2014 }
2015 
2016 static void
2017 fddrop(struct filedesc *fdp)
2018 {
2019 
2020 	if (fdp->fd_holdcnt > 1) {
2021 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2022 			return;
2023 	}
2024 
2025 	FILEDESC_LOCK_DESTROY(fdp);
2026 	uma_zfree(filedesc0_zone, fdp);
2027 }
2028 
2029 /*
2030  * Share a filedesc structure.
2031  */
2032 struct filedesc *
2033 fdshare(struct filedesc *fdp)
2034 {
2035 
2036 	refcount_acquire(&fdp->fd_refcnt);
2037 	return (fdp);
2038 }
2039 
2040 /*
2041  * Unshare a filedesc structure, if necessary by making a copy
2042  */
2043 void
2044 fdunshare(struct thread *td)
2045 {
2046 	struct filedesc *tmp;
2047 	struct proc *p = td->td_proc;
2048 
2049 	if (p->p_fd->fd_refcnt == 1)
2050 		return;
2051 
2052 	tmp = fdcopy(p->p_fd);
2053 	fdescfree(td);
2054 	p->p_fd = tmp;
2055 }
2056 
2057 void
2058 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2059 {
2060 
2061 	fdescfree(td);
2062 	td->td_proc->p_fd = fdp;
2063 }
2064 
2065 /*
2066  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2067  * this is to ease callers, not catch errors.
2068  */
2069 struct filedesc *
2070 fdcopy(struct filedesc *fdp)
2071 {
2072 	struct filedesc *newfdp;
2073 	struct filedescent *nfde, *ofde;
2074 	int i;
2075 
2076 	MPASS(fdp != NULL);
2077 
2078 	newfdp = fdinit(fdp, true);
2079 	/* copy all passable descriptors (i.e. not kqueue) */
2080 	newfdp->fd_freefile = -1;
2081 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
2082 		ofde = &fdp->fd_ofiles[i];
2083 		if (ofde->fde_file == NULL ||
2084 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2085 		    !fhold(ofde->fde_file)) {
2086 			if (newfdp->fd_freefile == -1)
2087 				newfdp->fd_freefile = i;
2088 			continue;
2089 		}
2090 		nfde = &newfdp->fd_ofiles[i];
2091 		*nfde = *ofde;
2092 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2093 		fdused_init(newfdp, i);
2094 		newfdp->fd_lastfile = i;
2095 	}
2096 	if (newfdp->fd_freefile == -1)
2097 		newfdp->fd_freefile = i;
2098 	newfdp->fd_cmask = fdp->fd_cmask;
2099 	FILEDESC_SUNLOCK(fdp);
2100 	return (newfdp);
2101 }
2102 
2103 /*
2104  * Copies a filedesc structure, while remapping all file descriptors
2105  * stored inside using a translation table.
2106  *
2107  * File descriptors are copied over to the new file descriptor table,
2108  * regardless of whether the close-on-exec flag is set.
2109  */
2110 int
2111 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2112     struct filedesc **ret)
2113 {
2114 	struct filedesc *newfdp;
2115 	struct filedescent *nfde, *ofde;
2116 	int error, i;
2117 
2118 	MPASS(fdp != NULL);
2119 
2120 	newfdp = fdinit(fdp, true);
2121 	if (nfds > fdp->fd_lastfile + 1) {
2122 		/* New table cannot be larger than the old one. */
2123 		error = E2BIG;
2124 		goto bad;
2125 	}
2126 	/* Copy all passable descriptors (i.e. not kqueue). */
2127 	newfdp->fd_freefile = nfds;
2128 	for (i = 0; i < nfds; ++i) {
2129 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2130 			/* File descriptor out of bounds. */
2131 			error = EBADF;
2132 			goto bad;
2133 		}
2134 		ofde = &fdp->fd_ofiles[fds[i]];
2135 		if (ofde->fde_file == NULL) {
2136 			/* Unused file descriptor. */
2137 			error = EBADF;
2138 			goto bad;
2139 		}
2140 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2141 			/* File descriptor cannot be passed. */
2142 			error = EINVAL;
2143 			goto bad;
2144 		}
2145 		if (!fhold(nfde->fde_file)) {
2146 			error = EBADF;
2147 			goto bad;
2148 		}
2149 		nfde = &newfdp->fd_ofiles[i];
2150 		*nfde = *ofde;
2151 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2152 		fdused_init(newfdp, i);
2153 		newfdp->fd_lastfile = i;
2154 	}
2155 	newfdp->fd_cmask = fdp->fd_cmask;
2156 	FILEDESC_SUNLOCK(fdp);
2157 	*ret = newfdp;
2158 	return (0);
2159 bad:
2160 	FILEDESC_SUNLOCK(fdp);
2161 	fdescfree_remapped(newfdp);
2162 	return (error);
2163 }
2164 
2165 /*
2166  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2167  * one of processes using it exits) and the table used to be shared.
2168  */
2169 static void
2170 fdclearlocks(struct thread *td)
2171 {
2172 	struct filedesc *fdp;
2173 	struct filedesc_to_leader *fdtol;
2174 	struct flock lf;
2175 	struct file *fp;
2176 	struct proc *p;
2177 	struct vnode *vp;
2178 	int i;
2179 
2180 	p = td->td_proc;
2181 	fdp = p->p_fd;
2182 	fdtol = p->p_fdtol;
2183 	MPASS(fdtol != NULL);
2184 
2185 	FILEDESC_XLOCK(fdp);
2186 	KASSERT(fdtol->fdl_refcount > 0,
2187 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2188 	    fdtol->fdl_refcount));
2189 	if (fdtol->fdl_refcount == 1 &&
2190 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2191 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2192 			fp = fdp->fd_ofiles[i].fde_file;
2193 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2194 			    !fhold(fp))
2195 				continue;
2196 			FILEDESC_XUNLOCK(fdp);
2197 			lf.l_whence = SEEK_SET;
2198 			lf.l_start = 0;
2199 			lf.l_len = 0;
2200 			lf.l_type = F_UNLCK;
2201 			vp = fp->f_vnode;
2202 			(void) VOP_ADVLOCK(vp,
2203 			    (caddr_t)p->p_leader, F_UNLCK,
2204 			    &lf, F_POSIX);
2205 			FILEDESC_XLOCK(fdp);
2206 			fdrop(fp, td);
2207 		}
2208 	}
2209 retry:
2210 	if (fdtol->fdl_refcount == 1) {
2211 		if (fdp->fd_holdleaderscount > 0 &&
2212 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2213 			/*
2214 			 * close() or kern_dup() has cleared a reference
2215 			 * in a shared file descriptor table.
2216 			 */
2217 			fdp->fd_holdleaderswakeup = 1;
2218 			sx_sleep(&fdp->fd_holdleaderscount,
2219 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2220 			goto retry;
2221 		}
2222 		if (fdtol->fdl_holdcount > 0) {
2223 			/*
2224 			 * Ensure that fdtol->fdl_leader remains
2225 			 * valid in closef().
2226 			 */
2227 			fdtol->fdl_wakeup = 1;
2228 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2229 			    "fdlhold", 0);
2230 			goto retry;
2231 		}
2232 	}
2233 	fdtol->fdl_refcount--;
2234 	if (fdtol->fdl_refcount == 0 &&
2235 	    fdtol->fdl_holdcount == 0) {
2236 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2237 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2238 	} else
2239 		fdtol = NULL;
2240 	p->p_fdtol = NULL;
2241 	FILEDESC_XUNLOCK(fdp);
2242 	if (fdtol != NULL)
2243 		free(fdtol, M_FILEDESC_TO_LEADER);
2244 }
2245 
2246 /*
2247  * Release a filedesc structure.
2248  */
2249 static void
2250 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2251 {
2252 	struct filedesc0 *fdp0;
2253 	struct freetable *ft, *tft;
2254 	struct filedescent *fde;
2255 	struct file *fp;
2256 	int i;
2257 
2258 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2259 		fde = &fdp->fd_ofiles[i];
2260 		fp = fde->fde_file;
2261 		if (fp != NULL) {
2262 			fdefree_last(fde);
2263 			if (needclose)
2264 				(void) closef(fp, td);
2265 			else
2266 				fdrop(fp, td);
2267 		}
2268 	}
2269 
2270 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2271 		free(fdp->fd_map, M_FILEDESC);
2272 	if (fdp->fd_nfiles > NDFILE)
2273 		free(fdp->fd_files, M_FILEDESC);
2274 
2275 	fdp0 = (struct filedesc0 *)fdp;
2276 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2277 		free(ft->ft_table, M_FILEDESC);
2278 
2279 	fddrop(fdp);
2280 }
2281 
2282 void
2283 fdescfree(struct thread *td)
2284 {
2285 	struct proc *p;
2286 	struct filedesc *fdp;
2287 	struct vnode *cdir, *jdir, *rdir;
2288 
2289 	p = td->td_proc;
2290 	fdp = p->p_fd;
2291 	MPASS(fdp != NULL);
2292 
2293 #ifdef RACCT
2294 	if (RACCT_ENABLED())
2295 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2296 #endif
2297 
2298 	if (p->p_fdtol != NULL)
2299 		fdclearlocks(td);
2300 
2301 	PROC_LOCK(p);
2302 	p->p_fd = NULL;
2303 	PROC_UNLOCK(p);
2304 
2305 	if (refcount_release(&fdp->fd_refcnt) == 0)
2306 		return;
2307 
2308 	FILEDESC_XLOCK(fdp);
2309 	cdir = fdp->fd_cdir;
2310 	fdp->fd_cdir = NULL;
2311 	rdir = fdp->fd_rdir;
2312 	fdp->fd_rdir = NULL;
2313 	jdir = fdp->fd_jdir;
2314 	fdp->fd_jdir = NULL;
2315 	FILEDESC_XUNLOCK(fdp);
2316 
2317 	if (cdir != NULL)
2318 		vrele(cdir);
2319 	if (rdir != NULL)
2320 		vrele(rdir);
2321 	if (jdir != NULL)
2322 		vrele(jdir);
2323 
2324 	fdescfree_fds(td, fdp, 1);
2325 }
2326 
2327 void
2328 fdescfree_remapped(struct filedesc *fdp)
2329 {
2330 
2331 	if (fdp->fd_cdir != NULL)
2332 		vrele(fdp->fd_cdir);
2333 	if (fdp->fd_rdir != NULL)
2334 		vrele(fdp->fd_rdir);
2335 	if (fdp->fd_jdir != NULL)
2336 		vrele(fdp->fd_jdir);
2337 
2338 	fdescfree_fds(curthread, fdp, 0);
2339 }
2340 
2341 /*
2342  * For setugid programs, we don't want to people to use that setugidness
2343  * to generate error messages which write to a file which otherwise would
2344  * otherwise be off-limits to the process.  We check for filesystems where
2345  * the vnode can change out from under us after execve (like [lin]procfs).
2346  *
2347  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2348  * sufficient.  We also don't check for setugidness since we know we are.
2349  */
2350 static bool
2351 is_unsafe(struct file *fp)
2352 {
2353 	struct vnode *vp;
2354 
2355 	if (fp->f_type != DTYPE_VNODE)
2356 		return (false);
2357 
2358 	vp = fp->f_vnode;
2359 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2360 }
2361 
2362 /*
2363  * Make this setguid thing safe, if at all possible.
2364  */
2365 void
2366 fdsetugidsafety(struct thread *td)
2367 {
2368 	struct filedesc *fdp;
2369 	struct file *fp;
2370 	int i;
2371 
2372 	fdp = td->td_proc->p_fd;
2373 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2374 	MPASS(fdp->fd_nfiles >= 3);
2375 	for (i = 0; i <= 2; i++) {
2376 		fp = fdp->fd_ofiles[i].fde_file;
2377 		if (fp != NULL && is_unsafe(fp)) {
2378 			FILEDESC_XLOCK(fdp);
2379 			knote_fdclose(td, i);
2380 			/*
2381 			 * NULL-out descriptor prior to close to avoid
2382 			 * a race while close blocks.
2383 			 */
2384 			fdfree(fdp, i);
2385 			FILEDESC_XUNLOCK(fdp);
2386 			(void) closef(fp, td);
2387 		}
2388 	}
2389 }
2390 
2391 /*
2392  * If a specific file object occupies a specific file descriptor, close the
2393  * file descriptor entry and drop a reference on the file object.  This is a
2394  * convenience function to handle a subsequent error in a function that calls
2395  * falloc() that handles the race that another thread might have closed the
2396  * file descriptor out from under the thread creating the file object.
2397  */
2398 void
2399 fdclose(struct thread *td, struct file *fp, int idx)
2400 {
2401 	struct filedesc *fdp = td->td_proc->p_fd;
2402 
2403 	FILEDESC_XLOCK(fdp);
2404 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2405 		fdfree(fdp, idx);
2406 		FILEDESC_XUNLOCK(fdp);
2407 		fdrop(fp, td);
2408 	} else
2409 		FILEDESC_XUNLOCK(fdp);
2410 }
2411 
2412 /*
2413  * Close any files on exec?
2414  */
2415 void
2416 fdcloseexec(struct thread *td)
2417 {
2418 	struct filedesc *fdp;
2419 	struct filedescent *fde;
2420 	struct file *fp;
2421 	int i;
2422 
2423 	fdp = td->td_proc->p_fd;
2424 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2425 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2426 		fde = &fdp->fd_ofiles[i];
2427 		fp = fde->fde_file;
2428 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2429 		    (fde->fde_flags & UF_EXCLOSE))) {
2430 			FILEDESC_XLOCK(fdp);
2431 			fdfree(fdp, i);
2432 			(void) closefp(fdp, i, fp, td, 0);
2433 			FILEDESC_UNLOCK_ASSERT(fdp);
2434 		}
2435 	}
2436 }
2437 
2438 /*
2439  * It is unsafe for set[ug]id processes to be started with file
2440  * descriptors 0..2 closed, as these descriptors are given implicit
2441  * significance in the Standard C library.  fdcheckstd() will create a
2442  * descriptor referencing /dev/null for each of stdin, stdout, and
2443  * stderr that is not already open.
2444  */
2445 int
2446 fdcheckstd(struct thread *td)
2447 {
2448 	struct filedesc *fdp;
2449 	register_t save;
2450 	int i, error, devnull;
2451 
2452 	fdp = td->td_proc->p_fd;
2453 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2454 	MPASS(fdp->fd_nfiles >= 3);
2455 	devnull = -1;
2456 	for (i = 0; i <= 2; i++) {
2457 		if (fdp->fd_ofiles[i].fde_file != NULL)
2458 			continue;
2459 
2460 		save = td->td_retval[0];
2461 		if (devnull != -1) {
2462 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2463 		} else {
2464 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2465 			    UIO_SYSSPACE, O_RDWR, 0);
2466 			if (error == 0) {
2467 				devnull = td->td_retval[0];
2468 				KASSERT(devnull == i, ("we didn't get our fd"));
2469 			}
2470 		}
2471 		td->td_retval[0] = save;
2472 		if (error != 0)
2473 			return (error);
2474 	}
2475 	return (0);
2476 }
2477 
2478 /*
2479  * Internal form of close.  Decrement reference count on file structure.
2480  * Note: td may be NULL when closing a file that was being passed in a
2481  * message.
2482  */
2483 int
2484 closef(struct file *fp, struct thread *td)
2485 {
2486 	struct vnode *vp;
2487 	struct flock lf;
2488 	struct filedesc_to_leader *fdtol;
2489 	struct filedesc *fdp;
2490 
2491 	/*
2492 	 * POSIX record locking dictates that any close releases ALL
2493 	 * locks owned by this process.  This is handled by setting
2494 	 * a flag in the unlock to free ONLY locks obeying POSIX
2495 	 * semantics, and not to free BSD-style file locks.
2496 	 * If the descriptor was in a message, POSIX-style locks
2497 	 * aren't passed with the descriptor, and the thread pointer
2498 	 * will be NULL.  Callers should be careful only to pass a
2499 	 * NULL thread pointer when there really is no owning
2500 	 * context that might have locks, or the locks will be
2501 	 * leaked.
2502 	 */
2503 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2504 		vp = fp->f_vnode;
2505 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2506 			lf.l_whence = SEEK_SET;
2507 			lf.l_start = 0;
2508 			lf.l_len = 0;
2509 			lf.l_type = F_UNLCK;
2510 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2511 			    F_UNLCK, &lf, F_POSIX);
2512 		}
2513 		fdtol = td->td_proc->p_fdtol;
2514 		if (fdtol != NULL) {
2515 			/*
2516 			 * Handle special case where file descriptor table is
2517 			 * shared between multiple process leaders.
2518 			 */
2519 			fdp = td->td_proc->p_fd;
2520 			FILEDESC_XLOCK(fdp);
2521 			for (fdtol = fdtol->fdl_next;
2522 			    fdtol != td->td_proc->p_fdtol;
2523 			    fdtol = fdtol->fdl_next) {
2524 				if ((fdtol->fdl_leader->p_flag &
2525 				    P_ADVLOCK) == 0)
2526 					continue;
2527 				fdtol->fdl_holdcount++;
2528 				FILEDESC_XUNLOCK(fdp);
2529 				lf.l_whence = SEEK_SET;
2530 				lf.l_start = 0;
2531 				lf.l_len = 0;
2532 				lf.l_type = F_UNLCK;
2533 				vp = fp->f_vnode;
2534 				(void) VOP_ADVLOCK(vp,
2535 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2536 				    F_POSIX);
2537 				FILEDESC_XLOCK(fdp);
2538 				fdtol->fdl_holdcount--;
2539 				if (fdtol->fdl_holdcount == 0 &&
2540 				    fdtol->fdl_wakeup != 0) {
2541 					fdtol->fdl_wakeup = 0;
2542 					wakeup(fdtol);
2543 				}
2544 			}
2545 			FILEDESC_XUNLOCK(fdp);
2546 		}
2547 	}
2548 	return (fdrop(fp, td));
2549 }
2550 
2551 /*
2552  * Initialize the file pointer with the specified properties.
2553  *
2554  * The ops are set with release semantics to be certain that the flags, type,
2555  * and data are visible when ops is.  This is to prevent ops methods from being
2556  * called with bad data.
2557  */
2558 void
2559 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2560 {
2561 	fp->f_data = data;
2562 	fp->f_flag = flag;
2563 	fp->f_type = type;
2564 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2565 }
2566 
2567 int
2568 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2569     struct file **fpp, struct filecaps *havecapsp)
2570 {
2571 	struct filedescent *fde;
2572 	int error;
2573 
2574 	FILEDESC_LOCK_ASSERT(fdp);
2575 
2576 	fde = fdeget_locked(fdp, fd);
2577 	if (fde == NULL) {
2578 		error = EBADF;
2579 		goto out;
2580 	}
2581 
2582 #ifdef CAPABILITIES
2583 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2584 	if (error != 0)
2585 		goto out;
2586 #endif
2587 
2588 	if (havecapsp != NULL)
2589 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2590 
2591 	*fpp = fde->fde_file;
2592 
2593 	error = 0;
2594 out:
2595 	return (error);
2596 }
2597 
2598 int
2599 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2600     struct file **fpp, struct filecaps *havecapsp)
2601 {
2602 	struct filedesc *fdp = td->td_proc->p_fd;
2603 	int error;
2604 #ifndef CAPABILITIES
2605 	error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2606 	if (error == 0 && havecapsp != NULL)
2607 		filecaps_fill(havecapsp);
2608 #else
2609 	struct file *fp;
2610 	seqc_t seq;
2611 
2612 	for (;;) {
2613 		error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2614 		if (error != 0)
2615 			return (error);
2616 
2617 		if (havecapsp != NULL) {
2618 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2619 			    havecapsp, false)) {
2620 				fdrop(fp, td);
2621 				goto get_locked;
2622 			}
2623 		}
2624 
2625 		if (!fd_modified(fdp, fd, seq))
2626 			break;
2627 		fdrop(fp, td);
2628 	}
2629 
2630 	*fpp = fp;
2631 	return (0);
2632 
2633 get_locked:
2634 	FILEDESC_SLOCK(fdp);
2635 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2636 	if (error == 0 && !fhold(*fpp))
2637 		error = EBADF;
2638 	FILEDESC_SUNLOCK(fdp);
2639 #endif
2640 	return (error);
2641 }
2642 
2643 int
2644 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2645     struct file **fpp, seqc_t *seqp)
2646 {
2647 #ifdef CAPABILITIES
2648 	const struct filedescent *fde;
2649 #endif
2650 	const struct fdescenttbl *fdt;
2651 	struct file *fp;
2652 	u_int count;
2653 #ifdef CAPABILITIES
2654 	seqc_t seq;
2655 	cap_rights_t haverights;
2656 	int error;
2657 #endif
2658 
2659 	fdt = fdp->fd_files;
2660 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2661 		return (EBADF);
2662 	/*
2663 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2664 	 * never raising a refcount above 0.  To accomplish this we have
2665 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2666 	 * must be re-verified once we acquire a reference to be certain
2667 	 * that the identity is still correct and we did not lose a race
2668 	 * due to preemption.
2669 	 */
2670 	for (;;) {
2671 #ifdef CAPABILITIES
2672 		seq = seqc_read(fd_seqc(fdt, fd));
2673 		fde = &fdt->fdt_ofiles[fd];
2674 		haverights = *cap_rights_fde_inline(fde);
2675 		fp = fde->fde_file;
2676 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2677 			continue;
2678 #else
2679 		fp = fdt->fdt_ofiles[fd].fde_file;
2680 #endif
2681 		if (fp == NULL)
2682 			return (EBADF);
2683 #ifdef CAPABILITIES
2684 		error = cap_check(&haverights, needrightsp);
2685 		if (error != 0)
2686 			return (error);
2687 #endif
2688 		count = fp->f_count;
2689 	retry:
2690 		if (count == 0) {
2691 			/*
2692 			 * Force a reload. Other thread could reallocate the
2693 			 * table before this fd was closed, so it possible that
2694 			 * there is a stale fp pointer in cached version.
2695 			 */
2696 			fdt = *(const struct fdescenttbl * const volatile *)
2697 			    &(fdp->fd_files);
2698 			continue;
2699 		}
2700 		if (__predict_false(count + 1 < count))
2701 			return (EBADF);
2702 
2703 		/*
2704 		 * Use an acquire barrier to force re-reading of fdt so it is
2705 		 * refreshed for verification.
2706 		 */
2707 		if (__predict_false(atomic_fcmpset_acq_int(&fp->f_count,
2708 		    &count, count + 1) == 0))
2709 			goto retry;
2710 		fdt = fdp->fd_files;
2711 #ifdef	CAPABILITIES
2712 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2713 #else
2714 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2715 #endif
2716 			break;
2717 		fdrop(fp, curthread);
2718 	}
2719 	*fpp = fp;
2720 	if (seqp != NULL) {
2721 #ifdef CAPABILITIES
2722 		*seqp = seq;
2723 #endif
2724 	}
2725 	return (0);
2726 }
2727 
2728 /*
2729  * Extract the file pointer associated with the specified descriptor for the
2730  * current user process.
2731  *
2732  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2733  * returned.
2734  *
2735  * File's rights will be checked against the capability rights mask.
2736  *
2737  * If an error occurred the non-zero error is returned and *fpp is set to
2738  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2739  * responsible for fdrop().
2740  */
2741 static __inline int
2742 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2743     cap_rights_t *needrightsp, seqc_t *seqp)
2744 {
2745 	struct filedesc *fdp;
2746 	struct file *fp;
2747 	int error;
2748 
2749 	*fpp = NULL;
2750 	fdp = td->td_proc->p_fd;
2751 	error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2752 	if (error != 0)
2753 		return (error);
2754 	if (fp->f_ops == &badfileops) {
2755 		fdrop(fp, td);
2756 		return (EBADF);
2757 	}
2758 
2759 	/*
2760 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2761 	 */
2762 	error = 0;
2763 	switch (flags) {
2764 	case FREAD:
2765 	case FWRITE:
2766 		if ((fp->f_flag & flags) == 0)
2767 			error = EBADF;
2768 		break;
2769 	case FEXEC:
2770 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2771 		    ((fp->f_flag & FWRITE) != 0))
2772 			error = EBADF;
2773 		break;
2774 	case 0:
2775 		break;
2776 	default:
2777 		KASSERT(0, ("wrong flags"));
2778 	}
2779 
2780 	if (error != 0) {
2781 		fdrop(fp, td);
2782 		return (error);
2783 	}
2784 
2785 	*fpp = fp;
2786 	return (0);
2787 }
2788 
2789 int
2790 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2791 {
2792 
2793 	return (_fget(td, fd, fpp, 0, rightsp, NULL));
2794 }
2795 
2796 int
2797 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2798     struct file **fpp)
2799 {
2800 	int error;
2801 #ifndef CAPABILITIES
2802 	error = _fget(td, fd, fpp, 0, rightsp, NULL);
2803 	if (maxprotp != NULL)
2804 		*maxprotp = VM_PROT_ALL;
2805 #else
2806 	cap_rights_t fdrights;
2807 	struct filedesc *fdp = td->td_proc->p_fd;
2808 	seqc_t seq;
2809 
2810 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2811 	for (;;) {
2812 		error = _fget(td, fd, fpp, 0, rightsp, &seq);
2813 		if (error != 0)
2814 			return (error);
2815 		if (maxprotp != NULL)
2816 			fdrights = *cap_rights(fdp, fd);
2817 		if (!fd_modified(fdp, fd, seq))
2818 			break;
2819 		fdrop(*fpp, td);
2820 	}
2821 
2822 	/*
2823 	 * If requested, convert capability rights to access flags.
2824 	 */
2825 	if (maxprotp != NULL)
2826 		*maxprotp = cap_rights_to_vmprot(&fdrights);
2827 #endif
2828 	return (error);
2829 }
2830 
2831 int
2832 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2833 {
2834 
2835 	return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2836 }
2837 
2838 int
2839 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2840 {
2841 
2842 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2843 }
2844 
2845 int
2846 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2847     struct file **fpp)
2848 {
2849 	struct filedesc *fdp = td->td_proc->p_fd;
2850 #ifndef CAPABILITIES
2851 	return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2852 #else
2853 	int error;
2854 	seqc_t seq;
2855 
2856 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2857 	for (;;) {
2858 		error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2859 		if (error != 0)
2860 			return (error);
2861 		error = cap_fcntl_check(fdp, fd, needfcntl);
2862 		if (!fd_modified(fdp, fd, seq))
2863 			break;
2864 		fdrop(*fpp, td);
2865 	}
2866 	if (error != 0) {
2867 		fdrop(*fpp, td);
2868 		*fpp = NULL;
2869 	}
2870 	return (error);
2871 #endif
2872 }
2873 
2874 /*
2875  * Like fget() but loads the underlying vnode, or returns an error if the
2876  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2877  * never have VM objects.  The returned vnode will be vref()'d.
2878  *
2879  * XXX: what about the unused flags ?
2880  */
2881 static __inline int
2882 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2883     struct vnode **vpp)
2884 {
2885 	struct file *fp;
2886 	int error;
2887 
2888 	*vpp = NULL;
2889 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2890 	if (error != 0)
2891 		return (error);
2892 	if (fp->f_vnode == NULL) {
2893 		error = EINVAL;
2894 	} else {
2895 		*vpp = fp->f_vnode;
2896 		vrefact(*vpp);
2897 	}
2898 	fdrop(fp, td);
2899 
2900 	return (error);
2901 }
2902 
2903 int
2904 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2905 {
2906 
2907 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2908 }
2909 
2910 int
2911 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2912     struct filecaps *havecaps, struct vnode **vpp)
2913 {
2914 	struct filedesc *fdp;
2915 	struct filecaps caps;
2916 	struct file *fp;
2917 	int error;
2918 
2919 	fdp = td->td_proc->p_fd;
2920 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2921 	if (error != 0)
2922 		return (error);
2923 	if (fp->f_ops == &badfileops) {
2924 		error = EBADF;
2925 		goto out;
2926 	}
2927 	if (fp->f_vnode == NULL) {
2928 		error = EINVAL;
2929 		goto out;
2930 	}
2931 
2932 	*havecaps = caps;
2933 	*vpp = fp->f_vnode;
2934 	vrefact(*vpp);
2935 
2936 	return (0);
2937 out:
2938 	filecaps_free(&caps);
2939 	return (error);
2940 }
2941 
2942 int
2943 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2944 {
2945 
2946 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2947 }
2948 
2949 int
2950 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2951 {
2952 
2953 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2954 }
2955 
2956 #ifdef notyet
2957 int
2958 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2959     struct vnode **vpp)
2960 {
2961 
2962 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2963 }
2964 #endif
2965 
2966 /*
2967  * Handle the last reference to a file being closed.
2968  *
2969  * Without the noinline attribute clang keeps inlining the func thorough this
2970  * file when fdrop is used.
2971  */
2972 int __noinline
2973 _fdrop(struct file *fp, struct thread *td)
2974 {
2975 	int error;
2976 
2977 	if (fp->f_count != 0)
2978 		panic("fdrop: count %d", fp->f_count);
2979 	error = fo_close(fp, td);
2980 	atomic_subtract_int(&openfiles, 1);
2981 	crfree(fp->f_cred);
2982 	free(fp->f_advice, M_FADVISE);
2983 	uma_zfree(file_zone, fp);
2984 
2985 	return (error);
2986 }
2987 
2988 /*
2989  * Apply an advisory lock on a file descriptor.
2990  *
2991  * Just attempt to get a record lock of the requested type on the entire file
2992  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2993  */
2994 #ifndef _SYS_SYSPROTO_H_
2995 struct flock_args {
2996 	int	fd;
2997 	int	how;
2998 };
2999 #endif
3000 /* ARGSUSED */
3001 int
3002 sys_flock(struct thread *td, struct flock_args *uap)
3003 {
3004 	struct file *fp;
3005 	struct vnode *vp;
3006 	struct flock lf;
3007 	int error;
3008 
3009 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3010 	if (error != 0)
3011 		return (error);
3012 	if (fp->f_type != DTYPE_VNODE) {
3013 		fdrop(fp, td);
3014 		return (EOPNOTSUPP);
3015 	}
3016 
3017 	vp = fp->f_vnode;
3018 	lf.l_whence = SEEK_SET;
3019 	lf.l_start = 0;
3020 	lf.l_len = 0;
3021 	if (uap->how & LOCK_UN) {
3022 		lf.l_type = F_UNLCK;
3023 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3024 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3025 		goto done2;
3026 	}
3027 	if (uap->how & LOCK_EX)
3028 		lf.l_type = F_WRLCK;
3029 	else if (uap->how & LOCK_SH)
3030 		lf.l_type = F_RDLCK;
3031 	else {
3032 		error = EBADF;
3033 		goto done2;
3034 	}
3035 	atomic_set_int(&fp->f_flag, FHASLOCK);
3036 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3037 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3038 done2:
3039 	fdrop(fp, td);
3040 	return (error);
3041 }
3042 /*
3043  * Duplicate the specified descriptor to a free descriptor.
3044  */
3045 int
3046 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3047     int openerror, int *indxp)
3048 {
3049 	struct filedescent *newfde, *oldfde;
3050 	struct file *fp;
3051 	u_long *ioctls;
3052 	int error, indx;
3053 
3054 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3055 	    ("unexpected error %d in %s", openerror, __func__));
3056 
3057 	/*
3058 	 * If the to-be-dup'd fd number is greater than the allowed number
3059 	 * of file descriptors, or the fd to be dup'd has already been
3060 	 * closed, then reject.
3061 	 */
3062 	FILEDESC_XLOCK(fdp);
3063 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3064 		FILEDESC_XUNLOCK(fdp);
3065 		return (EBADF);
3066 	}
3067 
3068 	error = fdalloc(td, 0, &indx);
3069 	if (error != 0) {
3070 		FILEDESC_XUNLOCK(fdp);
3071 		return (error);
3072 	}
3073 
3074 	/*
3075 	 * There are two cases of interest here.
3076 	 *
3077 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3078 	 *
3079 	 * For ENXIO steal away the file structure from (dfd) and store it in
3080 	 * (indx).  (dfd) is effectively closed by this operation.
3081 	 */
3082 	switch (openerror) {
3083 	case ENODEV:
3084 		/*
3085 		 * Check that the mode the file is being opened for is a
3086 		 * subset of the mode of the existing descriptor.
3087 		 */
3088 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3089 			fdunused(fdp, indx);
3090 			FILEDESC_XUNLOCK(fdp);
3091 			return (EACCES);
3092 		}
3093 		if (!fhold(fp)) {
3094 			fdunused(fdp, indx);
3095 			FILEDESC_XUNLOCK(fdp);
3096 			return (EBADF);
3097 		}
3098 		newfde = &fdp->fd_ofiles[indx];
3099 		oldfde = &fdp->fd_ofiles[dfd];
3100 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3101 #ifdef CAPABILITIES
3102 		seqc_write_begin(&newfde->fde_seqc);
3103 #endif
3104 		memcpy(newfde, oldfde, fde_change_size);
3105 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3106 		    ioctls);
3107 #ifdef CAPABILITIES
3108 		seqc_write_end(&newfde->fde_seqc);
3109 #endif
3110 		break;
3111 	case ENXIO:
3112 		/*
3113 		 * Steal away the file pointer from dfd and stuff it into indx.
3114 		 */
3115 		newfde = &fdp->fd_ofiles[indx];
3116 		oldfde = &fdp->fd_ofiles[dfd];
3117 #ifdef CAPABILITIES
3118 		seqc_write_begin(&newfde->fde_seqc);
3119 #endif
3120 		memcpy(newfde, oldfde, fde_change_size);
3121 		oldfde->fde_file = NULL;
3122 		fdunused(fdp, dfd);
3123 #ifdef CAPABILITIES
3124 		seqc_write_end(&newfde->fde_seqc);
3125 #endif
3126 		break;
3127 	}
3128 	FILEDESC_XUNLOCK(fdp);
3129 	*indxp = indx;
3130 	return (0);
3131 }
3132 
3133 /*
3134  * This sysctl determines if we will allow a process to chroot(2) if it
3135  * has a directory open:
3136  *	0: disallowed for all processes.
3137  *	1: allowed for processes that were not already chroot(2)'ed.
3138  *	2: allowed for all processes.
3139  */
3140 
3141 static int chroot_allow_open_directories = 1;
3142 
3143 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3144     &chroot_allow_open_directories, 0,
3145     "Allow a process to chroot(2) if it has a directory open");
3146 
3147 /*
3148  * Helper function for raised chroot(2) security function:  Refuse if
3149  * any filedescriptors are open directories.
3150  */
3151 static int
3152 chroot_refuse_vdir_fds(struct filedesc *fdp)
3153 {
3154 	struct vnode *vp;
3155 	struct file *fp;
3156 	int fd;
3157 
3158 	FILEDESC_LOCK_ASSERT(fdp);
3159 
3160 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3161 		fp = fget_locked(fdp, fd);
3162 		if (fp == NULL)
3163 			continue;
3164 		if (fp->f_type == DTYPE_VNODE) {
3165 			vp = fp->f_vnode;
3166 			if (vp->v_type == VDIR)
3167 				return (EPERM);
3168 		}
3169 	}
3170 	return (0);
3171 }
3172 
3173 /*
3174  * Common routine for kern_chroot() and jail_attach().  The caller is
3175  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3176  * authorize this operation.
3177  */
3178 int
3179 pwd_chroot(struct thread *td, struct vnode *vp)
3180 {
3181 	struct filedesc *fdp;
3182 	struct vnode *oldvp;
3183 	int error;
3184 
3185 	fdp = td->td_proc->p_fd;
3186 	FILEDESC_XLOCK(fdp);
3187 	if (chroot_allow_open_directories == 0 ||
3188 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3189 		error = chroot_refuse_vdir_fds(fdp);
3190 		if (error != 0) {
3191 			FILEDESC_XUNLOCK(fdp);
3192 			return (error);
3193 		}
3194 	}
3195 	oldvp = fdp->fd_rdir;
3196 	vrefact(vp);
3197 	fdp->fd_rdir = vp;
3198 	if (fdp->fd_jdir == NULL) {
3199 		vrefact(vp);
3200 		fdp->fd_jdir = vp;
3201 	}
3202 	FILEDESC_XUNLOCK(fdp);
3203 	vrele(oldvp);
3204 	return (0);
3205 }
3206 
3207 void
3208 pwd_chdir(struct thread *td, struct vnode *vp)
3209 {
3210 	struct filedesc *fdp;
3211 	struct vnode *oldvp;
3212 
3213 	fdp = td->td_proc->p_fd;
3214 	FILEDESC_XLOCK(fdp);
3215 	VNASSERT(vp->v_usecount > 0, vp,
3216 	    ("chdir to a vnode with zero usecount"));
3217 	oldvp = fdp->fd_cdir;
3218 	fdp->fd_cdir = vp;
3219 	FILEDESC_XUNLOCK(fdp);
3220 	vrele(oldvp);
3221 }
3222 
3223 /*
3224  * Scan all active processes and prisons to see if any of them have a current
3225  * or root directory of `olddp'. If so, replace them with the new mount point.
3226  */
3227 void
3228 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3229 {
3230 	struct filedesc *fdp;
3231 	struct prison *pr;
3232 	struct proc *p;
3233 	int nrele;
3234 
3235 	if (vrefcnt(olddp) == 1)
3236 		return;
3237 	nrele = 0;
3238 	sx_slock(&allproc_lock);
3239 	FOREACH_PROC_IN_SYSTEM(p) {
3240 		PROC_LOCK(p);
3241 		fdp = fdhold(p);
3242 		PROC_UNLOCK(p);
3243 		if (fdp == NULL)
3244 			continue;
3245 		FILEDESC_XLOCK(fdp);
3246 		if (fdp->fd_cdir == olddp) {
3247 			vrefact(newdp);
3248 			fdp->fd_cdir = newdp;
3249 			nrele++;
3250 		}
3251 		if (fdp->fd_rdir == olddp) {
3252 			vrefact(newdp);
3253 			fdp->fd_rdir = newdp;
3254 			nrele++;
3255 		}
3256 		if (fdp->fd_jdir == olddp) {
3257 			vrefact(newdp);
3258 			fdp->fd_jdir = newdp;
3259 			nrele++;
3260 		}
3261 		FILEDESC_XUNLOCK(fdp);
3262 		fddrop(fdp);
3263 	}
3264 	sx_sunlock(&allproc_lock);
3265 	if (rootvnode == olddp) {
3266 		vrefact(newdp);
3267 		rootvnode = newdp;
3268 		nrele++;
3269 	}
3270 	mtx_lock(&prison0.pr_mtx);
3271 	if (prison0.pr_root == olddp) {
3272 		vrefact(newdp);
3273 		prison0.pr_root = newdp;
3274 		nrele++;
3275 	}
3276 	mtx_unlock(&prison0.pr_mtx);
3277 	sx_slock(&allprison_lock);
3278 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3279 		mtx_lock(&pr->pr_mtx);
3280 		if (pr->pr_root == olddp) {
3281 			vrefact(newdp);
3282 			pr->pr_root = newdp;
3283 			nrele++;
3284 		}
3285 		mtx_unlock(&pr->pr_mtx);
3286 	}
3287 	sx_sunlock(&allprison_lock);
3288 	while (nrele--)
3289 		vrele(olddp);
3290 }
3291 
3292 struct filedesc_to_leader *
3293 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3294 {
3295 	struct filedesc_to_leader *fdtol;
3296 
3297 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3298 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3299 	fdtol->fdl_refcount = 1;
3300 	fdtol->fdl_holdcount = 0;
3301 	fdtol->fdl_wakeup = 0;
3302 	fdtol->fdl_leader = leader;
3303 	if (old != NULL) {
3304 		FILEDESC_XLOCK(fdp);
3305 		fdtol->fdl_next = old->fdl_next;
3306 		fdtol->fdl_prev = old;
3307 		old->fdl_next = fdtol;
3308 		fdtol->fdl_next->fdl_prev = fdtol;
3309 		FILEDESC_XUNLOCK(fdp);
3310 	} else {
3311 		fdtol->fdl_next = fdtol;
3312 		fdtol->fdl_prev = fdtol;
3313 	}
3314 	return (fdtol);
3315 }
3316 
3317 static int
3318 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3319 {
3320 	struct filedesc *fdp;
3321 	int i, count, slots;
3322 
3323 	if (*(int *)arg1 != 0)
3324 		return (EINVAL);
3325 
3326 	fdp = curproc->p_fd;
3327 	count = 0;
3328 	FILEDESC_SLOCK(fdp);
3329 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3330 	for (i = 0; i < slots; i++)
3331 		count += bitcountl(fdp->fd_map[i]);
3332 	FILEDESC_SUNLOCK(fdp);
3333 
3334 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3335 }
3336 
3337 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3338     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3339     "Number of open file descriptors");
3340 
3341 /*
3342  * Get file structures globally.
3343  */
3344 static int
3345 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3346 {
3347 	struct xfile xf;
3348 	struct filedesc *fdp;
3349 	struct file *fp;
3350 	struct proc *p;
3351 	int error, n;
3352 
3353 	error = sysctl_wire_old_buffer(req, 0);
3354 	if (error != 0)
3355 		return (error);
3356 	if (req->oldptr == NULL) {
3357 		n = 0;
3358 		sx_slock(&allproc_lock);
3359 		FOREACH_PROC_IN_SYSTEM(p) {
3360 			PROC_LOCK(p);
3361 			if (p->p_state == PRS_NEW) {
3362 				PROC_UNLOCK(p);
3363 				continue;
3364 			}
3365 			fdp = fdhold(p);
3366 			PROC_UNLOCK(p);
3367 			if (fdp == NULL)
3368 				continue;
3369 			/* overestimates sparse tables. */
3370 			if (fdp->fd_lastfile > 0)
3371 				n += fdp->fd_lastfile;
3372 			fddrop(fdp);
3373 		}
3374 		sx_sunlock(&allproc_lock);
3375 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3376 	}
3377 	error = 0;
3378 	bzero(&xf, sizeof(xf));
3379 	xf.xf_size = sizeof(xf);
3380 	sx_slock(&allproc_lock);
3381 	FOREACH_PROC_IN_SYSTEM(p) {
3382 		PROC_LOCK(p);
3383 		if (p->p_state == PRS_NEW) {
3384 			PROC_UNLOCK(p);
3385 			continue;
3386 		}
3387 		if (p_cansee(req->td, p) != 0) {
3388 			PROC_UNLOCK(p);
3389 			continue;
3390 		}
3391 		xf.xf_pid = p->p_pid;
3392 		xf.xf_uid = p->p_ucred->cr_uid;
3393 		fdp = fdhold(p);
3394 		PROC_UNLOCK(p);
3395 		if (fdp == NULL)
3396 			continue;
3397 		FILEDESC_SLOCK(fdp);
3398 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3399 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3400 				continue;
3401 			xf.xf_fd = n;
3402 			xf.xf_file = (uintptr_t)fp;
3403 			xf.xf_data = (uintptr_t)fp->f_data;
3404 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3405 			xf.xf_type = (uintptr_t)fp->f_type;
3406 			xf.xf_count = fp->f_count;
3407 			xf.xf_msgcount = 0;
3408 			xf.xf_offset = foffset_get(fp);
3409 			xf.xf_flag = fp->f_flag;
3410 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3411 			if (error)
3412 				break;
3413 		}
3414 		FILEDESC_SUNLOCK(fdp);
3415 		fddrop(fdp);
3416 		if (error)
3417 			break;
3418 	}
3419 	sx_sunlock(&allproc_lock);
3420 	return (error);
3421 }
3422 
3423 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3424     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3425 
3426 #ifdef KINFO_FILE_SIZE
3427 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3428 #endif
3429 
3430 static int
3431 xlate_fflags(int fflags)
3432 {
3433 	static const struct {
3434 		int	fflag;
3435 		int	kf_fflag;
3436 	} fflags_table[] = {
3437 		{ FAPPEND, KF_FLAG_APPEND },
3438 		{ FASYNC, KF_FLAG_ASYNC },
3439 		{ FFSYNC, KF_FLAG_FSYNC },
3440 		{ FHASLOCK, KF_FLAG_HASLOCK },
3441 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3442 		{ FREAD, KF_FLAG_READ },
3443 		{ FWRITE, KF_FLAG_WRITE },
3444 		{ O_CREAT, KF_FLAG_CREAT },
3445 		{ O_DIRECT, KF_FLAG_DIRECT },
3446 		{ O_EXCL, KF_FLAG_EXCL },
3447 		{ O_EXEC, KF_FLAG_EXEC },
3448 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3449 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3450 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3451 		{ O_TRUNC, KF_FLAG_TRUNC }
3452 	};
3453 	unsigned int i;
3454 	int kflags;
3455 
3456 	kflags = 0;
3457 	for (i = 0; i < nitems(fflags_table); i++)
3458 		if (fflags & fflags_table[i].fflag)
3459 			kflags |=  fflags_table[i].kf_fflag;
3460 	return (kflags);
3461 }
3462 
3463 /* Trim unused data from kf_path by truncating the structure size. */
3464 void
3465 pack_kinfo(struct kinfo_file *kif)
3466 {
3467 
3468 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3469 	    strlen(kif->kf_path) + 1;
3470 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3471 }
3472 
3473 static void
3474 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3475     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3476 {
3477 	int error;
3478 
3479 	bzero(kif, sizeof(*kif));
3480 
3481 	/* Set a default type to allow for empty fill_kinfo() methods. */
3482 	kif->kf_type = KF_TYPE_UNKNOWN;
3483 	kif->kf_flags = xlate_fflags(fp->f_flag);
3484 	if (rightsp != NULL)
3485 		kif->kf_cap_rights = *rightsp;
3486 	else
3487 		cap_rights_init(&kif->kf_cap_rights);
3488 	kif->kf_fd = fd;
3489 	kif->kf_ref_count = fp->f_count;
3490 	kif->kf_offset = foffset_get(fp);
3491 
3492 	/*
3493 	 * This may drop the filedesc lock, so the 'fp' cannot be
3494 	 * accessed after this call.
3495 	 */
3496 	error = fo_fill_kinfo(fp, kif, fdp);
3497 	if (error == 0)
3498 		kif->kf_status |= KF_ATTR_VALID;
3499 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3500 		pack_kinfo(kif);
3501 	else
3502 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3503 }
3504 
3505 static void
3506 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3507     struct kinfo_file *kif, int flags)
3508 {
3509 	int error;
3510 
3511 	bzero(kif, sizeof(*kif));
3512 
3513 	kif->kf_type = KF_TYPE_VNODE;
3514 	error = vn_fill_kinfo_vnode(vp, kif);
3515 	if (error == 0)
3516 		kif->kf_status |= KF_ATTR_VALID;
3517 	kif->kf_flags = xlate_fflags(fflags);
3518 	cap_rights_init(&kif->kf_cap_rights);
3519 	kif->kf_fd = fd;
3520 	kif->kf_ref_count = -1;
3521 	kif->kf_offset = -1;
3522 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3523 		pack_kinfo(kif);
3524 	else
3525 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3526 	vrele(vp);
3527 }
3528 
3529 struct export_fd_buf {
3530 	struct filedesc		*fdp;
3531 	struct sbuf 		*sb;
3532 	ssize_t			remainder;
3533 	struct kinfo_file	kif;
3534 	int			flags;
3535 };
3536 
3537 static int
3538 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3539 {
3540 	struct kinfo_file *kif;
3541 
3542 	kif = &efbuf->kif;
3543 	if (efbuf->remainder != -1) {
3544 		if (efbuf->remainder < kif->kf_structsize) {
3545 			/* Terminate export. */
3546 			efbuf->remainder = 0;
3547 			return (0);
3548 		}
3549 		efbuf->remainder -= kif->kf_structsize;
3550 	}
3551 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3552 }
3553 
3554 static int
3555 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3556     struct export_fd_buf *efbuf)
3557 {
3558 	int error;
3559 
3560 	if (efbuf->remainder == 0)
3561 		return (0);
3562 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3563 	    efbuf->flags);
3564 	FILEDESC_SUNLOCK(efbuf->fdp);
3565 	error = export_kinfo_to_sb(efbuf);
3566 	FILEDESC_SLOCK(efbuf->fdp);
3567 	return (error);
3568 }
3569 
3570 static int
3571 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3572     struct export_fd_buf *efbuf)
3573 {
3574 	int error;
3575 
3576 	if (efbuf->remainder == 0)
3577 		return (0);
3578 	if (efbuf->fdp != NULL)
3579 		FILEDESC_SUNLOCK(efbuf->fdp);
3580 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3581 	error = export_kinfo_to_sb(efbuf);
3582 	if (efbuf->fdp != NULL)
3583 		FILEDESC_SLOCK(efbuf->fdp);
3584 	return (error);
3585 }
3586 
3587 /*
3588  * Store a process file descriptor information to sbuf.
3589  *
3590  * Takes a locked proc as argument, and returns with the proc unlocked.
3591  */
3592 int
3593 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3594     int flags)
3595 {
3596 	struct file *fp;
3597 	struct filedesc *fdp;
3598 	struct export_fd_buf *efbuf;
3599 	struct vnode *cttyvp, *textvp, *tracevp;
3600 	int error, i;
3601 	cap_rights_t rights;
3602 
3603 	PROC_LOCK_ASSERT(p, MA_OWNED);
3604 
3605 	/* ktrace vnode */
3606 	tracevp = p->p_tracevp;
3607 	if (tracevp != NULL)
3608 		vrefact(tracevp);
3609 	/* text vnode */
3610 	textvp = p->p_textvp;
3611 	if (textvp != NULL)
3612 		vrefact(textvp);
3613 	/* Controlling tty. */
3614 	cttyvp = NULL;
3615 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3616 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3617 		if (cttyvp != NULL)
3618 			vrefact(cttyvp);
3619 	}
3620 	fdp = fdhold(p);
3621 	PROC_UNLOCK(p);
3622 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3623 	efbuf->fdp = NULL;
3624 	efbuf->sb = sb;
3625 	efbuf->remainder = maxlen;
3626 	efbuf->flags = flags;
3627 	if (tracevp != NULL)
3628 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3629 		    efbuf);
3630 	if (textvp != NULL)
3631 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3632 	if (cttyvp != NULL)
3633 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3634 		    efbuf);
3635 	error = 0;
3636 	if (fdp == NULL)
3637 		goto fail;
3638 	efbuf->fdp = fdp;
3639 	FILEDESC_SLOCK(fdp);
3640 	/* working directory */
3641 	if (fdp->fd_cdir != NULL) {
3642 		vrefact(fdp->fd_cdir);
3643 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3644 	}
3645 	/* root directory */
3646 	if (fdp->fd_rdir != NULL) {
3647 		vrefact(fdp->fd_rdir);
3648 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3649 	}
3650 	/* jail directory */
3651 	if (fdp->fd_jdir != NULL) {
3652 		vrefact(fdp->fd_jdir);
3653 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3654 	}
3655 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3656 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3657 			continue;
3658 #ifdef CAPABILITIES
3659 		rights = *cap_rights(fdp, i);
3660 #else /* !CAPABILITIES */
3661 		rights = cap_no_rights;
3662 #endif
3663 		/*
3664 		 * Create sysctl entry.  It is OK to drop the filedesc
3665 		 * lock inside of export_file_to_sb() as we will
3666 		 * re-validate and re-evaluate its properties when the
3667 		 * loop continues.
3668 		 */
3669 		error = export_file_to_sb(fp, i, &rights, efbuf);
3670 		if (error != 0 || efbuf->remainder == 0)
3671 			break;
3672 	}
3673 	FILEDESC_SUNLOCK(fdp);
3674 	fddrop(fdp);
3675 fail:
3676 	free(efbuf, M_TEMP);
3677 	return (error);
3678 }
3679 
3680 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3681 
3682 /*
3683  * Get per-process file descriptors for use by procstat(1), et al.
3684  */
3685 static int
3686 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3687 {
3688 	struct sbuf sb;
3689 	struct proc *p;
3690 	ssize_t maxlen;
3691 	int error, error2, *name;
3692 
3693 	name = (int *)arg1;
3694 
3695 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3696 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3697 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3698 	if (error != 0) {
3699 		sbuf_delete(&sb);
3700 		return (error);
3701 	}
3702 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3703 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3704 	    KERN_FILEDESC_PACK_KINFO);
3705 	error2 = sbuf_finish(&sb);
3706 	sbuf_delete(&sb);
3707 	return (error != 0 ? error : error2);
3708 }
3709 
3710 #ifdef COMPAT_FREEBSD7
3711 #ifdef KINFO_OFILE_SIZE
3712 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3713 #endif
3714 
3715 static void
3716 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3717 {
3718 
3719 	okif->kf_structsize = sizeof(*okif);
3720 	okif->kf_type = kif->kf_type;
3721 	okif->kf_fd = kif->kf_fd;
3722 	okif->kf_ref_count = kif->kf_ref_count;
3723 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3724 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3725 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3726 	okif->kf_offset = kif->kf_offset;
3727 	if (kif->kf_type == KF_TYPE_VNODE)
3728 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3729 	else
3730 		okif->kf_vnode_type = KF_VTYPE_VNON;
3731 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3732 	if (kif->kf_type == KF_TYPE_SOCKET) {
3733 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3734 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3735 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3736 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3737 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3738 	} else {
3739 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3740 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3741 	}
3742 }
3743 
3744 static int
3745 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3746     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3747 {
3748 	int error;
3749 
3750 	vrefact(vp);
3751 	FILEDESC_SUNLOCK(fdp);
3752 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3753 	kinfo_to_okinfo(kif, okif);
3754 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3755 	FILEDESC_SLOCK(fdp);
3756 	return (error);
3757 }
3758 
3759 /*
3760  * Get per-process file descriptors for use by procstat(1), et al.
3761  */
3762 static int
3763 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3764 {
3765 	struct kinfo_ofile *okif;
3766 	struct kinfo_file *kif;
3767 	struct filedesc *fdp;
3768 	int error, i, *name;
3769 	struct file *fp;
3770 	struct proc *p;
3771 
3772 	name = (int *)arg1;
3773 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3774 	if (error != 0)
3775 		return (error);
3776 	fdp = fdhold(p);
3777 	PROC_UNLOCK(p);
3778 	if (fdp == NULL)
3779 		return (ENOENT);
3780 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3781 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3782 	FILEDESC_SLOCK(fdp);
3783 	if (fdp->fd_cdir != NULL)
3784 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3785 		    okif, fdp, req);
3786 	if (fdp->fd_rdir != NULL)
3787 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3788 		    okif, fdp, req);
3789 	if (fdp->fd_jdir != NULL)
3790 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3791 		    okif, fdp, req);
3792 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3793 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3794 			continue;
3795 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3796 		    KERN_FILEDESC_PACK_KINFO);
3797 		FILEDESC_SUNLOCK(fdp);
3798 		kinfo_to_okinfo(kif, okif);
3799 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3800 		FILEDESC_SLOCK(fdp);
3801 		if (error)
3802 			break;
3803 	}
3804 	FILEDESC_SUNLOCK(fdp);
3805 	fddrop(fdp);
3806 	free(kif, M_TEMP);
3807 	free(okif, M_TEMP);
3808 	return (0);
3809 }
3810 
3811 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3812     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3813     "Process ofiledesc entries");
3814 #endif	/* COMPAT_FREEBSD7 */
3815 
3816 int
3817 vntype_to_kinfo(int vtype)
3818 {
3819 	struct {
3820 		int	vtype;
3821 		int	kf_vtype;
3822 	} vtypes_table[] = {
3823 		{ VBAD, KF_VTYPE_VBAD },
3824 		{ VBLK, KF_VTYPE_VBLK },
3825 		{ VCHR, KF_VTYPE_VCHR },
3826 		{ VDIR, KF_VTYPE_VDIR },
3827 		{ VFIFO, KF_VTYPE_VFIFO },
3828 		{ VLNK, KF_VTYPE_VLNK },
3829 		{ VNON, KF_VTYPE_VNON },
3830 		{ VREG, KF_VTYPE_VREG },
3831 		{ VSOCK, KF_VTYPE_VSOCK }
3832 	};
3833 	unsigned int i;
3834 
3835 	/*
3836 	 * Perform vtype translation.
3837 	 */
3838 	for (i = 0; i < nitems(vtypes_table); i++)
3839 		if (vtypes_table[i].vtype == vtype)
3840 			return (vtypes_table[i].kf_vtype);
3841 
3842 	return (KF_VTYPE_UNKNOWN);
3843 }
3844 
3845 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3846     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3847     "Process filedesc entries");
3848 
3849 /*
3850  * Store a process current working directory information to sbuf.
3851  *
3852  * Takes a locked proc as argument, and returns with the proc unlocked.
3853  */
3854 int
3855 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3856 {
3857 	struct filedesc *fdp;
3858 	struct export_fd_buf *efbuf;
3859 	int error;
3860 
3861 	PROC_LOCK_ASSERT(p, MA_OWNED);
3862 
3863 	fdp = fdhold(p);
3864 	PROC_UNLOCK(p);
3865 	if (fdp == NULL)
3866 		return (EINVAL);
3867 
3868 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3869 	efbuf->fdp = fdp;
3870 	efbuf->sb = sb;
3871 	efbuf->remainder = maxlen;
3872 
3873 	FILEDESC_SLOCK(fdp);
3874 	if (fdp->fd_cdir == NULL)
3875 		error = EINVAL;
3876 	else {
3877 		vrefact(fdp->fd_cdir);
3878 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3879 		    FREAD, efbuf);
3880 	}
3881 	FILEDESC_SUNLOCK(fdp);
3882 	fddrop(fdp);
3883 	free(efbuf, M_TEMP);
3884 	return (error);
3885 }
3886 
3887 /*
3888  * Get per-process current working directory.
3889  */
3890 static int
3891 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3892 {
3893 	struct sbuf sb;
3894 	struct proc *p;
3895 	ssize_t maxlen;
3896 	int error, error2, *name;
3897 
3898 	name = (int *)arg1;
3899 
3900 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3901 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3902 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3903 	if (error != 0) {
3904 		sbuf_delete(&sb);
3905 		return (error);
3906 	}
3907 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3908 	error = kern_proc_cwd_out(p, &sb, maxlen);
3909 	error2 = sbuf_finish(&sb);
3910 	sbuf_delete(&sb);
3911 	return (error != 0 ? error : error2);
3912 }
3913 
3914 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3915     sysctl_kern_proc_cwd, "Process current working directory");
3916 
3917 #ifdef DDB
3918 /*
3919  * For the purposes of debugging, generate a human-readable string for the
3920  * file type.
3921  */
3922 static const char *
3923 file_type_to_name(short type)
3924 {
3925 
3926 	switch (type) {
3927 	case 0:
3928 		return ("zero");
3929 	case DTYPE_VNODE:
3930 		return ("vnode");
3931 	case DTYPE_SOCKET:
3932 		return ("socket");
3933 	case DTYPE_PIPE:
3934 		return ("pipe");
3935 	case DTYPE_FIFO:
3936 		return ("fifo");
3937 	case DTYPE_KQUEUE:
3938 		return ("kqueue");
3939 	case DTYPE_CRYPTO:
3940 		return ("crypto");
3941 	case DTYPE_MQUEUE:
3942 		return ("mqueue");
3943 	case DTYPE_SHM:
3944 		return ("shm");
3945 	case DTYPE_SEM:
3946 		return ("ksem");
3947 	case DTYPE_PTS:
3948 		return ("pts");
3949 	case DTYPE_DEV:
3950 		return ("dev");
3951 	case DTYPE_PROCDESC:
3952 		return ("proc");
3953 	case DTYPE_LINUXEFD:
3954 		return ("levent");
3955 	case DTYPE_LINUXTFD:
3956 		return ("ltimer");
3957 	default:
3958 		return ("unkn");
3959 	}
3960 }
3961 
3962 /*
3963  * For the purposes of debugging, identify a process (if any, perhaps one of
3964  * many) that references the passed file in its file descriptor array. Return
3965  * NULL if none.
3966  */
3967 static struct proc *
3968 file_to_first_proc(struct file *fp)
3969 {
3970 	struct filedesc *fdp;
3971 	struct proc *p;
3972 	int n;
3973 
3974 	FOREACH_PROC_IN_SYSTEM(p) {
3975 		if (p->p_state == PRS_NEW)
3976 			continue;
3977 		fdp = p->p_fd;
3978 		if (fdp == NULL)
3979 			continue;
3980 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3981 			if (fp == fdp->fd_ofiles[n].fde_file)
3982 				return (p);
3983 		}
3984 	}
3985 	return (NULL);
3986 }
3987 
3988 static void
3989 db_print_file(struct file *fp, int header)
3990 {
3991 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3992 	struct proc *p;
3993 
3994 	if (header)
3995 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3996 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3997 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3998 		    "FCmd");
3999 	p = file_to_first_proc(fp);
4000 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4001 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4002 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4003 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4004 
4005 #undef XPTRWIDTH
4006 }
4007 
4008 DB_SHOW_COMMAND(file, db_show_file)
4009 {
4010 	struct file *fp;
4011 
4012 	if (!have_addr) {
4013 		db_printf("usage: show file <addr>\n");
4014 		return;
4015 	}
4016 	fp = (struct file *)addr;
4017 	db_print_file(fp, 1);
4018 }
4019 
4020 DB_SHOW_COMMAND(files, db_show_files)
4021 {
4022 	struct filedesc *fdp;
4023 	struct file *fp;
4024 	struct proc *p;
4025 	int header;
4026 	int n;
4027 
4028 	header = 1;
4029 	FOREACH_PROC_IN_SYSTEM(p) {
4030 		if (p->p_state == PRS_NEW)
4031 			continue;
4032 		if ((fdp = p->p_fd) == NULL)
4033 			continue;
4034 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
4035 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4036 				continue;
4037 			db_print_file(fp, header);
4038 			header = 0;
4039 		}
4040 	}
4041 }
4042 #endif
4043 
4044 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4045     &maxfilesperproc, 0, "Maximum files allowed open per process");
4046 
4047 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4048     &maxfiles, 0, "Maximum number of files");
4049 
4050 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4051     &openfiles, 0, "System-wide number of open files");
4052 
4053 /* ARGSUSED*/
4054 static void
4055 filelistinit(void *dummy)
4056 {
4057 
4058 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4059 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4060 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4061 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4062 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4063 }
4064 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4065 
4066 /*-------------------------------------------------------------------*/
4067 
4068 static int
4069 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4070     int flags, struct thread *td)
4071 {
4072 
4073 	return (EBADF);
4074 }
4075 
4076 static int
4077 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4078     struct thread *td)
4079 {
4080 
4081 	return (EINVAL);
4082 }
4083 
4084 static int
4085 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4086     struct thread *td)
4087 {
4088 
4089 	return (EBADF);
4090 }
4091 
4092 static int
4093 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4094     struct thread *td)
4095 {
4096 
4097 	return (0);
4098 }
4099 
4100 static int
4101 badfo_kqfilter(struct file *fp, struct knote *kn)
4102 {
4103 
4104 	return (EBADF);
4105 }
4106 
4107 static int
4108 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4109     struct thread *td)
4110 {
4111 
4112 	return (EBADF);
4113 }
4114 
4115 static int
4116 badfo_close(struct file *fp, struct thread *td)
4117 {
4118 
4119 	return (0);
4120 }
4121 
4122 static int
4123 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4124     struct thread *td)
4125 {
4126 
4127 	return (EBADF);
4128 }
4129 
4130 static int
4131 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4132     struct thread *td)
4133 {
4134 
4135 	return (EBADF);
4136 }
4137 
4138 static int
4139 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4140     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4141     struct thread *td)
4142 {
4143 
4144 	return (EBADF);
4145 }
4146 
4147 static int
4148 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4149 {
4150 
4151 	return (0);
4152 }
4153 
4154 struct fileops badfileops = {
4155 	.fo_read = badfo_readwrite,
4156 	.fo_write = badfo_readwrite,
4157 	.fo_truncate = badfo_truncate,
4158 	.fo_ioctl = badfo_ioctl,
4159 	.fo_poll = badfo_poll,
4160 	.fo_kqfilter = badfo_kqfilter,
4161 	.fo_stat = badfo_stat,
4162 	.fo_close = badfo_close,
4163 	.fo_chmod = badfo_chmod,
4164 	.fo_chown = badfo_chown,
4165 	.fo_sendfile = badfo_sendfile,
4166 	.fo_fill_kinfo = badfo_fill_kinfo,
4167 };
4168 
4169 int
4170 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4171     int flags, struct thread *td)
4172 {
4173 
4174 	return (EOPNOTSUPP);
4175 }
4176 
4177 int
4178 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4179     struct thread *td)
4180 {
4181 
4182 	return (EINVAL);
4183 }
4184 
4185 int
4186 invfo_ioctl(struct file *fp, u_long com, void *data,
4187     struct ucred *active_cred, struct thread *td)
4188 {
4189 
4190 	return (ENOTTY);
4191 }
4192 
4193 int
4194 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4195     struct thread *td)
4196 {
4197 
4198 	return (poll_no_poll(events));
4199 }
4200 
4201 int
4202 invfo_kqfilter(struct file *fp, struct knote *kn)
4203 {
4204 
4205 	return (EINVAL);
4206 }
4207 
4208 int
4209 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4210     struct thread *td)
4211 {
4212 
4213 	return (EINVAL);
4214 }
4215 
4216 int
4217 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4218     struct thread *td)
4219 {
4220 
4221 	return (EINVAL);
4222 }
4223 
4224 int
4225 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4226     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4227     struct thread *td)
4228 {
4229 
4230 	return (EINVAL);
4231 }
4232 
4233 /*-------------------------------------------------------------------*/
4234 
4235 /*
4236  * File Descriptor pseudo-device driver (/dev/fd/).
4237  *
4238  * Opening minor device N dup()s the file (if any) connected to file
4239  * descriptor N belonging to the calling process.  Note that this driver
4240  * consists of only the ``open()'' routine, because all subsequent
4241  * references to this file will be direct to the other driver.
4242  *
4243  * XXX: we could give this one a cloning event handler if necessary.
4244  */
4245 
4246 /* ARGSUSED */
4247 static int
4248 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4249 {
4250 
4251 	/*
4252 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4253 	 * the file descriptor being sought for duplication. The error
4254 	 * return ensures that the vnode for this device will be released
4255 	 * by vn_open. Open will detect this special error and take the
4256 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4257 	 * will simply report the error.
4258 	 */
4259 	td->td_dupfd = dev2unit(dev);
4260 	return (ENODEV);
4261 }
4262 
4263 static struct cdevsw fildesc_cdevsw = {
4264 	.d_version =	D_VERSION,
4265 	.d_open =	fdopen,
4266 	.d_name =	"FD",
4267 };
4268 
4269 static void
4270 fildesc_drvinit(void *unused)
4271 {
4272 	struct cdev *dev;
4273 
4274 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4275 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4276 	make_dev_alias(dev, "stdin");
4277 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4278 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4279 	make_dev_alias(dev, "stdout");
4280 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4281 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4282 	make_dev_alias(dev, "stderr");
4283 }
4284 
4285 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4286