xref: /freebsd/sys/kern/kern_descrip.c (revision 2c52512caf6ec10f49038e6884b9c2aea905cc4e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98     "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 
102 MALLOC_DECLARE(M_FADVISE);
103 
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
107 VFS_SMR_DECLARE;
108 
109 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
110 		    struct thread *td, bool holdleaders, bool audit);
111 static int	fd_first_free(struct filedesc *fdp, int low, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the last used fd.
217  *
218  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219  * Otherwise use fdlastfile.
220  */
221 int
222 fdlastfile_single(struct filedesc *fdp)
223 {
224 	NDSLOTTYPE *map = fdp->fd_map;
225 	int off, minoff;
226 
227 	off = NDSLOT(fdp->fd_nfiles - 1);
228 	for (minoff = NDSLOT(0); off >= minoff; --off)
229 		if (map[off] != 0)
230 			return (off * NDENTRIES + flsl(map[off]) - 1);
231 	return (-1);
232 }
233 
234 int
235 fdlastfile(struct filedesc *fdp)
236 {
237 
238 	FILEDESC_LOCK_ASSERT(fdp);
239 	return (fdlastfile_single(fdp));
240 }
241 
242 static int
243 fdisused(struct filedesc *fdp, int fd)
244 {
245 
246 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 
249 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
250 }
251 
252 /*
253  * Mark a file descriptor as used.
254  */
255 static void
256 fdused_init(struct filedesc *fdp, int fd)
257 {
258 
259 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 
261 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
262 }
263 
264 static void
265 fdused(struct filedesc *fdp, int fd)
266 {
267 
268 	FILEDESC_XLOCK_ASSERT(fdp);
269 
270 	fdused_init(fdp, fd);
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile++;
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	FILEDESC_XLOCK_ASSERT(fdp);
311 	fde = &fdp->fd_ofiles[fd];
312 #ifdef CAPABILITIES
313 	seqc_write_begin(&fde->fde_seqc);
314 #endif
315 	fde->fde_file = NULL;
316 #ifdef CAPABILITIES
317 	seqc_write_end(&fde->fde_seqc);
318 #endif
319 	fdefree_last(fde);
320 	fdunused(fdp, fd);
321 }
322 
323 /*
324  * System calls on descriptors.
325  */
326 #ifndef _SYS_SYSPROTO_H_
327 struct getdtablesize_args {
328 	int	dummy;
329 };
330 #endif
331 /* ARGSUSED */
332 int
333 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
334 {
335 #ifdef	RACCT
336 	uint64_t lim;
337 #endif
338 
339 	td->td_retval[0] = getmaxfd(td);
340 #ifdef	RACCT
341 	PROC_LOCK(td->td_proc);
342 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
343 	PROC_UNLOCK(td->td_proc);
344 	if (lim < td->td_retval[0])
345 		td->td_retval[0] = lim;
346 #endif
347 	return (0);
348 }
349 
350 /*
351  * Duplicate a file descriptor to a particular value.
352  *
353  * Note: keep in mind that a potential race condition exists when closing
354  * descriptors from a shared descriptor table (via rfork).
355  */
356 #ifndef _SYS_SYSPROTO_H_
357 struct dup2_args {
358 	u_int	from;
359 	u_int	to;
360 };
361 #endif
362 /* ARGSUSED */
363 int
364 sys_dup2(struct thread *td, struct dup2_args *uap)
365 {
366 
367 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
368 }
369 
370 /*
371  * Duplicate a file descriptor.
372  */
373 #ifndef _SYS_SYSPROTO_H_
374 struct dup_args {
375 	u_int	fd;
376 };
377 #endif
378 /* ARGSUSED */
379 int
380 sys_dup(struct thread *td, struct dup_args *uap)
381 {
382 
383 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
384 }
385 
386 /*
387  * The file control system call.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct fcntl_args {
391 	int	fd;
392 	int	cmd;
393 	long	arg;
394 };
395 #endif
396 /* ARGSUSED */
397 int
398 sys_fcntl(struct thread *td, struct fcntl_args *uap)
399 {
400 
401 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
402 }
403 
404 int
405 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
406 {
407 	struct flock fl;
408 	struct __oflock ofl;
409 	intptr_t arg1;
410 	int error, newcmd;
411 
412 	error = 0;
413 	newcmd = cmd;
414 	switch (cmd) {
415 	case F_OGETLK:
416 	case F_OSETLK:
417 	case F_OSETLKW:
418 		/*
419 		 * Convert old flock structure to new.
420 		 */
421 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
422 		fl.l_start = ofl.l_start;
423 		fl.l_len = ofl.l_len;
424 		fl.l_pid = ofl.l_pid;
425 		fl.l_type = ofl.l_type;
426 		fl.l_whence = ofl.l_whence;
427 		fl.l_sysid = 0;
428 
429 		switch (cmd) {
430 		case F_OGETLK:
431 			newcmd = F_GETLK;
432 			break;
433 		case F_OSETLK:
434 			newcmd = F_SETLK;
435 			break;
436 		case F_OSETLKW:
437 			newcmd = F_SETLKW;
438 			break;
439 		}
440 		arg1 = (intptr_t)&fl;
441 		break;
442 	case F_GETLK:
443 	case F_SETLK:
444 	case F_SETLKW:
445 	case F_SETLK_REMOTE:
446 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
447 		arg1 = (intptr_t)&fl;
448 		break;
449 	default:
450 		arg1 = arg;
451 		break;
452 	}
453 	if (error)
454 		return (error);
455 	error = kern_fcntl(td, fd, newcmd, arg1);
456 	if (error)
457 		return (error);
458 	if (cmd == F_OGETLK) {
459 		ofl.l_start = fl.l_start;
460 		ofl.l_len = fl.l_len;
461 		ofl.l_pid = fl.l_pid;
462 		ofl.l_type = fl.l_type;
463 		ofl.l_whence = fl.l_whence;
464 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
465 	} else if (cmd == F_GETLK) {
466 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
467 	}
468 	return (error);
469 }
470 
471 int
472 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 {
474 	struct filedesc *fdp;
475 	struct flock *flp;
476 	struct file *fp, *fp2;
477 	struct filedescent *fde;
478 	struct proc *p;
479 	struct vnode *vp;
480 	struct mount *mp;
481 	int error, flg, seals, tmp;
482 	uint64_t bsize;
483 	off_t foffset;
484 
485 	error = 0;
486 	flg = F_POSIX;
487 	p = td->td_proc;
488 	fdp = p->p_fd;
489 
490 	AUDIT_ARG_FD(cmd);
491 	AUDIT_ARG_CMD(cmd);
492 	switch (cmd) {
493 	case F_DUPFD:
494 		tmp = arg;
495 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
496 		break;
497 
498 	case F_DUPFD_CLOEXEC:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
501 		break;
502 
503 	case F_DUP2FD:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD_CLOEXEC:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
511 		break;
512 
513 	case F_GETFD:
514 		error = EBADF;
515 		FILEDESC_SLOCK(fdp);
516 		fde = fdeget_locked(fdp, fd);
517 		if (fde != NULL) {
518 			td->td_retval[0] =
519 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
520 			error = 0;
521 		}
522 		FILEDESC_SUNLOCK(fdp);
523 		break;
524 
525 	case F_SETFD:
526 		error = EBADF;
527 		FILEDESC_XLOCK(fdp);
528 		fde = fdeget_locked(fdp, fd);
529 		if (fde != NULL) {
530 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
531 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
532 			error = 0;
533 		}
534 		FILEDESC_XUNLOCK(fdp);
535 		break;
536 
537 	case F_GETFL:
538 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
539 		if (error != 0)
540 			break;
541 		td->td_retval[0] = OFLAGS(fp->f_flag);
542 		fdrop(fp, td);
543 		break;
544 
545 	case F_SETFL:
546 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
547 		if (error != 0)
548 			break;
549 		do {
550 			tmp = flg = fp->f_flag;
551 			tmp &= ~FCNTLFLAGS;
552 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
553 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
554 		tmp = fp->f_flag & FNONBLOCK;
555 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
556 		if (error != 0) {
557 			fdrop(fp, td);
558 			break;
559 		}
560 		tmp = fp->f_flag & FASYNC;
561 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
562 		if (error == 0) {
563 			fdrop(fp, td);
564 			break;
565 		}
566 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
567 		tmp = 0;
568 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
569 		fdrop(fp, td);
570 		break;
571 
572 	case F_GETOWN:
573 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
574 		if (error != 0)
575 			break;
576 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 		if (error == 0)
578 			td->td_retval[0] = tmp;
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_SETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		tmp = arg;
587 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
588 		fdrop(fp, td);
589 		break;
590 
591 	case F_SETLK_REMOTE:
592 		error = priv_check(td, PRIV_NFS_LOCKD);
593 		if (error != 0)
594 			return (error);
595 		flg = F_REMOTE;
596 		goto do_setlk;
597 
598 	case F_SETLKW:
599 		flg |= F_WAIT;
600 		/* FALLTHROUGH F_SETLK */
601 
602 	case F_SETLK:
603 	do_setlk:
604 		flp = (struct flock *)arg;
605 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
606 			error = EINVAL;
607 			break;
608 		}
609 
610 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
611 		if (error != 0)
612 			break;
613 		if (fp->f_type != DTYPE_VNODE) {
614 			error = EBADF;
615 			fdrop(fp, td);
616 			break;
617 		}
618 
619 		if (flp->l_whence == SEEK_CUR) {
620 			foffset = foffset_get(fp);
621 			if (foffset < 0 ||
622 			    (flp->l_start > 0 &&
623 			     foffset > OFF_MAX - flp->l_start)) {
624 				error = EOVERFLOW;
625 				fdrop(fp, td);
626 				break;
627 			}
628 			flp->l_start += foffset;
629 		}
630 
631 		vp = fp->f_vnode;
632 		switch (flp->l_type) {
633 		case F_RDLCK:
634 			if ((fp->f_flag & FREAD) == 0) {
635 				error = EBADF;
636 				break;
637 			}
638 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
639 				PROC_LOCK(p->p_leader);
640 				p->p_leader->p_flag |= P_ADVLOCK;
641 				PROC_UNLOCK(p->p_leader);
642 			}
643 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
644 			    flp, flg);
645 			break;
646 		case F_WRLCK:
647 			if ((fp->f_flag & FWRITE) == 0) {
648 				error = EBADF;
649 				break;
650 			}
651 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
652 				PROC_LOCK(p->p_leader);
653 				p->p_leader->p_flag |= P_ADVLOCK;
654 				PROC_UNLOCK(p->p_leader);
655 			}
656 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
657 			    flp, flg);
658 			break;
659 		case F_UNLCK:
660 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
661 			    flp, flg);
662 			break;
663 		case F_UNLCKSYS:
664 			if (flg != F_REMOTE) {
665 				error = EINVAL;
666 				break;
667 			}
668 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
669 			    F_UNLCKSYS, flp, flg);
670 			break;
671 		default:
672 			error = EINVAL;
673 			break;
674 		}
675 		if (error != 0 || flp->l_type == F_UNLCK ||
676 		    flp->l_type == F_UNLCKSYS) {
677 			fdrop(fp, td);
678 			break;
679 		}
680 
681 		/*
682 		 * Check for a race with close.
683 		 *
684 		 * The vnode is now advisory locked (or unlocked, but this case
685 		 * is not really important) as the caller requested.
686 		 * We had to drop the filedesc lock, so we need to recheck if
687 		 * the descriptor is still valid, because if it was closed
688 		 * in the meantime we need to remove advisory lock from the
689 		 * vnode - close on any descriptor leading to an advisory
690 		 * locked vnode, removes that lock.
691 		 * We will return 0 on purpose in that case, as the result of
692 		 * successful advisory lock might have been externally visible
693 		 * already. This is fine - effectively we pretend to the caller
694 		 * that the closing thread was a bit slower and that the
695 		 * advisory lock succeeded before the close.
696 		 */
697 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
698 		if (error != 0) {
699 			fdrop(fp, td);
700 			break;
701 		}
702 		if (fp != fp2) {
703 			flp->l_whence = SEEK_SET;
704 			flp->l_start = 0;
705 			flp->l_len = 0;
706 			flp->l_type = F_UNLCK;
707 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
708 			    F_UNLCK, flp, F_POSIX);
709 		}
710 		fdrop(fp, td);
711 		fdrop(fp2, td);
712 		break;
713 
714 	case F_GETLK:
715 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
716 		if (error != 0)
717 			break;
718 		if (fp->f_type != DTYPE_VNODE) {
719 			error = EBADF;
720 			fdrop(fp, td);
721 			break;
722 		}
723 		flp = (struct flock *)arg;
724 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
725 		    flp->l_type != F_UNLCK) {
726 			error = EINVAL;
727 			fdrop(fp, td);
728 			break;
729 		}
730 		if (flp->l_whence == SEEK_CUR) {
731 			foffset = foffset_get(fp);
732 			if ((flp->l_start > 0 &&
733 			    foffset > OFF_MAX - flp->l_start) ||
734 			    (flp->l_start < 0 &&
735 			    foffset < OFF_MIN - flp->l_start)) {
736 				error = EOVERFLOW;
737 				fdrop(fp, td);
738 				break;
739 			}
740 			flp->l_start += foffset;
741 		}
742 		vp = fp->f_vnode;
743 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
744 		    F_POSIX);
745 		fdrop(fp, td);
746 		break;
747 
748 	case F_ADD_SEALS:
749 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
750 		if (error != 0)
751 			break;
752 		error = fo_add_seals(fp, arg);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_GET_SEALS:
757 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
758 		if (error != 0)
759 			break;
760 		if (fo_get_seals(fp, &seals) == 0)
761 			td->td_retval[0] = seals;
762 		else
763 			error = EINVAL;
764 		fdrop(fp, td);
765 		break;
766 
767 	case F_RDAHEAD:
768 		arg = arg ? 128 * 1024: 0;
769 		/* FALLTHROUGH */
770 	case F_READAHEAD:
771 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
772 		if (error != 0)
773 			break;
774 		if (fp->f_type != DTYPE_VNODE) {
775 			fdrop(fp, td);
776 			error = EBADF;
777 			break;
778 		}
779 		vp = fp->f_vnode;
780 		if (vp->v_type != VREG) {
781 			fdrop(fp, td);
782 			error = ENOTTY;
783 			break;
784 		}
785 
786 		/*
787 		 * Exclusive lock synchronizes against f_seqcount reads and
788 		 * writes in sequential_heuristic().
789 		 */
790 		error = vn_lock(vp, LK_EXCLUSIVE);
791 		if (error != 0) {
792 			fdrop(fp, td);
793 			break;
794 		}
795 		if (arg >= 0) {
796 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
797 			arg = MIN(arg, INT_MAX - bsize + 1);
798 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
799 			    (arg + bsize - 1) / bsize);
800 			atomic_set_int(&fp->f_flag, FRDAHEAD);
801 		} else {
802 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
803 		}
804 		VOP_UNLOCK(vp);
805 		fdrop(fp, td);
806 		break;
807 
808 	case F_ISUNIONSTACK:
809 		/*
810 		 * Check if the vnode is part of a union stack (either the
811 		 * "union" flag from mount(2) or unionfs).
812 		 *
813 		 * Prior to introduction of this op libc's readdir would call
814 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
815 		 * data just to check fs name and a mount flag.
816 		 *
817 		 * Fixing the code to handle everything in the kernel instead
818 		 * is a non-trivial endeavor and has low priority, thus this
819 		 * horrible kludge facilitates the current behavior in a much
820 		 * cheaper manner until someone(tm) sorts this out.
821 		 */
822 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
823 		if (error != 0)
824 			break;
825 		if (fp->f_type != DTYPE_VNODE) {
826 			fdrop(fp, td);
827 			error = EBADF;
828 			break;
829 		}
830 		vp = fp->f_vnode;
831 		/*
832 		 * Since we don't prevent dooming the vnode even non-null mp
833 		 * found can become immediately stale. This is tolerable since
834 		 * mount points are type-stable (providing safe memory access)
835 		 * and any vfs op on this vnode going forward will return an
836 		 * error (meaning return value in this case is meaningless).
837 		 */
838 		mp = atomic_load_ptr(&vp->v_mount);
839 		if (__predict_false(mp == NULL)) {
840 			fdrop(fp, td);
841 			error = EBADF;
842 			break;
843 		}
844 		td->td_retval[0] = 0;
845 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
846 		    mp->mnt_flag & MNT_UNION)
847 			td->td_retval[0] = 1;
848 		fdrop(fp, td);
849 		break;
850 
851 	default:
852 		error = EINVAL;
853 		break;
854 	}
855 	return (error);
856 }
857 
858 static int
859 getmaxfd(struct thread *td)
860 {
861 
862 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
863 }
864 
865 /*
866  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
867  */
868 int
869 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
870 {
871 	struct filedesc *fdp;
872 	struct filedescent *oldfde, *newfde;
873 	struct proc *p;
874 	struct file *delfp, *oldfp;
875 	u_long *oioctls, *nioctls;
876 	int error, maxfd;
877 
878 	p = td->td_proc;
879 	fdp = p->p_fd;
880 	oioctls = NULL;
881 
882 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
883 	MPASS(mode < FDDUP_LASTMODE);
884 
885 	AUDIT_ARG_FD(old);
886 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
887 
888 	/*
889 	 * Verify we have a valid descriptor to dup from and possibly to
890 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
891 	 * return EINVAL when the new descriptor is out of bounds.
892 	 */
893 	if (old < 0)
894 		return (EBADF);
895 	if (new < 0)
896 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
897 	maxfd = getmaxfd(td);
898 	if (new >= maxfd)
899 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
900 
901 	error = EBADF;
902 	FILEDESC_XLOCK(fdp);
903 	if (fget_locked(fdp, old) == NULL)
904 		goto unlock;
905 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
906 		td->td_retval[0] = new;
907 		if (flags & FDDUP_FLAG_CLOEXEC)
908 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
909 		error = 0;
910 		goto unlock;
911 	}
912 
913 	oldfde = &fdp->fd_ofiles[old];
914 	oldfp = oldfde->fde_file;
915 	if (!fhold(oldfp))
916 		goto unlock;
917 
918 	/*
919 	 * If the caller specified a file descriptor, make sure the file
920 	 * table is large enough to hold it, and grab it.  Otherwise, just
921 	 * allocate a new descriptor the usual way.
922 	 */
923 	switch (mode) {
924 	case FDDUP_NORMAL:
925 	case FDDUP_FCNTL:
926 		if ((error = fdalloc(td, new, &new)) != 0) {
927 			fdrop(oldfp, td);
928 			goto unlock;
929 		}
930 		break;
931 	case FDDUP_MUSTREPLACE:
932 		/* Target file descriptor must exist. */
933 		if (fget_locked(fdp, new) == NULL) {
934 			fdrop(oldfp, td);
935 			goto unlock;
936 		}
937 		break;
938 	case FDDUP_FIXED:
939 		if (new >= fdp->fd_nfiles) {
940 			/*
941 			 * The resource limits are here instead of e.g.
942 			 * fdalloc(), because the file descriptor table may be
943 			 * shared between processes, so we can't really use
944 			 * racct_add()/racct_sub().  Instead of counting the
945 			 * number of actually allocated descriptors, just put
946 			 * the limit on the size of the file descriptor table.
947 			 */
948 #ifdef RACCT
949 			if (RACCT_ENABLED()) {
950 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
951 				if (error != 0) {
952 					error = EMFILE;
953 					fdrop(oldfp, td);
954 					goto unlock;
955 				}
956 			}
957 #endif
958 			fdgrowtable_exp(fdp, new + 1);
959 		}
960 		if (!fdisused(fdp, new))
961 			fdused(fdp, new);
962 		break;
963 	default:
964 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
965 	}
966 
967 	KASSERT(old != new, ("new fd is same as old"));
968 
969 	/* Refetch oldfde because the table may have grown and old one freed. */
970 	oldfde = &fdp->fd_ofiles[old];
971 	KASSERT(oldfp == oldfde->fde_file,
972 	    ("fdt_ofiles shift from growth observed at fd %d",
973 	    old));
974 
975 	newfde = &fdp->fd_ofiles[new];
976 	delfp = newfde->fde_file;
977 
978 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
979 
980 	/*
981 	 * Duplicate the source descriptor.
982 	 */
983 #ifdef CAPABILITIES
984 	seqc_write_begin(&newfde->fde_seqc);
985 #endif
986 	oioctls = filecaps_free_prep(&newfde->fde_caps);
987 	memcpy(newfde, oldfde, fde_change_size);
988 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
989 	    nioctls);
990 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
991 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
992 	else
993 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
994 #ifdef CAPABILITIES
995 	seqc_write_end(&newfde->fde_seqc);
996 #endif
997 	td->td_retval[0] = new;
998 
999 	error = 0;
1000 
1001 	if (delfp != NULL) {
1002 		(void) closefp(fdp, new, delfp, td, true, false);
1003 		FILEDESC_UNLOCK_ASSERT(fdp);
1004 	} else {
1005 unlock:
1006 		FILEDESC_XUNLOCK(fdp);
1007 	}
1008 
1009 	filecaps_free_finish(oioctls);
1010 	return (error);
1011 }
1012 
1013 static void
1014 sigiofree(struct sigio *sigio)
1015 {
1016 	crfree(sigio->sio_ucred);
1017 	free(sigio, M_SIGIO);
1018 }
1019 
1020 static struct sigio *
1021 funsetown_locked(struct sigio *sigio)
1022 {
1023 	struct proc *p;
1024 	struct pgrp *pg;
1025 
1026 	SIGIO_ASSERT_LOCKED();
1027 
1028 	if (sigio == NULL)
1029 		return (NULL);
1030 	*(sigio->sio_myref) = NULL;
1031 	if (sigio->sio_pgid < 0) {
1032 		pg = sigio->sio_pgrp;
1033 		PGRP_LOCK(pg);
1034 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1035 		    sigio, sio_pgsigio);
1036 		PGRP_UNLOCK(pg);
1037 	} else {
1038 		p = sigio->sio_proc;
1039 		PROC_LOCK(p);
1040 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1041 		    sigio, sio_pgsigio);
1042 		PROC_UNLOCK(p);
1043 	}
1044 	return (sigio);
1045 }
1046 
1047 /*
1048  * If sigio is on the list associated with a process or process group,
1049  * disable signalling from the device, remove sigio from the list and
1050  * free sigio.
1051  */
1052 void
1053 funsetown(struct sigio **sigiop)
1054 {
1055 	struct sigio *sigio;
1056 
1057 	/* Racy check, consumers must provide synchronization. */
1058 	if (*sigiop == NULL)
1059 		return;
1060 
1061 	SIGIO_LOCK();
1062 	sigio = funsetown_locked(*sigiop);
1063 	SIGIO_UNLOCK();
1064 	if (sigio != NULL)
1065 		sigiofree(sigio);
1066 }
1067 
1068 /*
1069  * Free a list of sigio structures.  The caller must ensure that new sigio
1070  * structures cannot be added after this point.  For process groups this is
1071  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1072  * as an interlock.
1073  */
1074 void
1075 funsetownlst(struct sigiolst *sigiolst)
1076 {
1077 	struct proc *p;
1078 	struct pgrp *pg;
1079 	struct sigio *sigio, *tmp;
1080 
1081 	/* Racy check. */
1082 	sigio = SLIST_FIRST(sigiolst);
1083 	if (sigio == NULL)
1084 		return;
1085 
1086 	p = NULL;
1087 	pg = NULL;
1088 
1089 	SIGIO_LOCK();
1090 	sigio = SLIST_FIRST(sigiolst);
1091 	if (sigio == NULL) {
1092 		SIGIO_UNLOCK();
1093 		return;
1094 	}
1095 
1096 	/*
1097 	 * Every entry of the list should belong to a single proc or pgrp.
1098 	 */
1099 	if (sigio->sio_pgid < 0) {
1100 		pg = sigio->sio_pgrp;
1101 		sx_assert(&proctree_lock, SX_XLOCKED);
1102 		PGRP_LOCK(pg);
1103 	} else /* if (sigio->sio_pgid > 0) */ {
1104 		p = sigio->sio_proc;
1105 		PROC_LOCK(p);
1106 		KASSERT((p->p_flag & P_WEXIT) != 0,
1107 		    ("%s: process %p is not exiting", __func__, p));
1108 	}
1109 
1110 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1111 		*sigio->sio_myref = NULL;
1112 		if (pg != NULL) {
1113 			KASSERT(sigio->sio_pgid < 0,
1114 			    ("Proc sigio in pgrp sigio list"));
1115 			KASSERT(sigio->sio_pgrp == pg,
1116 			    ("Bogus pgrp in sigio list"));
1117 		} else /* if (p != NULL) */ {
1118 			KASSERT(sigio->sio_pgid > 0,
1119 			    ("Pgrp sigio in proc sigio list"));
1120 			KASSERT(sigio->sio_proc == p,
1121 			    ("Bogus proc in sigio list"));
1122 		}
1123 	}
1124 
1125 	if (pg != NULL)
1126 		PGRP_UNLOCK(pg);
1127 	else
1128 		PROC_UNLOCK(p);
1129 	SIGIO_UNLOCK();
1130 
1131 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1132 		sigiofree(sigio);
1133 }
1134 
1135 /*
1136  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1137  *
1138  * After permission checking, add a sigio structure to the sigio list for
1139  * the process or process group.
1140  */
1141 int
1142 fsetown(pid_t pgid, struct sigio **sigiop)
1143 {
1144 	struct proc *proc;
1145 	struct pgrp *pgrp;
1146 	struct sigio *osigio, *sigio;
1147 	int ret;
1148 
1149 	if (pgid == 0) {
1150 		funsetown(sigiop);
1151 		return (0);
1152 	}
1153 
1154 	ret = 0;
1155 
1156 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1157 	sigio->sio_pgid = pgid;
1158 	sigio->sio_ucred = crhold(curthread->td_ucred);
1159 	sigio->sio_myref = sigiop;
1160 
1161 	sx_slock(&proctree_lock);
1162 	SIGIO_LOCK();
1163 	osigio = funsetown_locked(*sigiop);
1164 	if (pgid > 0) {
1165 		proc = pfind(pgid);
1166 		if (proc == NULL) {
1167 			ret = ESRCH;
1168 			goto fail;
1169 		}
1170 
1171 		/*
1172 		 * Policy - Don't allow a process to FSETOWN a process
1173 		 * in another session.
1174 		 *
1175 		 * Remove this test to allow maximum flexibility or
1176 		 * restrict FSETOWN to the current process or process
1177 		 * group for maximum safety.
1178 		 */
1179 		if (proc->p_session != curthread->td_proc->p_session) {
1180 			PROC_UNLOCK(proc);
1181 			ret = EPERM;
1182 			goto fail;
1183 		}
1184 
1185 		sigio->sio_proc = proc;
1186 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1187 		PROC_UNLOCK(proc);
1188 	} else /* if (pgid < 0) */ {
1189 		pgrp = pgfind(-pgid);
1190 		if (pgrp == NULL) {
1191 			ret = ESRCH;
1192 			goto fail;
1193 		}
1194 
1195 		/*
1196 		 * Policy - Don't allow a process to FSETOWN a process
1197 		 * in another session.
1198 		 *
1199 		 * Remove this test to allow maximum flexibility or
1200 		 * restrict FSETOWN to the current process or process
1201 		 * group for maximum safety.
1202 		 */
1203 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1204 			PGRP_UNLOCK(pgrp);
1205 			ret = EPERM;
1206 			goto fail;
1207 		}
1208 
1209 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1210 		sigio->sio_pgrp = pgrp;
1211 		PGRP_UNLOCK(pgrp);
1212 	}
1213 	sx_sunlock(&proctree_lock);
1214 	*sigiop = sigio;
1215 	SIGIO_UNLOCK();
1216 	if (osigio != NULL)
1217 		sigiofree(osigio);
1218 	return (0);
1219 
1220 fail:
1221 	SIGIO_UNLOCK();
1222 	sx_sunlock(&proctree_lock);
1223 	sigiofree(sigio);
1224 	if (osigio != NULL)
1225 		sigiofree(osigio);
1226 	return (ret);
1227 }
1228 
1229 /*
1230  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1231  */
1232 pid_t
1233 fgetown(struct sigio **sigiop)
1234 {
1235 	pid_t pgid;
1236 
1237 	SIGIO_LOCK();
1238 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1239 	SIGIO_UNLOCK();
1240 	return (pgid);
1241 }
1242 
1243 static int
1244 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1245     bool audit)
1246 {
1247 	int error;
1248 
1249 	FILEDESC_XLOCK_ASSERT(fdp);
1250 
1251 	/*
1252 	 * We now hold the fp reference that used to be owned by the
1253 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1254 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1255 	 * added, and deleteing a knote for the new fd.
1256 	 */
1257 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1258 		knote_fdclose(td, fd);
1259 
1260 	/*
1261 	 * We need to notify mqueue if the object is of type mqueue.
1262 	 */
1263 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1264 		mq_fdclose(td, fd, fp);
1265 	FILEDESC_XUNLOCK(fdp);
1266 
1267 #ifdef AUDIT
1268 	if (AUDITING_TD(td) && audit)
1269 		audit_sysclose(td, fd, fp);
1270 #endif
1271 	error = closef(fp, td);
1272 
1273 	/*
1274 	 * All paths leading up to closefp() will have already removed or
1275 	 * replaced the fd in the filedesc table, so a restart would not
1276 	 * operate on the same file.
1277 	 */
1278 	if (error == ERESTART)
1279 		error = EINTR;
1280 
1281 	return (error);
1282 }
1283 
1284 static int
1285 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1286     bool holdleaders, bool audit)
1287 {
1288 	int error;
1289 
1290 	FILEDESC_XLOCK_ASSERT(fdp);
1291 
1292 	if (holdleaders) {
1293 		if (td->td_proc->p_fdtol != NULL) {
1294 			/*
1295 			 * Ask fdfree() to sleep to ensure that all relevant
1296 			 * process leaders can be traversed in closef().
1297 			 */
1298 			fdp->fd_holdleaderscount++;
1299 		} else {
1300 			holdleaders = false;
1301 		}
1302 	}
1303 
1304 	error = closefp_impl(fdp, fd, fp, td, audit);
1305 	if (holdleaders) {
1306 		FILEDESC_XLOCK(fdp);
1307 		fdp->fd_holdleaderscount--;
1308 		if (fdp->fd_holdleaderscount == 0 &&
1309 		    fdp->fd_holdleaderswakeup != 0) {
1310 			fdp->fd_holdleaderswakeup = 0;
1311 			wakeup(&fdp->fd_holdleaderscount);
1312 		}
1313 		FILEDESC_XUNLOCK(fdp);
1314 	}
1315 	return (error);
1316 }
1317 
1318 static int
1319 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1320     bool holdleaders, bool audit)
1321 {
1322 
1323 	FILEDESC_XLOCK_ASSERT(fdp);
1324 
1325 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1326 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1327 	} else {
1328 		return (closefp_impl(fdp, fd, fp, td, audit));
1329 	}
1330 }
1331 
1332 /*
1333  * Close a file descriptor.
1334  */
1335 #ifndef _SYS_SYSPROTO_H_
1336 struct close_args {
1337 	int     fd;
1338 };
1339 #endif
1340 /* ARGSUSED */
1341 int
1342 sys_close(struct thread *td, struct close_args *uap)
1343 {
1344 
1345 	return (kern_close(td, uap->fd));
1346 }
1347 
1348 int
1349 kern_close(struct thread *td, int fd)
1350 {
1351 	struct filedesc *fdp;
1352 	struct file *fp;
1353 
1354 	fdp = td->td_proc->p_fd;
1355 
1356 	FILEDESC_XLOCK(fdp);
1357 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1358 		FILEDESC_XUNLOCK(fdp);
1359 		return (EBADF);
1360 	}
1361 	fdfree(fdp, fd);
1362 
1363 	/* closefp() drops the FILEDESC lock for us. */
1364 	return (closefp(fdp, fd, fp, td, true, true));
1365 }
1366 
1367 int
1368 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1369 {
1370 	struct filedesc *fdp;
1371 	const struct fdescenttbl *fdt;
1372 	struct file *fp;
1373 	int fd;
1374 
1375 	/*
1376 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1377 	 * open should not be a usage error.  From a close_range() perspective,
1378 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1379 	 * be a usage error as all fd above 3 are in-fact already closed.
1380 	 */
1381 	if (highfd < lowfd) {
1382 		return (EINVAL);
1383 	}
1384 
1385 	fdp = td->td_proc->p_fd;
1386 	FILEDESC_XLOCK(fdp);
1387 	fdt = atomic_load_ptr(&fdp->fd_files);
1388 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1389 	fd = lowfd;
1390 	if (__predict_false(fd > highfd)) {
1391 		goto out_locked;
1392 	}
1393 	for (;;) {
1394 		fp = fdt->fdt_ofiles[fd].fde_file;
1395 		if (fp == NULL) {
1396 			if (fd == highfd)
1397 				goto out_locked;
1398 		} else {
1399 			fdfree(fdp, fd);
1400 			(void) closefp(fdp, fd, fp, td, true, true);
1401 			if (fd == highfd)
1402 				goto out_unlocked;
1403 			FILEDESC_XLOCK(fdp);
1404 			fdt = atomic_load_ptr(&fdp->fd_files);
1405 		}
1406 		fd++;
1407 	}
1408 out_locked:
1409 	FILEDESC_XUNLOCK(fdp);
1410 out_unlocked:
1411 	return (0);
1412 }
1413 
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct close_range_args {
1416 	u_int	lowfd;
1417 	u_int	highfd;
1418 	int	flags;
1419 };
1420 #endif
1421 int
1422 sys_close_range(struct thread *td, struct close_range_args *uap)
1423 {
1424 
1425 	/* No flags currently defined */
1426 	if (uap->flags != 0)
1427 		return (EINVAL);
1428 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1429 }
1430 
1431 #ifdef COMPAT_FREEBSD12
1432 /*
1433  * Close open file descriptors.
1434  */
1435 #ifndef _SYS_SYSPROTO_H_
1436 struct freebsd12_closefrom_args {
1437 	int	lowfd;
1438 };
1439 #endif
1440 /* ARGSUSED */
1441 int
1442 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1443 {
1444 	u_int lowfd;
1445 
1446 	AUDIT_ARG_FD(uap->lowfd);
1447 
1448 	/*
1449 	 * Treat negative starting file descriptor values identical to
1450 	 * closefrom(0) which closes all files.
1451 	 */
1452 	lowfd = MAX(0, uap->lowfd);
1453 	return (kern_close_range(td, lowfd, ~0U));
1454 }
1455 #endif	/* COMPAT_FREEBSD12 */
1456 
1457 #if defined(COMPAT_43)
1458 /*
1459  * Return status information about a file descriptor.
1460  */
1461 #ifndef _SYS_SYSPROTO_H_
1462 struct ofstat_args {
1463 	int	fd;
1464 	struct	ostat *sb;
1465 };
1466 #endif
1467 /* ARGSUSED */
1468 int
1469 ofstat(struct thread *td, struct ofstat_args *uap)
1470 {
1471 	struct ostat oub;
1472 	struct stat ub;
1473 	int error;
1474 
1475 	error = kern_fstat(td, uap->fd, &ub);
1476 	if (error == 0) {
1477 		cvtstat(&ub, &oub);
1478 		error = copyout(&oub, uap->sb, sizeof(oub));
1479 	}
1480 	return (error);
1481 }
1482 #endif /* COMPAT_43 */
1483 
1484 #if defined(COMPAT_FREEBSD11)
1485 int
1486 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1487 {
1488 	struct stat sb;
1489 	struct freebsd11_stat osb;
1490 	int error;
1491 
1492 	error = kern_fstat(td, uap->fd, &sb);
1493 	if (error != 0)
1494 		return (error);
1495 	error = freebsd11_cvtstat(&sb, &osb);
1496 	if (error == 0)
1497 		error = copyout(&osb, uap->sb, sizeof(osb));
1498 	return (error);
1499 }
1500 #endif	/* COMPAT_FREEBSD11 */
1501 
1502 /*
1503  * Return status information about a file descriptor.
1504  */
1505 #ifndef _SYS_SYSPROTO_H_
1506 struct fstat_args {
1507 	int	fd;
1508 	struct	stat *sb;
1509 };
1510 #endif
1511 /* ARGSUSED */
1512 int
1513 sys_fstat(struct thread *td, struct fstat_args *uap)
1514 {
1515 	struct stat ub;
1516 	int error;
1517 
1518 	error = kern_fstat(td, uap->fd, &ub);
1519 	if (error == 0)
1520 		error = copyout(&ub, uap->sb, sizeof(ub));
1521 	return (error);
1522 }
1523 
1524 int
1525 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1526 {
1527 	struct file *fp;
1528 	int error;
1529 
1530 	AUDIT_ARG_FD(fd);
1531 
1532 	error = fget(td, fd, &cap_fstat_rights, &fp);
1533 	if (__predict_false(error != 0))
1534 		return (error);
1535 
1536 	AUDIT_ARG_FILE(td->td_proc, fp);
1537 
1538 	error = fo_stat(fp, sbp, td->td_ucred, td);
1539 	fdrop(fp, td);
1540 #ifdef __STAT_TIME_T_EXT
1541 	sbp->st_atim_ext = 0;
1542 	sbp->st_mtim_ext = 0;
1543 	sbp->st_ctim_ext = 0;
1544 	sbp->st_btim_ext = 0;
1545 #endif
1546 #ifdef KTRACE
1547 	if (KTRPOINT(td, KTR_STRUCT))
1548 		ktrstat_error(sbp, error);
1549 #endif
1550 	return (error);
1551 }
1552 
1553 #if defined(COMPAT_FREEBSD11)
1554 /*
1555  * Return status information about a file descriptor.
1556  */
1557 #ifndef _SYS_SYSPROTO_H_
1558 struct freebsd11_nfstat_args {
1559 	int	fd;
1560 	struct	nstat *sb;
1561 };
1562 #endif
1563 /* ARGSUSED */
1564 int
1565 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1566 {
1567 	struct nstat nub;
1568 	struct stat ub;
1569 	int error;
1570 
1571 	error = kern_fstat(td, uap->fd, &ub);
1572 	if (error == 0) {
1573 		freebsd11_cvtnstat(&ub, &nub);
1574 		error = copyout(&nub, uap->sb, sizeof(nub));
1575 	}
1576 	return (error);
1577 }
1578 #endif /* COMPAT_FREEBSD11 */
1579 
1580 /*
1581  * Return pathconf information about a file descriptor.
1582  */
1583 #ifndef _SYS_SYSPROTO_H_
1584 struct fpathconf_args {
1585 	int	fd;
1586 	int	name;
1587 };
1588 #endif
1589 /* ARGSUSED */
1590 int
1591 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1592 {
1593 	long value;
1594 	int error;
1595 
1596 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1597 	if (error == 0)
1598 		td->td_retval[0] = value;
1599 	return (error);
1600 }
1601 
1602 int
1603 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1604 {
1605 	struct file *fp;
1606 	struct vnode *vp;
1607 	int error;
1608 
1609 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1610 	if (error != 0)
1611 		return (error);
1612 
1613 	if (name == _PC_ASYNC_IO) {
1614 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1615 		goto out;
1616 	}
1617 	vp = fp->f_vnode;
1618 	if (vp != NULL) {
1619 		vn_lock(vp, LK_SHARED | LK_RETRY);
1620 		error = VOP_PATHCONF(vp, name, valuep);
1621 		VOP_UNLOCK(vp);
1622 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1623 		if (name != _PC_PIPE_BUF) {
1624 			error = EINVAL;
1625 		} else {
1626 			*valuep = PIPE_BUF;
1627 			error = 0;
1628 		}
1629 	} else {
1630 		error = EOPNOTSUPP;
1631 	}
1632 out:
1633 	fdrop(fp, td);
1634 	return (error);
1635 }
1636 
1637 /*
1638  * Copy filecaps structure allocating memory for ioctls array if needed.
1639  *
1640  * The last parameter indicates whether the fdtable is locked. If it is not and
1641  * ioctls are encountered, copying fails and the caller must lock the table.
1642  *
1643  * Note that if the table was not locked, the caller has to check the relevant
1644  * sequence counter to determine whether the operation was successful.
1645  */
1646 bool
1647 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1648 {
1649 	size_t size;
1650 
1651 	if (src->fc_ioctls != NULL && !locked)
1652 		return (false);
1653 	memcpy(dst, src, sizeof(*src));
1654 	if (src->fc_ioctls == NULL)
1655 		return (true);
1656 
1657 	KASSERT(src->fc_nioctls > 0,
1658 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1659 
1660 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1661 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1662 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1663 	return (true);
1664 }
1665 
1666 static u_long *
1667 filecaps_copy_prep(const struct filecaps *src)
1668 {
1669 	u_long *ioctls;
1670 	size_t size;
1671 
1672 	if (__predict_true(src->fc_ioctls == NULL))
1673 		return (NULL);
1674 
1675 	KASSERT(src->fc_nioctls > 0,
1676 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1677 
1678 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1679 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1680 	return (ioctls);
1681 }
1682 
1683 static void
1684 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1685     u_long *ioctls)
1686 {
1687 	size_t size;
1688 
1689 	*dst = *src;
1690 	if (__predict_true(src->fc_ioctls == NULL)) {
1691 		MPASS(ioctls == NULL);
1692 		return;
1693 	}
1694 
1695 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1696 	dst->fc_ioctls = ioctls;
1697 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1698 }
1699 
1700 /*
1701  * Move filecaps structure to the new place and clear the old place.
1702  */
1703 void
1704 filecaps_move(struct filecaps *src, struct filecaps *dst)
1705 {
1706 
1707 	*dst = *src;
1708 	bzero(src, sizeof(*src));
1709 }
1710 
1711 /*
1712  * Fill the given filecaps structure with full rights.
1713  */
1714 static void
1715 filecaps_fill(struct filecaps *fcaps)
1716 {
1717 
1718 	CAP_ALL(&fcaps->fc_rights);
1719 	fcaps->fc_ioctls = NULL;
1720 	fcaps->fc_nioctls = -1;
1721 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1722 }
1723 
1724 /*
1725  * Free memory allocated within filecaps structure.
1726  */
1727 void
1728 filecaps_free(struct filecaps *fcaps)
1729 {
1730 
1731 	free(fcaps->fc_ioctls, M_FILECAPS);
1732 	bzero(fcaps, sizeof(*fcaps));
1733 }
1734 
1735 static u_long *
1736 filecaps_free_prep(struct filecaps *fcaps)
1737 {
1738 	u_long *ioctls;
1739 
1740 	ioctls = fcaps->fc_ioctls;
1741 	bzero(fcaps, sizeof(*fcaps));
1742 	return (ioctls);
1743 }
1744 
1745 static void
1746 filecaps_free_finish(u_long *ioctls)
1747 {
1748 
1749 	free(ioctls, M_FILECAPS);
1750 }
1751 
1752 /*
1753  * Validate the given filecaps structure.
1754  */
1755 static void
1756 filecaps_validate(const struct filecaps *fcaps, const char *func)
1757 {
1758 
1759 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1760 	    ("%s: invalid rights", func));
1761 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1762 	    ("%s: invalid fcntls", func));
1763 	KASSERT(fcaps->fc_fcntls == 0 ||
1764 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1765 	    ("%s: fcntls without CAP_FCNTL", func));
1766 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1767 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1768 	    ("%s: invalid ioctls", func));
1769 	KASSERT(fcaps->fc_nioctls == 0 ||
1770 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1771 	    ("%s: ioctls without CAP_IOCTL", func));
1772 }
1773 
1774 static void
1775 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1776 {
1777 	int nfd1;
1778 
1779 	FILEDESC_XLOCK_ASSERT(fdp);
1780 
1781 	nfd1 = fdp->fd_nfiles * 2;
1782 	if (nfd1 < nfd)
1783 		nfd1 = nfd;
1784 	fdgrowtable(fdp, nfd1);
1785 }
1786 
1787 /*
1788  * Grow the file table to accommodate (at least) nfd descriptors.
1789  */
1790 static void
1791 fdgrowtable(struct filedesc *fdp, int nfd)
1792 {
1793 	struct filedesc0 *fdp0;
1794 	struct freetable *ft;
1795 	struct fdescenttbl *ntable;
1796 	struct fdescenttbl *otable;
1797 	int nnfiles, onfiles;
1798 	NDSLOTTYPE *nmap, *omap;
1799 
1800 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1801 
1802 	/* save old values */
1803 	onfiles = fdp->fd_nfiles;
1804 	otable = fdp->fd_files;
1805 	omap = fdp->fd_map;
1806 
1807 	/* compute the size of the new table */
1808 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1809 	if (nnfiles <= onfiles)
1810 		/* the table is already large enough */
1811 		return;
1812 
1813 	/*
1814 	 * Allocate a new table.  We need enough space for the number of
1815 	 * entries, file entries themselves and the struct freetable we will use
1816 	 * when we decommission the table and place it on the freelist.
1817 	 * We place the struct freetable in the middle so we don't have
1818 	 * to worry about padding.
1819 	 */
1820 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1821 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1822 	    sizeof(struct freetable),
1823 	    M_FILEDESC, M_ZERO | M_WAITOK);
1824 	/* copy the old data */
1825 	ntable->fdt_nfiles = nnfiles;
1826 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1827 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1828 
1829 	/*
1830 	 * Allocate a new map only if the old is not large enough.  It will
1831 	 * grow at a slower rate than the table as it can map more
1832 	 * entries than the table can hold.
1833 	 */
1834 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1835 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1836 		    M_ZERO | M_WAITOK);
1837 		/* copy over the old data and update the pointer */
1838 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1839 		fdp->fd_map = nmap;
1840 	}
1841 
1842 	/*
1843 	 * Make sure that ntable is correctly initialized before we replace
1844 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1845 	 * data.
1846 	 */
1847 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1848 
1849 	/*
1850 	 * Free the old file table when not shared by other threads or processes.
1851 	 * The old file table is considered to be shared when either are true:
1852 	 * - The process has more than one thread.
1853 	 * - The file descriptor table has been shared via fdshare().
1854 	 *
1855 	 * When shared, the old file table will be placed on a freelist
1856 	 * which will be processed when the struct filedesc is released.
1857 	 *
1858 	 * Note that if onfiles == NDFILE, we're dealing with the original
1859 	 * static allocation contained within (struct filedesc0 *)fdp,
1860 	 * which must not be freed.
1861 	 */
1862 	if (onfiles > NDFILE) {
1863 		if (curproc->p_numthreads == 1 &&
1864 		    refcount_load(&fdp->fd_refcnt) == 1)
1865 			free(otable, M_FILEDESC);
1866 		else {
1867 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1868 			fdp0 = (struct filedesc0 *)fdp;
1869 			ft->ft_table = otable;
1870 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1871 		}
1872 	}
1873 	/*
1874 	 * The map does not have the same possibility of threads still
1875 	 * holding references to it.  So always free it as long as it
1876 	 * does not reference the original static allocation.
1877 	 */
1878 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1879 		free(omap, M_FILEDESC);
1880 }
1881 
1882 /*
1883  * Allocate a file descriptor for the process.
1884  */
1885 int
1886 fdalloc(struct thread *td, int minfd, int *result)
1887 {
1888 	struct proc *p = td->td_proc;
1889 	struct filedesc *fdp = p->p_fd;
1890 	int fd, maxfd, allocfd;
1891 #ifdef RACCT
1892 	int error;
1893 #endif
1894 
1895 	FILEDESC_XLOCK_ASSERT(fdp);
1896 
1897 	if (fdp->fd_freefile > minfd)
1898 		minfd = fdp->fd_freefile;
1899 
1900 	maxfd = getmaxfd(td);
1901 
1902 	/*
1903 	 * Search the bitmap for a free descriptor starting at minfd.
1904 	 * If none is found, grow the file table.
1905 	 */
1906 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1907 	if (__predict_false(fd >= maxfd))
1908 		return (EMFILE);
1909 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1910 		allocfd = min(fd * 2, maxfd);
1911 #ifdef RACCT
1912 		if (RACCT_ENABLED()) {
1913 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1914 			if (error != 0)
1915 				return (EMFILE);
1916 		}
1917 #endif
1918 		/*
1919 		 * fd is already equal to first free descriptor >= minfd, so
1920 		 * we only need to grow the table and we are done.
1921 		 */
1922 		fdgrowtable_exp(fdp, allocfd);
1923 	}
1924 
1925 	/*
1926 	 * Perform some sanity checks, then mark the file descriptor as
1927 	 * used and return it to the caller.
1928 	 */
1929 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1930 	    ("invalid descriptor %d", fd));
1931 	KASSERT(!fdisused(fdp, fd),
1932 	    ("fd_first_free() returned non-free descriptor"));
1933 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1934 	    ("file descriptor isn't free"));
1935 	fdused(fdp, fd);
1936 	*result = fd;
1937 	return (0);
1938 }
1939 
1940 /*
1941  * Allocate n file descriptors for the process.
1942  */
1943 int
1944 fdallocn(struct thread *td, int minfd, int *fds, int n)
1945 {
1946 	struct proc *p = td->td_proc;
1947 	struct filedesc *fdp = p->p_fd;
1948 	int i;
1949 
1950 	FILEDESC_XLOCK_ASSERT(fdp);
1951 
1952 	for (i = 0; i < n; i++)
1953 		if (fdalloc(td, 0, &fds[i]) != 0)
1954 			break;
1955 
1956 	if (i < n) {
1957 		for (i--; i >= 0; i--)
1958 			fdunused(fdp, fds[i]);
1959 		return (EMFILE);
1960 	}
1961 
1962 	return (0);
1963 }
1964 
1965 /*
1966  * Create a new open file structure and allocate a file descriptor for the
1967  * process that refers to it.  We add one reference to the file for the
1968  * descriptor table and one reference for resultfp. This is to prevent us
1969  * being preempted and the entry in the descriptor table closed after we
1970  * release the FILEDESC lock.
1971  */
1972 int
1973 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1974     struct filecaps *fcaps)
1975 {
1976 	struct file *fp;
1977 	int error, fd;
1978 
1979 	MPASS(resultfp != NULL);
1980 	MPASS(resultfd != NULL);
1981 
1982 	error = falloc_noinstall(td, &fp);
1983 	if (error)
1984 		return (error);		/* no reference held on error */
1985 
1986 	error = finstall(td, fp, &fd, flags, fcaps);
1987 	if (error) {
1988 		fdrop(fp, td);		/* one reference (fp only) */
1989 		return (error);
1990 	}
1991 
1992 	*resultfp = fp;
1993 	*resultfd = fd;
1994 
1995 	return (0);
1996 }
1997 
1998 /*
1999  * Create a new open file structure without allocating a file descriptor.
2000  */
2001 int
2002 falloc_noinstall(struct thread *td, struct file **resultfp)
2003 {
2004 	struct file *fp;
2005 	int maxuserfiles = maxfiles - (maxfiles / 20);
2006 	int openfiles_new;
2007 	static struct timeval lastfail;
2008 	static int curfail;
2009 
2010 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2011 
2012 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2013 	if ((openfiles_new >= maxuserfiles &&
2014 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2015 	    openfiles_new >= maxfiles) {
2016 		atomic_subtract_int(&openfiles, 1);
2017 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2018 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2019 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2020 		}
2021 		return (ENFILE);
2022 	}
2023 	fp = uma_zalloc(file_zone, M_WAITOK);
2024 	bzero(fp, sizeof(*fp));
2025 	refcount_init(&fp->f_count, 1);
2026 	fp->f_cred = crhold(td->td_ucred);
2027 	fp->f_ops = &badfileops;
2028 	*resultfp = fp;
2029 	return (0);
2030 }
2031 
2032 /*
2033  * Install a file in a file descriptor table.
2034  */
2035 void
2036 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2037     struct filecaps *fcaps)
2038 {
2039 	struct filedescent *fde;
2040 
2041 	MPASS(fp != NULL);
2042 	if (fcaps != NULL)
2043 		filecaps_validate(fcaps, __func__);
2044 	FILEDESC_XLOCK_ASSERT(fdp);
2045 
2046 	fde = &fdp->fd_ofiles[fd];
2047 #ifdef CAPABILITIES
2048 	seqc_write_begin(&fde->fde_seqc);
2049 #endif
2050 	fde->fde_file = fp;
2051 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2052 	if (fcaps != NULL)
2053 		filecaps_move(fcaps, &fde->fde_caps);
2054 	else
2055 		filecaps_fill(&fde->fde_caps);
2056 #ifdef CAPABILITIES
2057 	seqc_write_end(&fde->fde_seqc);
2058 #endif
2059 }
2060 
2061 int
2062 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2063     struct filecaps *fcaps)
2064 {
2065 	struct filedesc *fdp = td->td_proc->p_fd;
2066 	int error;
2067 
2068 	MPASS(fd != NULL);
2069 
2070 	if (!fhold(fp))
2071 		return (EBADF);
2072 	FILEDESC_XLOCK(fdp);
2073 	error = fdalloc(td, 0, fd);
2074 	if (__predict_false(error != 0)) {
2075 		FILEDESC_XUNLOCK(fdp);
2076 		fdrop(fp, td);
2077 		return (error);
2078 	}
2079 	_finstall(fdp, fp, *fd, flags, fcaps);
2080 	FILEDESC_XUNLOCK(fdp);
2081 	return (0);
2082 }
2083 
2084 /*
2085  * Build a new filedesc structure from another.
2086  *
2087  * If fdp is not NULL, return with it shared locked.
2088  */
2089 struct filedesc *
2090 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2091 {
2092 	struct filedesc0 *newfdp0;
2093 	struct filedesc *newfdp;
2094 
2095 	if (prepfiles)
2096 		MPASS(lastfile != NULL);
2097 	else
2098 		MPASS(lastfile == NULL);
2099 
2100 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2101 	newfdp = &newfdp0->fd_fd;
2102 
2103 	/* Create the file descriptor table. */
2104 	FILEDESC_LOCK_INIT(newfdp);
2105 	refcount_init(&newfdp->fd_refcnt, 1);
2106 	refcount_init(&newfdp->fd_holdcnt, 1);
2107 	newfdp->fd_map = newfdp0->fd_dmap;
2108 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2109 	newfdp->fd_files->fdt_nfiles = NDFILE;
2110 
2111 	if (fdp == NULL)
2112 		return (newfdp);
2113 
2114 	FILEDESC_SLOCK(fdp);
2115 	if (!prepfiles) {
2116 		FILEDESC_SUNLOCK(fdp);
2117 		return (newfdp);
2118 	}
2119 
2120 	for (;;) {
2121 		*lastfile = fdlastfile(fdp);
2122 		if (*lastfile < newfdp->fd_nfiles)
2123 			break;
2124 		FILEDESC_SUNLOCK(fdp);
2125 		fdgrowtable(newfdp, *lastfile + 1);
2126 		FILEDESC_SLOCK(fdp);
2127 	}
2128 
2129 	return (newfdp);
2130 }
2131 
2132 /*
2133  * Build a pwddesc structure from another.
2134  * Copy the current, root, and jail root vnode references.
2135  *
2136  * If pdp is not NULL, return with it shared locked.
2137  */
2138 struct pwddesc *
2139 pdinit(struct pwddesc *pdp, bool keeplock)
2140 {
2141 	struct pwddesc *newpdp;
2142 	struct pwd *newpwd;
2143 
2144 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2145 
2146 	PWDDESC_LOCK_INIT(newpdp);
2147 	refcount_init(&newpdp->pd_refcount, 1);
2148 	newpdp->pd_cmask = CMASK;
2149 
2150 	if (pdp == NULL) {
2151 		newpwd = pwd_alloc();
2152 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2153 		return (newpdp);
2154 	}
2155 
2156 	PWDDESC_XLOCK(pdp);
2157 	newpwd = pwd_hold_pwddesc(pdp);
2158 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2159 	if (!keeplock)
2160 		PWDDESC_XUNLOCK(pdp);
2161 	return (newpdp);
2162 }
2163 
2164 static struct filedesc *
2165 fdhold(struct proc *p)
2166 {
2167 	struct filedesc *fdp;
2168 
2169 	PROC_LOCK_ASSERT(p, MA_OWNED);
2170 	fdp = p->p_fd;
2171 	if (fdp != NULL)
2172 		refcount_acquire(&fdp->fd_holdcnt);
2173 	return (fdp);
2174 }
2175 
2176 static struct pwddesc *
2177 pdhold(struct proc *p)
2178 {
2179 	struct pwddesc *pdp;
2180 
2181 	PROC_LOCK_ASSERT(p, MA_OWNED);
2182 	pdp = p->p_pd;
2183 	if (pdp != NULL)
2184 		refcount_acquire(&pdp->pd_refcount);
2185 	return (pdp);
2186 }
2187 
2188 static void
2189 fddrop(struct filedesc *fdp)
2190 {
2191 
2192 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2193 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2194 			return;
2195 	}
2196 
2197 	FILEDESC_LOCK_DESTROY(fdp);
2198 	uma_zfree(filedesc0_zone, fdp);
2199 }
2200 
2201 static void
2202 pddrop(struct pwddesc *pdp)
2203 {
2204 	struct pwd *pwd;
2205 
2206 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2207 		return;
2208 
2209 	PWDDESC_XLOCK(pdp);
2210 	if (refcount_release(&pdp->pd_refcount) == 0) {
2211 		PWDDESC_XUNLOCK(pdp);
2212 		return;
2213 	}
2214 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2215 	pwd_set(pdp, NULL);
2216 	PWDDESC_XUNLOCK(pdp);
2217 	pwd_drop(pwd);
2218 
2219 	PWDDESC_LOCK_DESTROY(pdp);
2220 	free(pdp, M_PWDDESC);
2221 }
2222 
2223 /*
2224  * Share a filedesc structure.
2225  */
2226 struct filedesc *
2227 fdshare(struct filedesc *fdp)
2228 {
2229 
2230 	refcount_acquire(&fdp->fd_refcnt);
2231 	return (fdp);
2232 }
2233 
2234 /*
2235  * Share a pwddesc structure.
2236  */
2237 struct pwddesc *
2238 pdshare(struct pwddesc *pdp)
2239 {
2240 	refcount_acquire(&pdp->pd_refcount);
2241 	return (pdp);
2242 }
2243 
2244 /*
2245  * Unshare a filedesc structure, if necessary by making a copy
2246  */
2247 void
2248 fdunshare(struct thread *td)
2249 {
2250 	struct filedesc *tmp;
2251 	struct proc *p = td->td_proc;
2252 
2253 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2254 		return;
2255 
2256 	tmp = fdcopy(p->p_fd);
2257 	fdescfree(td);
2258 	p->p_fd = tmp;
2259 }
2260 
2261 /*
2262  * Unshare a pwddesc structure.
2263  */
2264 void
2265 pdunshare(struct thread *td)
2266 {
2267 	struct pwddesc *pdp;
2268 	struct proc *p;
2269 
2270 	p = td->td_proc;
2271 	/* Not shared. */
2272 	if (p->p_pd->pd_refcount == 1)
2273 		return;
2274 
2275 	pdp = pdcopy(p->p_pd);
2276 	pdescfree(td);
2277 	p->p_pd = pdp;
2278 }
2279 
2280 void
2281 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2282 {
2283 
2284 	fdescfree(td);
2285 	td->td_proc->p_fd = fdp;
2286 }
2287 
2288 /*
2289  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2290  * this is to ease callers, not catch errors.
2291  */
2292 struct filedesc *
2293 fdcopy(struct filedesc *fdp)
2294 {
2295 	struct filedesc *newfdp;
2296 	struct filedescent *nfde, *ofde;
2297 	int i, lastfile;
2298 
2299 	MPASS(fdp != NULL);
2300 
2301 	newfdp = fdinit(fdp, true, &lastfile);
2302 	/* copy all passable descriptors (i.e. not kqueue) */
2303 	newfdp->fd_freefile = -1;
2304 	for (i = 0; i <= lastfile; ++i) {
2305 		ofde = &fdp->fd_ofiles[i];
2306 		if (ofde->fde_file == NULL ||
2307 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2308 		    !fhold(ofde->fde_file)) {
2309 			if (newfdp->fd_freefile == -1)
2310 				newfdp->fd_freefile = i;
2311 			continue;
2312 		}
2313 		nfde = &newfdp->fd_ofiles[i];
2314 		*nfde = *ofde;
2315 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2316 		fdused_init(newfdp, i);
2317 	}
2318 	if (newfdp->fd_freefile == -1)
2319 		newfdp->fd_freefile = i;
2320 	FILEDESC_SUNLOCK(fdp);
2321 	return (newfdp);
2322 }
2323 
2324 /*
2325  * Copy a pwddesc structure.
2326  */
2327 struct pwddesc *
2328 pdcopy(struct pwddesc *pdp)
2329 {
2330 	struct pwddesc *newpdp;
2331 
2332 	MPASS(pdp != NULL);
2333 
2334 	newpdp = pdinit(pdp, true);
2335 	newpdp->pd_cmask = pdp->pd_cmask;
2336 	PWDDESC_XUNLOCK(pdp);
2337 	return (newpdp);
2338 }
2339 
2340 /*
2341  * Copies a filedesc structure, while remapping all file descriptors
2342  * stored inside using a translation table.
2343  *
2344  * File descriptors are copied over to the new file descriptor table,
2345  * regardless of whether the close-on-exec flag is set.
2346  */
2347 int
2348 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2349     struct filedesc **ret)
2350 {
2351 	struct filedesc *newfdp;
2352 	struct filedescent *nfde, *ofde;
2353 	int error, i, lastfile;
2354 
2355 	MPASS(fdp != NULL);
2356 
2357 	newfdp = fdinit(fdp, true, &lastfile);
2358 	if (nfds > lastfile + 1) {
2359 		/* New table cannot be larger than the old one. */
2360 		error = E2BIG;
2361 		goto bad;
2362 	}
2363 	/* Copy all passable descriptors (i.e. not kqueue). */
2364 	newfdp->fd_freefile = nfds;
2365 	for (i = 0; i < nfds; ++i) {
2366 		if (fds[i] < 0 || fds[i] > lastfile) {
2367 			/* File descriptor out of bounds. */
2368 			error = EBADF;
2369 			goto bad;
2370 		}
2371 		ofde = &fdp->fd_ofiles[fds[i]];
2372 		if (ofde->fde_file == NULL) {
2373 			/* Unused file descriptor. */
2374 			error = EBADF;
2375 			goto bad;
2376 		}
2377 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2378 			/* File descriptor cannot be passed. */
2379 			error = EINVAL;
2380 			goto bad;
2381 		}
2382 		if (!fhold(ofde->fde_file)) {
2383 			error = EBADF;
2384 			goto bad;
2385 		}
2386 		nfde = &newfdp->fd_ofiles[i];
2387 		*nfde = *ofde;
2388 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2389 		fdused_init(newfdp, i);
2390 	}
2391 	FILEDESC_SUNLOCK(fdp);
2392 	*ret = newfdp;
2393 	return (0);
2394 bad:
2395 	FILEDESC_SUNLOCK(fdp);
2396 	fdescfree_remapped(newfdp);
2397 	return (error);
2398 }
2399 
2400 /*
2401  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2402  * one of processes using it exits) and the table used to be shared.
2403  */
2404 static void
2405 fdclearlocks(struct thread *td)
2406 {
2407 	struct filedesc *fdp;
2408 	struct filedesc_to_leader *fdtol;
2409 	struct flock lf;
2410 	struct file *fp;
2411 	struct proc *p;
2412 	struct vnode *vp;
2413 	int i, lastfile;
2414 
2415 	p = td->td_proc;
2416 	fdp = p->p_fd;
2417 	fdtol = p->p_fdtol;
2418 	MPASS(fdtol != NULL);
2419 
2420 	FILEDESC_XLOCK(fdp);
2421 	KASSERT(fdtol->fdl_refcount > 0,
2422 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2423 	    fdtol->fdl_refcount));
2424 	if (fdtol->fdl_refcount == 1 &&
2425 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2426 		lastfile = fdlastfile(fdp);
2427 		for (i = 0; i <= lastfile; i++) {
2428 			fp = fdp->fd_ofiles[i].fde_file;
2429 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2430 			    !fhold(fp))
2431 				continue;
2432 			FILEDESC_XUNLOCK(fdp);
2433 			lf.l_whence = SEEK_SET;
2434 			lf.l_start = 0;
2435 			lf.l_len = 0;
2436 			lf.l_type = F_UNLCK;
2437 			vp = fp->f_vnode;
2438 			(void) VOP_ADVLOCK(vp,
2439 			    (caddr_t)p->p_leader, F_UNLCK,
2440 			    &lf, F_POSIX);
2441 			FILEDESC_XLOCK(fdp);
2442 			fdrop(fp, td);
2443 		}
2444 	}
2445 retry:
2446 	if (fdtol->fdl_refcount == 1) {
2447 		if (fdp->fd_holdleaderscount > 0 &&
2448 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2449 			/*
2450 			 * close() or kern_dup() has cleared a reference
2451 			 * in a shared file descriptor table.
2452 			 */
2453 			fdp->fd_holdleaderswakeup = 1;
2454 			sx_sleep(&fdp->fd_holdleaderscount,
2455 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2456 			goto retry;
2457 		}
2458 		if (fdtol->fdl_holdcount > 0) {
2459 			/*
2460 			 * Ensure that fdtol->fdl_leader remains
2461 			 * valid in closef().
2462 			 */
2463 			fdtol->fdl_wakeup = 1;
2464 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2465 			    "fdlhold", 0);
2466 			goto retry;
2467 		}
2468 	}
2469 	fdtol->fdl_refcount--;
2470 	if (fdtol->fdl_refcount == 0 &&
2471 	    fdtol->fdl_holdcount == 0) {
2472 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2473 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2474 	} else
2475 		fdtol = NULL;
2476 	p->p_fdtol = NULL;
2477 	FILEDESC_XUNLOCK(fdp);
2478 	if (fdtol != NULL)
2479 		free(fdtol, M_FILEDESC_TO_LEADER);
2480 }
2481 
2482 /*
2483  * Release a filedesc structure.
2484  */
2485 static void
2486 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2487 {
2488 	struct filedesc0 *fdp0;
2489 	struct freetable *ft, *tft;
2490 	struct filedescent *fde;
2491 	struct file *fp;
2492 	int i, lastfile;
2493 
2494 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2495 	    ("%s: fd table %p carries references", __func__, fdp));
2496 
2497 	/*
2498 	 * Serialize with threads iterating over the table, if any.
2499 	 */
2500 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2501 		FILEDESC_XLOCK(fdp);
2502 		FILEDESC_XUNLOCK(fdp);
2503 	}
2504 
2505 	lastfile = fdlastfile_single(fdp);
2506 	for (i = 0; i <= lastfile; i++) {
2507 		fde = &fdp->fd_ofiles[i];
2508 		fp = fde->fde_file;
2509 		if (fp != NULL) {
2510 			fdefree_last(fde);
2511 			if (needclose)
2512 				(void) closef(fp, td);
2513 			else
2514 				fdrop(fp, td);
2515 		}
2516 	}
2517 
2518 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2519 		free(fdp->fd_map, M_FILEDESC);
2520 	if (fdp->fd_nfiles > NDFILE)
2521 		free(fdp->fd_files, M_FILEDESC);
2522 
2523 	fdp0 = (struct filedesc0 *)fdp;
2524 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2525 		free(ft->ft_table, M_FILEDESC);
2526 
2527 	fddrop(fdp);
2528 }
2529 
2530 void
2531 fdescfree(struct thread *td)
2532 {
2533 	struct proc *p;
2534 	struct filedesc *fdp;
2535 
2536 	p = td->td_proc;
2537 	fdp = p->p_fd;
2538 	MPASS(fdp != NULL);
2539 
2540 #ifdef RACCT
2541 	if (RACCT_ENABLED())
2542 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2543 #endif
2544 
2545 	if (p->p_fdtol != NULL)
2546 		fdclearlocks(td);
2547 
2548 	PROC_LOCK(p);
2549 	p->p_fd = NULL;
2550 	PROC_UNLOCK(p);
2551 
2552 	if (refcount_release(&fdp->fd_refcnt) == 0)
2553 		return;
2554 
2555 	fdescfree_fds(td, fdp, 1);
2556 }
2557 
2558 void
2559 pdescfree(struct thread *td)
2560 {
2561 	struct proc *p;
2562 	struct pwddesc *pdp;
2563 
2564 	p = td->td_proc;
2565 	pdp = p->p_pd;
2566 	MPASS(pdp != NULL);
2567 
2568 	PROC_LOCK(p);
2569 	p->p_pd = NULL;
2570 	PROC_UNLOCK(p);
2571 
2572 	pddrop(pdp);
2573 }
2574 
2575 void
2576 fdescfree_remapped(struct filedesc *fdp)
2577 {
2578 #ifdef INVARIANTS
2579 	/* fdescfree_fds() asserts that fd_refcnt == 0. */
2580 	if (!refcount_release(&fdp->fd_refcnt))
2581 		panic("%s: fd table %p has extra references", __func__, fdp);
2582 #endif
2583 	fdescfree_fds(curthread, fdp, 0);
2584 }
2585 
2586 /*
2587  * For setugid programs, we don't want to people to use that setugidness
2588  * to generate error messages which write to a file which otherwise would
2589  * otherwise be off-limits to the process.  We check for filesystems where
2590  * the vnode can change out from under us after execve (like [lin]procfs).
2591  *
2592  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2593  * sufficient.  We also don't check for setugidness since we know we are.
2594  */
2595 static bool
2596 is_unsafe(struct file *fp)
2597 {
2598 	struct vnode *vp;
2599 
2600 	if (fp->f_type != DTYPE_VNODE)
2601 		return (false);
2602 
2603 	vp = fp->f_vnode;
2604 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2605 }
2606 
2607 /*
2608  * Make this setguid thing safe, if at all possible.
2609  */
2610 void
2611 fdsetugidsafety(struct thread *td)
2612 {
2613 	struct filedesc *fdp;
2614 	struct file *fp;
2615 	int i;
2616 
2617 	fdp = td->td_proc->p_fd;
2618 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2619 	    ("the fdtable should not be shared"));
2620 	MPASS(fdp->fd_nfiles >= 3);
2621 	for (i = 0; i <= 2; i++) {
2622 		fp = fdp->fd_ofiles[i].fde_file;
2623 		if (fp != NULL && is_unsafe(fp)) {
2624 			FILEDESC_XLOCK(fdp);
2625 			knote_fdclose(td, i);
2626 			/*
2627 			 * NULL-out descriptor prior to close to avoid
2628 			 * a race while close blocks.
2629 			 */
2630 			fdfree(fdp, i);
2631 			FILEDESC_XUNLOCK(fdp);
2632 			(void) closef(fp, td);
2633 		}
2634 	}
2635 }
2636 
2637 /*
2638  * If a specific file object occupies a specific file descriptor, close the
2639  * file descriptor entry and drop a reference on the file object.  This is a
2640  * convenience function to handle a subsequent error in a function that calls
2641  * falloc() that handles the race that another thread might have closed the
2642  * file descriptor out from under the thread creating the file object.
2643  */
2644 void
2645 fdclose(struct thread *td, struct file *fp, int idx)
2646 {
2647 	struct filedesc *fdp = td->td_proc->p_fd;
2648 
2649 	FILEDESC_XLOCK(fdp);
2650 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2651 		fdfree(fdp, idx);
2652 		FILEDESC_XUNLOCK(fdp);
2653 		fdrop(fp, td);
2654 	} else
2655 		FILEDESC_XUNLOCK(fdp);
2656 }
2657 
2658 /*
2659  * Close any files on exec?
2660  */
2661 void
2662 fdcloseexec(struct thread *td)
2663 {
2664 	struct filedesc *fdp;
2665 	struct filedescent *fde;
2666 	struct file *fp;
2667 	int i, lastfile;
2668 
2669 	fdp = td->td_proc->p_fd;
2670 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2671 	    ("the fdtable should not be shared"));
2672 	lastfile = fdlastfile_single(fdp);
2673 	for (i = 0; i <= lastfile; i++) {
2674 		fde = &fdp->fd_ofiles[i];
2675 		fp = fde->fde_file;
2676 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2677 		    (fde->fde_flags & UF_EXCLOSE))) {
2678 			FILEDESC_XLOCK(fdp);
2679 			fdfree(fdp, i);
2680 			(void) closefp(fdp, i, fp, td, false, false);
2681 			FILEDESC_UNLOCK_ASSERT(fdp);
2682 		}
2683 	}
2684 }
2685 
2686 /*
2687  * It is unsafe for set[ug]id processes to be started with file
2688  * descriptors 0..2 closed, as these descriptors are given implicit
2689  * significance in the Standard C library.  fdcheckstd() will create a
2690  * descriptor referencing /dev/null for each of stdin, stdout, and
2691  * stderr that is not already open.
2692  */
2693 int
2694 fdcheckstd(struct thread *td)
2695 {
2696 	struct filedesc *fdp;
2697 	register_t save;
2698 	int i, error, devnull;
2699 
2700 	fdp = td->td_proc->p_fd;
2701 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2702 	    ("the fdtable should not be shared"));
2703 	MPASS(fdp->fd_nfiles >= 3);
2704 	devnull = -1;
2705 	for (i = 0; i <= 2; i++) {
2706 		if (fdp->fd_ofiles[i].fde_file != NULL)
2707 			continue;
2708 
2709 		save = td->td_retval[0];
2710 		if (devnull != -1) {
2711 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2712 		} else {
2713 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2714 			    UIO_SYSSPACE, O_RDWR, 0);
2715 			if (error == 0) {
2716 				devnull = td->td_retval[0];
2717 				KASSERT(devnull == i, ("we didn't get our fd"));
2718 			}
2719 		}
2720 		td->td_retval[0] = save;
2721 		if (error != 0)
2722 			return (error);
2723 	}
2724 	return (0);
2725 }
2726 
2727 /*
2728  * Internal form of close.  Decrement reference count on file structure.
2729  * Note: td may be NULL when closing a file that was being passed in a
2730  * message.
2731  */
2732 int
2733 closef(struct file *fp, struct thread *td)
2734 {
2735 	struct vnode *vp;
2736 	struct flock lf;
2737 	struct filedesc_to_leader *fdtol;
2738 	struct filedesc *fdp;
2739 
2740 	/*
2741 	 * POSIX record locking dictates that any close releases ALL
2742 	 * locks owned by this process.  This is handled by setting
2743 	 * a flag in the unlock to free ONLY locks obeying POSIX
2744 	 * semantics, and not to free BSD-style file locks.
2745 	 * If the descriptor was in a message, POSIX-style locks
2746 	 * aren't passed with the descriptor, and the thread pointer
2747 	 * will be NULL.  Callers should be careful only to pass a
2748 	 * NULL thread pointer when there really is no owning
2749 	 * context that might have locks, or the locks will be
2750 	 * leaked.
2751 	 */
2752 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2753 		vp = fp->f_vnode;
2754 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2755 			lf.l_whence = SEEK_SET;
2756 			lf.l_start = 0;
2757 			lf.l_len = 0;
2758 			lf.l_type = F_UNLCK;
2759 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2760 			    F_UNLCK, &lf, F_POSIX);
2761 		}
2762 		fdtol = td->td_proc->p_fdtol;
2763 		if (fdtol != NULL) {
2764 			/*
2765 			 * Handle special case where file descriptor table is
2766 			 * shared between multiple process leaders.
2767 			 */
2768 			fdp = td->td_proc->p_fd;
2769 			FILEDESC_XLOCK(fdp);
2770 			for (fdtol = fdtol->fdl_next;
2771 			    fdtol != td->td_proc->p_fdtol;
2772 			    fdtol = fdtol->fdl_next) {
2773 				if ((fdtol->fdl_leader->p_flag &
2774 				    P_ADVLOCK) == 0)
2775 					continue;
2776 				fdtol->fdl_holdcount++;
2777 				FILEDESC_XUNLOCK(fdp);
2778 				lf.l_whence = SEEK_SET;
2779 				lf.l_start = 0;
2780 				lf.l_len = 0;
2781 				lf.l_type = F_UNLCK;
2782 				vp = fp->f_vnode;
2783 				(void) VOP_ADVLOCK(vp,
2784 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2785 				    F_POSIX);
2786 				FILEDESC_XLOCK(fdp);
2787 				fdtol->fdl_holdcount--;
2788 				if (fdtol->fdl_holdcount == 0 &&
2789 				    fdtol->fdl_wakeup != 0) {
2790 					fdtol->fdl_wakeup = 0;
2791 					wakeup(fdtol);
2792 				}
2793 			}
2794 			FILEDESC_XUNLOCK(fdp);
2795 		}
2796 	}
2797 	return (fdrop_close(fp, td));
2798 }
2799 
2800 /*
2801  * Initialize the file pointer with the specified properties.
2802  *
2803  * The ops are set with release semantics to be certain that the flags, type,
2804  * and data are visible when ops is.  This is to prevent ops methods from being
2805  * called with bad data.
2806  */
2807 void
2808 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2809 {
2810 	fp->f_data = data;
2811 	fp->f_flag = flag;
2812 	fp->f_type = type;
2813 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2814 }
2815 
2816 void
2817 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2818 {
2819 	fp->f_seqcount[UIO_READ] = 1;
2820 	fp->f_seqcount[UIO_WRITE] = 1;
2821 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2822 	    data, ops);
2823 }
2824 
2825 int
2826 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2827     struct file **fpp, struct filecaps *havecapsp)
2828 {
2829 	struct filedescent *fde;
2830 	int error;
2831 
2832 	FILEDESC_LOCK_ASSERT(fdp);
2833 
2834 	fde = fdeget_locked(fdp, fd);
2835 	if (fde == NULL) {
2836 		error = EBADF;
2837 		goto out;
2838 	}
2839 
2840 #ifdef CAPABILITIES
2841 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2842 	if (error != 0)
2843 		goto out;
2844 #endif
2845 
2846 	if (havecapsp != NULL)
2847 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2848 
2849 	*fpp = fde->fde_file;
2850 
2851 	error = 0;
2852 out:
2853 	return (error);
2854 }
2855 
2856 int
2857 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2858     struct file **fpp, struct filecaps *havecapsp)
2859 {
2860 	struct filedesc *fdp = td->td_proc->p_fd;
2861 	int error;
2862 #ifndef CAPABILITIES
2863 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2864 	if (havecapsp != NULL && error == 0)
2865 		filecaps_fill(havecapsp);
2866 #else
2867 	struct file *fp;
2868 	seqc_t seq;
2869 
2870 	*fpp = NULL;
2871 	for (;;) {
2872 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2873 		if (error != 0)
2874 			return (error);
2875 
2876 		if (havecapsp != NULL) {
2877 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2878 			    havecapsp, false)) {
2879 				fdrop(fp, td);
2880 				goto get_locked;
2881 			}
2882 		}
2883 
2884 		if (!fd_modified(fdp, fd, seq))
2885 			break;
2886 		fdrop(fp, td);
2887 	}
2888 
2889 	*fpp = fp;
2890 	return (0);
2891 
2892 get_locked:
2893 	FILEDESC_SLOCK(fdp);
2894 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2895 	if (error == 0 && !fhold(*fpp))
2896 		error = EBADF;
2897 	FILEDESC_SUNLOCK(fdp);
2898 #endif
2899 	return (error);
2900 }
2901 
2902 #ifdef CAPABILITIES
2903 int
2904 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2905 {
2906 	const struct filedescent *fde;
2907 	const struct fdescenttbl *fdt;
2908 	struct filedesc *fdp;
2909 	struct file *fp;
2910 	struct vnode *vp;
2911 	const cap_rights_t *haverights;
2912 	cap_rights_t rights;
2913 	seqc_t seq;
2914 
2915 	VFS_SMR_ASSERT_ENTERED();
2916 
2917 	rights = *ndp->ni_rightsneeded;
2918 	cap_rights_set_one(&rights, CAP_LOOKUP);
2919 
2920 	fdp = curproc->p_fd;
2921 	fdt = fdp->fd_files;
2922 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2923 		return (EBADF);
2924 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2925 	fde = &fdt->fdt_ofiles[fd];
2926 	haverights = cap_rights_fde_inline(fde);
2927 	fp = fde->fde_file;
2928 	if (__predict_false(fp == NULL))
2929 		return (EAGAIN);
2930 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2931 		return (EAGAIN);
2932 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2933 	vp = fp->f_vnode;
2934 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2935 		return (EAGAIN);
2936 	}
2937 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2938 		return (EAGAIN);
2939 	}
2940 	/*
2941 	 * Use an acquire barrier to force re-reading of fdt so it is
2942 	 * refreshed for verification.
2943 	 */
2944 	atomic_thread_fence_acq();
2945 	fdt = fdp->fd_files;
2946 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2947 		return (EAGAIN);
2948 	/*
2949 	 * If file descriptor doesn't have all rights,
2950 	 * all lookups relative to it must also be
2951 	 * strictly relative.
2952 	 *
2953 	 * Not yet supported by fast path.
2954 	 */
2955 	CAP_ALL(&rights);
2956 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2957 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2958 	    ndp->ni_filecaps.fc_nioctls != -1) {
2959 #ifdef notyet
2960 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2961 #else
2962 		return (EAGAIN);
2963 #endif
2964 	}
2965 	*vpp = vp;
2966 	return (0);
2967 }
2968 #else
2969 int
2970 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2971 {
2972 	const struct fdescenttbl *fdt;
2973 	struct filedesc *fdp;
2974 	struct file *fp;
2975 	struct vnode *vp;
2976 
2977 	VFS_SMR_ASSERT_ENTERED();
2978 
2979 	fdp = curproc->p_fd;
2980 	fdt = fdp->fd_files;
2981 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2982 		return (EBADF);
2983 	fp = fdt->fdt_ofiles[fd].fde_file;
2984 	if (__predict_false(fp == NULL))
2985 		return (EAGAIN);
2986 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2987 	vp = fp->f_vnode;
2988 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2989 		return (EAGAIN);
2990 	}
2991 	/*
2992 	 * Use an acquire barrier to force re-reading of fdt so it is
2993 	 * refreshed for verification.
2994 	 */
2995 	atomic_thread_fence_acq();
2996 	fdt = fdp->fd_files;
2997 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2998 		return (EAGAIN);
2999 	filecaps_fill(&ndp->ni_filecaps);
3000 	*vpp = vp;
3001 	return (0);
3002 }
3003 #endif
3004 
3005 int
3006 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3007     struct file **fpp, seqc_t *seqp)
3008 {
3009 #ifdef CAPABILITIES
3010 	const struct filedescent *fde;
3011 #endif
3012 	const struct fdescenttbl *fdt;
3013 	struct file *fp;
3014 #ifdef CAPABILITIES
3015 	seqc_t seq;
3016 	cap_rights_t haverights;
3017 	int error;
3018 #endif
3019 
3020 	fdt = fdp->fd_files;
3021 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3022 		return (EBADF);
3023 	/*
3024 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
3025 	 * never raising a refcount above 0.  To accomplish this we have
3026 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3027 	 * must be re-verified once we acquire a reference to be certain
3028 	 * that the identity is still correct and we did not lose a race
3029 	 * due to preemption.
3030 	 */
3031 	for (;;) {
3032 #ifdef CAPABILITIES
3033 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3034 		fde = &fdt->fdt_ofiles[fd];
3035 		haverights = *cap_rights_fde_inline(fde);
3036 		fp = fde->fde_file;
3037 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3038 			continue;
3039 #else
3040 		fp = fdt->fdt_ofiles[fd].fde_file;
3041 #endif
3042 		if (fp == NULL)
3043 			return (EBADF);
3044 #ifdef CAPABILITIES
3045 		error = cap_check_inline(&haverights, needrightsp);
3046 		if (error != 0)
3047 			return (error);
3048 #endif
3049 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3050 			/*
3051 			 * Force a reload. Other thread could reallocate the
3052 			 * table before this fd was closed, so it is possible
3053 			 * that there is a stale fp pointer in cached version.
3054 			 */
3055 			fdt = atomic_load_ptr(&fdp->fd_files);
3056 			continue;
3057 		}
3058 		/*
3059 		 * Use an acquire barrier to force re-reading of fdt so it is
3060 		 * refreshed for verification.
3061 		 */
3062 		atomic_thread_fence_acq();
3063 		fdt = fdp->fd_files;
3064 #ifdef	CAPABILITIES
3065 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3066 #else
3067 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3068 #endif
3069 			break;
3070 		fdrop(fp, curthread);
3071 	}
3072 	*fpp = fp;
3073 	if (seqp != NULL) {
3074 #ifdef CAPABILITIES
3075 		*seqp = seq;
3076 #endif
3077 	}
3078 	return (0);
3079 }
3080 
3081 /*
3082  * See the comments in fget_unlocked_seq for an explanation of how this works.
3083  *
3084  * This is a simplified variant which bails out to the aforementioned routine
3085  * if anything goes wrong. In practice this only happens when userspace is
3086  * racing with itself.
3087  */
3088 int
3089 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3090     struct file **fpp)
3091 {
3092 #ifdef CAPABILITIES
3093 	const struct filedescent *fde;
3094 #endif
3095 	const struct fdescenttbl *fdt;
3096 	struct file *fp;
3097 #ifdef CAPABILITIES
3098 	seqc_t seq;
3099 	const cap_rights_t *haverights;
3100 #endif
3101 
3102 	fdt = fdp->fd_files;
3103 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3104 		return (EBADF);
3105 #ifdef CAPABILITIES
3106 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3107 	fde = &fdt->fdt_ofiles[fd];
3108 	haverights = cap_rights_fde_inline(fde);
3109 	fp = fde->fde_file;
3110 #else
3111 	fp = fdt->fdt_ofiles[fd].fde_file;
3112 #endif
3113 	if (__predict_false(fp == NULL))
3114 		goto out_fallback;
3115 #ifdef CAPABILITIES
3116 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3117 		goto out_fallback;
3118 #endif
3119 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3120 		goto out_fallback;
3121 
3122 	/*
3123 	 * Use an acquire barrier to force re-reading of fdt so it is
3124 	 * refreshed for verification.
3125 	 */
3126 	atomic_thread_fence_acq();
3127 	fdt = fdp->fd_files;
3128 #ifdef	CAPABILITIES
3129 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3130 #else
3131 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3132 #endif
3133 		goto out_fdrop;
3134 	*fpp = fp;
3135 	return (0);
3136 out_fdrop:
3137 	fdrop(fp, curthread);
3138 out_fallback:
3139 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3140 }
3141 
3142 /*
3143  * Extract the file pointer associated with the specified descriptor for the
3144  * current user process.
3145  *
3146  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3147  * returned.
3148  *
3149  * File's rights will be checked against the capability rights mask.
3150  *
3151  * If an error occurred the non-zero error is returned and *fpp is set to
3152  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3153  * responsible for fdrop().
3154  */
3155 static __inline int
3156 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3157     cap_rights_t *needrightsp)
3158 {
3159 	struct filedesc *fdp;
3160 	struct file *fp;
3161 	int error;
3162 
3163 	*fpp = NULL;
3164 	fdp = td->td_proc->p_fd;
3165 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3166 	if (__predict_false(error != 0))
3167 		return (error);
3168 	if (__predict_false(fp->f_ops == &badfileops)) {
3169 		fdrop(fp, td);
3170 		return (EBADF);
3171 	}
3172 
3173 	/*
3174 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3175 	 */
3176 	error = 0;
3177 	switch (flags) {
3178 	case FREAD:
3179 	case FWRITE:
3180 		if ((fp->f_flag & flags) == 0)
3181 			error = EBADF;
3182 		break;
3183 	case FEXEC:
3184 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3185 		    ((fp->f_flag & FWRITE) != 0))
3186 			error = EBADF;
3187 		break;
3188 	case 0:
3189 		break;
3190 	default:
3191 		KASSERT(0, ("wrong flags"));
3192 	}
3193 
3194 	if (error != 0) {
3195 		fdrop(fp, td);
3196 		return (error);
3197 	}
3198 
3199 	*fpp = fp;
3200 	return (0);
3201 }
3202 
3203 int
3204 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3205 {
3206 
3207 	return (_fget(td, fd, fpp, 0, rightsp));
3208 }
3209 
3210 int
3211 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3212     struct file **fpp)
3213 {
3214 	int error;
3215 #ifndef CAPABILITIES
3216 	error = _fget(td, fd, fpp, 0, rightsp);
3217 	if (maxprotp != NULL)
3218 		*maxprotp = VM_PROT_ALL;
3219 	return (error);
3220 #else
3221 	cap_rights_t fdrights;
3222 	struct filedesc *fdp;
3223 	struct file *fp;
3224 	seqc_t seq;
3225 
3226 	*fpp = NULL;
3227 	fdp = td->td_proc->p_fd;
3228 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3229 	for (;;) {
3230 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3231 		if (__predict_false(error != 0))
3232 			return (error);
3233 		if (__predict_false(fp->f_ops == &badfileops)) {
3234 			fdrop(fp, td);
3235 			return (EBADF);
3236 		}
3237 		if (maxprotp != NULL)
3238 			fdrights = *cap_rights(fdp, fd);
3239 		if (!fd_modified(fdp, fd, seq))
3240 			break;
3241 		fdrop(fp, td);
3242 	}
3243 
3244 	/*
3245 	 * If requested, convert capability rights to access flags.
3246 	 */
3247 	if (maxprotp != NULL)
3248 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3249 	*fpp = fp;
3250 	return (0);
3251 #endif
3252 }
3253 
3254 int
3255 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3256 {
3257 
3258 	return (_fget(td, fd, fpp, FREAD, rightsp));
3259 }
3260 
3261 int
3262 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3263 {
3264 
3265 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3266 }
3267 
3268 int
3269 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3270     struct file **fpp)
3271 {
3272 	struct filedesc *fdp = td->td_proc->p_fd;
3273 #ifndef CAPABILITIES
3274 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3275 #else
3276 	struct file *fp;
3277 	int error;
3278 	seqc_t seq;
3279 
3280 	*fpp = NULL;
3281 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3282 	for (;;) {
3283 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3284 		if (error != 0)
3285 			return (error);
3286 		error = cap_fcntl_check(fdp, fd, needfcntl);
3287 		if (!fd_modified(fdp, fd, seq))
3288 			break;
3289 		fdrop(fp, td);
3290 	}
3291 	if (error != 0) {
3292 		fdrop(fp, td);
3293 		return (error);
3294 	}
3295 	*fpp = fp;
3296 	return (0);
3297 #endif
3298 }
3299 
3300 /*
3301  * Like fget() but loads the underlying vnode, or returns an error if the
3302  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3303  * never have VM objects.  The returned vnode will be vref()'d.
3304  *
3305  * XXX: what about the unused flags ?
3306  */
3307 static __inline int
3308 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3309     struct vnode **vpp)
3310 {
3311 	struct file *fp;
3312 	int error;
3313 
3314 	*vpp = NULL;
3315 	error = _fget(td, fd, &fp, flags, needrightsp);
3316 	if (error != 0)
3317 		return (error);
3318 	if (fp->f_vnode == NULL) {
3319 		error = EINVAL;
3320 	} else {
3321 		*vpp = fp->f_vnode;
3322 		vrefact(*vpp);
3323 	}
3324 	fdrop(fp, td);
3325 
3326 	return (error);
3327 }
3328 
3329 int
3330 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3331 {
3332 
3333 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3334 }
3335 
3336 int
3337 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3338     struct filecaps *havecaps, struct vnode **vpp)
3339 {
3340 	struct filecaps caps;
3341 	struct file *fp;
3342 	int error;
3343 
3344 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3345 	if (error != 0)
3346 		return (error);
3347 	if (fp->f_ops == &badfileops) {
3348 		error = EBADF;
3349 		goto out;
3350 	}
3351 	if (fp->f_vnode == NULL) {
3352 		error = EINVAL;
3353 		goto out;
3354 	}
3355 
3356 	*havecaps = caps;
3357 	*vpp = fp->f_vnode;
3358 	vrefact(*vpp);
3359 	fdrop(fp, td);
3360 
3361 	return (0);
3362 out:
3363 	filecaps_free(&caps);
3364 	fdrop(fp, td);
3365 	return (error);
3366 }
3367 
3368 int
3369 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3370 {
3371 
3372 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3373 }
3374 
3375 int
3376 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3377 {
3378 
3379 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3380 }
3381 
3382 #ifdef notyet
3383 int
3384 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3385     struct vnode **vpp)
3386 {
3387 
3388 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3389 }
3390 #endif
3391 
3392 /*
3393  * Handle the last reference to a file being closed.
3394  *
3395  * Without the noinline attribute clang keeps inlining the func thorough this
3396  * file when fdrop is used.
3397  */
3398 int __noinline
3399 _fdrop(struct file *fp, struct thread *td)
3400 {
3401 	int error;
3402 #ifdef INVARIANTS
3403 	int count;
3404 
3405 	count = refcount_load(&fp->f_count);
3406 	if (count != 0)
3407 		panic("fdrop: fp %p count %d", fp, count);
3408 #endif
3409 	error = fo_close(fp, td);
3410 	atomic_subtract_int(&openfiles, 1);
3411 	crfree(fp->f_cred);
3412 	free(fp->f_advice, M_FADVISE);
3413 	uma_zfree(file_zone, fp);
3414 
3415 	return (error);
3416 }
3417 
3418 /*
3419  * Apply an advisory lock on a file descriptor.
3420  *
3421  * Just attempt to get a record lock of the requested type on the entire file
3422  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3423  */
3424 #ifndef _SYS_SYSPROTO_H_
3425 struct flock_args {
3426 	int	fd;
3427 	int	how;
3428 };
3429 #endif
3430 /* ARGSUSED */
3431 int
3432 sys_flock(struct thread *td, struct flock_args *uap)
3433 {
3434 	struct file *fp;
3435 	struct vnode *vp;
3436 	struct flock lf;
3437 	int error;
3438 
3439 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3440 	if (error != 0)
3441 		return (error);
3442 	if (fp->f_type != DTYPE_VNODE) {
3443 		fdrop(fp, td);
3444 		return (EOPNOTSUPP);
3445 	}
3446 
3447 	vp = fp->f_vnode;
3448 	lf.l_whence = SEEK_SET;
3449 	lf.l_start = 0;
3450 	lf.l_len = 0;
3451 	if (uap->how & LOCK_UN) {
3452 		lf.l_type = F_UNLCK;
3453 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3454 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3455 		goto done2;
3456 	}
3457 	if (uap->how & LOCK_EX)
3458 		lf.l_type = F_WRLCK;
3459 	else if (uap->how & LOCK_SH)
3460 		lf.l_type = F_RDLCK;
3461 	else {
3462 		error = EBADF;
3463 		goto done2;
3464 	}
3465 	atomic_set_int(&fp->f_flag, FHASLOCK);
3466 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3467 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3468 done2:
3469 	fdrop(fp, td);
3470 	return (error);
3471 }
3472 /*
3473  * Duplicate the specified descriptor to a free descriptor.
3474  */
3475 int
3476 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3477     int openerror, int *indxp)
3478 {
3479 	struct filedescent *newfde, *oldfde;
3480 	struct file *fp;
3481 	u_long *ioctls;
3482 	int error, indx;
3483 
3484 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3485 	    ("unexpected error %d in %s", openerror, __func__));
3486 
3487 	/*
3488 	 * If the to-be-dup'd fd number is greater than the allowed number
3489 	 * of file descriptors, or the fd to be dup'd has already been
3490 	 * closed, then reject.
3491 	 */
3492 	FILEDESC_XLOCK(fdp);
3493 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3494 		FILEDESC_XUNLOCK(fdp);
3495 		return (EBADF);
3496 	}
3497 
3498 	error = fdalloc(td, 0, &indx);
3499 	if (error != 0) {
3500 		FILEDESC_XUNLOCK(fdp);
3501 		return (error);
3502 	}
3503 
3504 	/*
3505 	 * There are two cases of interest here.
3506 	 *
3507 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3508 	 *
3509 	 * For ENXIO steal away the file structure from (dfd) and store it in
3510 	 * (indx).  (dfd) is effectively closed by this operation.
3511 	 */
3512 	switch (openerror) {
3513 	case ENODEV:
3514 		/*
3515 		 * Check that the mode the file is being opened for is a
3516 		 * subset of the mode of the existing descriptor.
3517 		 */
3518 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3519 			fdunused(fdp, indx);
3520 			FILEDESC_XUNLOCK(fdp);
3521 			return (EACCES);
3522 		}
3523 		if (!fhold(fp)) {
3524 			fdunused(fdp, indx);
3525 			FILEDESC_XUNLOCK(fdp);
3526 			return (EBADF);
3527 		}
3528 		newfde = &fdp->fd_ofiles[indx];
3529 		oldfde = &fdp->fd_ofiles[dfd];
3530 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3531 #ifdef CAPABILITIES
3532 		seqc_write_begin(&newfde->fde_seqc);
3533 #endif
3534 		memcpy(newfde, oldfde, fde_change_size);
3535 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3536 		    ioctls);
3537 #ifdef CAPABILITIES
3538 		seqc_write_end(&newfde->fde_seqc);
3539 #endif
3540 		break;
3541 	case ENXIO:
3542 		/*
3543 		 * Steal away the file pointer from dfd and stuff it into indx.
3544 		 */
3545 		newfde = &fdp->fd_ofiles[indx];
3546 		oldfde = &fdp->fd_ofiles[dfd];
3547 #ifdef CAPABILITIES
3548 		seqc_write_begin(&newfde->fde_seqc);
3549 #endif
3550 		memcpy(newfde, oldfde, fde_change_size);
3551 		oldfde->fde_file = NULL;
3552 		fdunused(fdp, dfd);
3553 #ifdef CAPABILITIES
3554 		seqc_write_end(&newfde->fde_seqc);
3555 #endif
3556 		break;
3557 	}
3558 	FILEDESC_XUNLOCK(fdp);
3559 	*indxp = indx;
3560 	return (0);
3561 }
3562 
3563 /*
3564  * This sysctl determines if we will allow a process to chroot(2) if it
3565  * has a directory open:
3566  *	0: disallowed for all processes.
3567  *	1: allowed for processes that were not already chroot(2)'ed.
3568  *	2: allowed for all processes.
3569  */
3570 
3571 static int chroot_allow_open_directories = 1;
3572 
3573 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3574     &chroot_allow_open_directories, 0,
3575     "Allow a process to chroot(2) if it has a directory open");
3576 
3577 /*
3578  * Helper function for raised chroot(2) security function:  Refuse if
3579  * any filedescriptors are open directories.
3580  */
3581 static int
3582 chroot_refuse_vdir_fds(struct filedesc *fdp)
3583 {
3584 	struct vnode *vp;
3585 	struct file *fp;
3586 	int fd, lastfile;
3587 
3588 	FILEDESC_LOCK_ASSERT(fdp);
3589 
3590 	lastfile = fdlastfile(fdp);
3591 	for (fd = 0; fd <= lastfile; fd++) {
3592 		fp = fget_locked(fdp, fd);
3593 		if (fp == NULL)
3594 			continue;
3595 		if (fp->f_type == DTYPE_VNODE) {
3596 			vp = fp->f_vnode;
3597 			if (vp->v_type == VDIR)
3598 				return (EPERM);
3599 		}
3600 	}
3601 	return (0);
3602 }
3603 
3604 static void
3605 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3606 {
3607 
3608 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3609 		vrefact(oldpwd->pwd_cdir);
3610 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3611 	}
3612 
3613 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3614 		vrefact(oldpwd->pwd_rdir);
3615 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3616 	}
3617 
3618 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3619 		vrefact(oldpwd->pwd_jdir);
3620 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3621 	}
3622 }
3623 
3624 struct pwd *
3625 pwd_hold_pwddesc(struct pwddesc *pdp)
3626 {
3627 	struct pwd *pwd;
3628 
3629 	PWDDESC_ASSERT_XLOCKED(pdp);
3630 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3631 	if (pwd != NULL)
3632 		refcount_acquire(&pwd->pwd_refcount);
3633 	return (pwd);
3634 }
3635 
3636 bool
3637 pwd_hold_smr(struct pwd *pwd)
3638 {
3639 
3640 	MPASS(pwd != NULL);
3641 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3642 		return (true);
3643 	}
3644 	return (false);
3645 }
3646 
3647 struct pwd *
3648 pwd_hold(struct thread *td)
3649 {
3650 	struct pwddesc *pdp;
3651 	struct pwd *pwd;
3652 
3653 	pdp = td->td_proc->p_pd;
3654 
3655 	vfs_smr_enter();
3656 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3657 	if (pwd_hold_smr(pwd)) {
3658 		vfs_smr_exit();
3659 		return (pwd);
3660 	}
3661 	vfs_smr_exit();
3662 	PWDDESC_XLOCK(pdp);
3663 	pwd = pwd_hold_pwddesc(pdp);
3664 	MPASS(pwd != NULL);
3665 	PWDDESC_XUNLOCK(pdp);
3666 	return (pwd);
3667 }
3668 
3669 static struct pwd *
3670 pwd_alloc(void)
3671 {
3672 	struct pwd *pwd;
3673 
3674 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3675 	bzero(pwd, sizeof(*pwd));
3676 	refcount_init(&pwd->pwd_refcount, 1);
3677 	return (pwd);
3678 }
3679 
3680 void
3681 pwd_drop(struct pwd *pwd)
3682 {
3683 
3684 	if (!refcount_release(&pwd->pwd_refcount))
3685 		return;
3686 
3687 	if (pwd->pwd_cdir != NULL)
3688 		vrele(pwd->pwd_cdir);
3689 	if (pwd->pwd_rdir != NULL)
3690 		vrele(pwd->pwd_rdir);
3691 	if (pwd->pwd_jdir != NULL)
3692 		vrele(pwd->pwd_jdir);
3693 	uma_zfree_smr(pwd_zone, pwd);
3694 }
3695 
3696 /*
3697 * Common routine for kern_chroot() and jail_attach().  The caller is
3698 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3699 * authorize this operation.
3700 */
3701 int
3702 pwd_chroot(struct thread *td, struct vnode *vp)
3703 {
3704 	struct pwddesc *pdp;
3705 	struct filedesc *fdp;
3706 	struct pwd *newpwd, *oldpwd;
3707 	int error;
3708 
3709 	fdp = td->td_proc->p_fd;
3710 	pdp = td->td_proc->p_pd;
3711 	newpwd = pwd_alloc();
3712 	FILEDESC_SLOCK(fdp);
3713 	PWDDESC_XLOCK(pdp);
3714 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3715 	if (chroot_allow_open_directories == 0 ||
3716 	    (chroot_allow_open_directories == 1 &&
3717 	    oldpwd->pwd_rdir != rootvnode)) {
3718 		error = chroot_refuse_vdir_fds(fdp);
3719 		FILEDESC_SUNLOCK(fdp);
3720 		if (error != 0) {
3721 			PWDDESC_XUNLOCK(pdp);
3722 			pwd_drop(newpwd);
3723 			return (error);
3724 		}
3725 	} else {
3726 		FILEDESC_SUNLOCK(fdp);
3727 	}
3728 
3729 	vrefact(vp);
3730 	newpwd->pwd_rdir = vp;
3731 	if (oldpwd->pwd_jdir == NULL) {
3732 		vrefact(vp);
3733 		newpwd->pwd_jdir = vp;
3734 	}
3735 	pwd_fill(oldpwd, newpwd);
3736 	pwd_set(pdp, newpwd);
3737 	PWDDESC_XUNLOCK(pdp);
3738 	pwd_drop(oldpwd);
3739 	return (0);
3740 }
3741 
3742 void
3743 pwd_chdir(struct thread *td, struct vnode *vp)
3744 {
3745 	struct pwddesc *pdp;
3746 	struct pwd *newpwd, *oldpwd;
3747 
3748 	VNPASS(vp->v_usecount > 0, vp);
3749 
3750 	newpwd = pwd_alloc();
3751 	pdp = td->td_proc->p_pd;
3752 	PWDDESC_XLOCK(pdp);
3753 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3754 	newpwd->pwd_cdir = vp;
3755 	pwd_fill(oldpwd, newpwd);
3756 	pwd_set(pdp, newpwd);
3757 	PWDDESC_XUNLOCK(pdp);
3758 	pwd_drop(oldpwd);
3759 }
3760 
3761 void
3762 pwd_ensure_dirs(void)
3763 {
3764 	struct pwddesc *pdp;
3765 	struct pwd *oldpwd, *newpwd;
3766 
3767 	pdp = curproc->p_pd;
3768 	PWDDESC_XLOCK(pdp);
3769 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3770 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3771 		PWDDESC_XUNLOCK(pdp);
3772 		return;
3773 	}
3774 	PWDDESC_XUNLOCK(pdp);
3775 
3776 	newpwd = pwd_alloc();
3777 	PWDDESC_XLOCK(pdp);
3778 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3779 	pwd_fill(oldpwd, newpwd);
3780 	if (newpwd->pwd_cdir == NULL) {
3781 		vrefact(rootvnode);
3782 		newpwd->pwd_cdir = rootvnode;
3783 	}
3784 	if (newpwd->pwd_rdir == NULL) {
3785 		vrefact(rootvnode);
3786 		newpwd->pwd_rdir = rootvnode;
3787 	}
3788 	pwd_set(pdp, newpwd);
3789 	PWDDESC_XUNLOCK(pdp);
3790 	pwd_drop(oldpwd);
3791 }
3792 
3793 void
3794 pwd_set_rootvnode(void)
3795 {
3796 	struct pwddesc *pdp;
3797 	struct pwd *oldpwd, *newpwd;
3798 
3799 	pdp = curproc->p_pd;
3800 
3801 	newpwd = pwd_alloc();
3802 	PWDDESC_XLOCK(pdp);
3803 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3804 	vrefact(rootvnode);
3805 	newpwd->pwd_cdir = rootvnode;
3806 	vrefact(rootvnode);
3807 	newpwd->pwd_rdir = rootvnode;
3808 	pwd_fill(oldpwd, newpwd);
3809 	pwd_set(pdp, newpwd);
3810 	PWDDESC_XUNLOCK(pdp);
3811 	pwd_drop(oldpwd);
3812 }
3813 
3814 /*
3815  * Scan all active processes and prisons to see if any of them have a current
3816  * or root directory of `olddp'. If so, replace them with the new mount point.
3817  */
3818 void
3819 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3820 {
3821 	struct pwddesc *pdp;
3822 	struct pwd *newpwd, *oldpwd;
3823 	struct prison *pr;
3824 	struct proc *p;
3825 	int nrele;
3826 
3827 	if (vrefcnt(olddp) == 1)
3828 		return;
3829 	nrele = 0;
3830 	newpwd = pwd_alloc();
3831 	sx_slock(&allproc_lock);
3832 	FOREACH_PROC_IN_SYSTEM(p) {
3833 		PROC_LOCK(p);
3834 		pdp = pdhold(p);
3835 		PROC_UNLOCK(p);
3836 		if (pdp == NULL)
3837 			continue;
3838 		PWDDESC_XLOCK(pdp);
3839 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3840 		if (oldpwd == NULL ||
3841 		    (oldpwd->pwd_cdir != olddp &&
3842 		    oldpwd->pwd_rdir != olddp &&
3843 		    oldpwd->pwd_jdir != olddp)) {
3844 			PWDDESC_XUNLOCK(pdp);
3845 			pddrop(pdp);
3846 			continue;
3847 		}
3848 		if (oldpwd->pwd_cdir == olddp) {
3849 			vrefact(newdp);
3850 			newpwd->pwd_cdir = newdp;
3851 		}
3852 		if (oldpwd->pwd_rdir == olddp) {
3853 			vrefact(newdp);
3854 			newpwd->pwd_rdir = newdp;
3855 		}
3856 		if (oldpwd->pwd_jdir == olddp) {
3857 			vrefact(newdp);
3858 			newpwd->pwd_jdir = newdp;
3859 		}
3860 		pwd_fill(oldpwd, newpwd);
3861 		pwd_set(pdp, newpwd);
3862 		PWDDESC_XUNLOCK(pdp);
3863 		pwd_drop(oldpwd);
3864 		pddrop(pdp);
3865 		newpwd = pwd_alloc();
3866 	}
3867 	sx_sunlock(&allproc_lock);
3868 	pwd_drop(newpwd);
3869 	if (rootvnode == olddp) {
3870 		vrefact(newdp);
3871 		rootvnode = newdp;
3872 		nrele++;
3873 	}
3874 	mtx_lock(&prison0.pr_mtx);
3875 	if (prison0.pr_root == olddp) {
3876 		vrefact(newdp);
3877 		prison0.pr_root = newdp;
3878 		nrele++;
3879 	}
3880 	mtx_unlock(&prison0.pr_mtx);
3881 	sx_slock(&allprison_lock);
3882 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3883 		mtx_lock(&pr->pr_mtx);
3884 		if (pr->pr_root == olddp) {
3885 			vrefact(newdp);
3886 			pr->pr_root = newdp;
3887 			nrele++;
3888 		}
3889 		mtx_unlock(&pr->pr_mtx);
3890 	}
3891 	sx_sunlock(&allprison_lock);
3892 	while (nrele--)
3893 		vrele(olddp);
3894 }
3895 
3896 struct filedesc_to_leader *
3897 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3898 {
3899 	struct filedesc_to_leader *fdtol;
3900 
3901 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3902 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3903 	fdtol->fdl_refcount = 1;
3904 	fdtol->fdl_holdcount = 0;
3905 	fdtol->fdl_wakeup = 0;
3906 	fdtol->fdl_leader = leader;
3907 	if (old != NULL) {
3908 		FILEDESC_XLOCK(fdp);
3909 		fdtol->fdl_next = old->fdl_next;
3910 		fdtol->fdl_prev = old;
3911 		old->fdl_next = fdtol;
3912 		fdtol->fdl_next->fdl_prev = fdtol;
3913 		FILEDESC_XUNLOCK(fdp);
3914 	} else {
3915 		fdtol->fdl_next = fdtol;
3916 		fdtol->fdl_prev = fdtol;
3917 	}
3918 	return (fdtol);
3919 }
3920 
3921 static int
3922 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3923 {
3924 	NDSLOTTYPE *map;
3925 	struct filedesc *fdp;
3926 	int count, off, minoff;
3927 
3928 	if (*(int *)arg1 != 0)
3929 		return (EINVAL);
3930 
3931 	fdp = curproc->p_fd;
3932 	count = 0;
3933 	FILEDESC_SLOCK(fdp);
3934 	map = fdp->fd_map;
3935 	off = NDSLOT(fdp->fd_nfiles - 1);
3936 	for (minoff = NDSLOT(0); off >= minoff; --off)
3937 		count += bitcountl(map[off]);
3938 	FILEDESC_SUNLOCK(fdp);
3939 
3940 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3941 }
3942 
3943 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3944     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3945     "Number of open file descriptors");
3946 
3947 /*
3948  * Get file structures globally.
3949  */
3950 static int
3951 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3952 {
3953 	struct xfile xf;
3954 	struct filedesc *fdp;
3955 	struct file *fp;
3956 	struct proc *p;
3957 	int error, n, lastfile;
3958 
3959 	error = sysctl_wire_old_buffer(req, 0);
3960 	if (error != 0)
3961 		return (error);
3962 	if (req->oldptr == NULL) {
3963 		n = 0;
3964 		sx_slock(&allproc_lock);
3965 		FOREACH_PROC_IN_SYSTEM(p) {
3966 			PROC_LOCK(p);
3967 			if (p->p_state == PRS_NEW) {
3968 				PROC_UNLOCK(p);
3969 				continue;
3970 			}
3971 			fdp = fdhold(p);
3972 			PROC_UNLOCK(p);
3973 			if (fdp == NULL)
3974 				continue;
3975 			/* overestimates sparse tables. */
3976 			n += fdp->fd_nfiles;
3977 			fddrop(fdp);
3978 		}
3979 		sx_sunlock(&allproc_lock);
3980 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3981 	}
3982 	error = 0;
3983 	bzero(&xf, sizeof(xf));
3984 	xf.xf_size = sizeof(xf);
3985 	sx_slock(&allproc_lock);
3986 	FOREACH_PROC_IN_SYSTEM(p) {
3987 		PROC_LOCK(p);
3988 		if (p->p_state == PRS_NEW) {
3989 			PROC_UNLOCK(p);
3990 			continue;
3991 		}
3992 		if (p_cansee(req->td, p) != 0) {
3993 			PROC_UNLOCK(p);
3994 			continue;
3995 		}
3996 		xf.xf_pid = p->p_pid;
3997 		xf.xf_uid = p->p_ucred->cr_uid;
3998 		fdp = fdhold(p);
3999 		PROC_UNLOCK(p);
4000 		if (fdp == NULL)
4001 			continue;
4002 		FILEDESC_SLOCK(fdp);
4003 		lastfile = fdlastfile(fdp);
4004 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4005 		    n++) {
4006 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4007 				continue;
4008 			xf.xf_fd = n;
4009 			xf.xf_file = (uintptr_t)fp;
4010 			xf.xf_data = (uintptr_t)fp->f_data;
4011 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4012 			xf.xf_type = (uintptr_t)fp->f_type;
4013 			xf.xf_count = refcount_load(&fp->f_count);
4014 			xf.xf_msgcount = 0;
4015 			xf.xf_offset = foffset_get(fp);
4016 			xf.xf_flag = fp->f_flag;
4017 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4018 			if (error)
4019 				break;
4020 		}
4021 		FILEDESC_SUNLOCK(fdp);
4022 		fddrop(fdp);
4023 		if (error)
4024 			break;
4025 	}
4026 	sx_sunlock(&allproc_lock);
4027 	return (error);
4028 }
4029 
4030 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4031     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4032 
4033 #ifdef KINFO_FILE_SIZE
4034 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4035 #endif
4036 
4037 static int
4038 xlate_fflags(int fflags)
4039 {
4040 	static const struct {
4041 		int	fflag;
4042 		int	kf_fflag;
4043 	} fflags_table[] = {
4044 		{ FAPPEND, KF_FLAG_APPEND },
4045 		{ FASYNC, KF_FLAG_ASYNC },
4046 		{ FFSYNC, KF_FLAG_FSYNC },
4047 		{ FHASLOCK, KF_FLAG_HASLOCK },
4048 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4049 		{ FREAD, KF_FLAG_READ },
4050 		{ FWRITE, KF_FLAG_WRITE },
4051 		{ O_CREAT, KF_FLAG_CREAT },
4052 		{ O_DIRECT, KF_FLAG_DIRECT },
4053 		{ O_EXCL, KF_FLAG_EXCL },
4054 		{ O_EXEC, KF_FLAG_EXEC },
4055 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4056 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4057 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4058 		{ O_TRUNC, KF_FLAG_TRUNC }
4059 	};
4060 	unsigned int i;
4061 	int kflags;
4062 
4063 	kflags = 0;
4064 	for (i = 0; i < nitems(fflags_table); i++)
4065 		if (fflags & fflags_table[i].fflag)
4066 			kflags |=  fflags_table[i].kf_fflag;
4067 	return (kflags);
4068 }
4069 
4070 /* Trim unused data from kf_path by truncating the structure size. */
4071 void
4072 pack_kinfo(struct kinfo_file *kif)
4073 {
4074 
4075 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4076 	    strlen(kif->kf_path) + 1;
4077 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4078 }
4079 
4080 static void
4081 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4082     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4083 {
4084 	int error;
4085 
4086 	bzero(kif, sizeof(*kif));
4087 
4088 	/* Set a default type to allow for empty fill_kinfo() methods. */
4089 	kif->kf_type = KF_TYPE_UNKNOWN;
4090 	kif->kf_flags = xlate_fflags(fp->f_flag);
4091 	if (rightsp != NULL)
4092 		kif->kf_cap_rights = *rightsp;
4093 	else
4094 		cap_rights_init_zero(&kif->kf_cap_rights);
4095 	kif->kf_fd = fd;
4096 	kif->kf_ref_count = refcount_load(&fp->f_count);
4097 	kif->kf_offset = foffset_get(fp);
4098 
4099 	/*
4100 	 * This may drop the filedesc lock, so the 'fp' cannot be
4101 	 * accessed after this call.
4102 	 */
4103 	error = fo_fill_kinfo(fp, kif, fdp);
4104 	if (error == 0)
4105 		kif->kf_status |= KF_ATTR_VALID;
4106 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4107 		pack_kinfo(kif);
4108 	else
4109 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4110 }
4111 
4112 static void
4113 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4114     struct kinfo_file *kif, int flags)
4115 {
4116 	int error;
4117 
4118 	bzero(kif, sizeof(*kif));
4119 
4120 	kif->kf_type = KF_TYPE_VNODE;
4121 	error = vn_fill_kinfo_vnode(vp, kif);
4122 	if (error == 0)
4123 		kif->kf_status |= KF_ATTR_VALID;
4124 	kif->kf_flags = xlate_fflags(fflags);
4125 	cap_rights_init_zero(&kif->kf_cap_rights);
4126 	kif->kf_fd = fd;
4127 	kif->kf_ref_count = -1;
4128 	kif->kf_offset = -1;
4129 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4130 		pack_kinfo(kif);
4131 	else
4132 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4133 	vrele(vp);
4134 }
4135 
4136 struct export_fd_buf {
4137 	struct filedesc		*fdp;
4138 	struct pwddesc	*pdp;
4139 	struct sbuf 		*sb;
4140 	ssize_t			remainder;
4141 	struct kinfo_file	kif;
4142 	int			flags;
4143 };
4144 
4145 static int
4146 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4147 {
4148 	struct kinfo_file *kif;
4149 
4150 	kif = &efbuf->kif;
4151 	if (efbuf->remainder != -1) {
4152 		if (efbuf->remainder < kif->kf_structsize) {
4153 			/* Terminate export. */
4154 			efbuf->remainder = 0;
4155 			return (0);
4156 		}
4157 		efbuf->remainder -= kif->kf_structsize;
4158 	}
4159 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4160 }
4161 
4162 static int
4163 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4164     struct export_fd_buf *efbuf)
4165 {
4166 	int error;
4167 
4168 	if (efbuf->remainder == 0)
4169 		return (0);
4170 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4171 	    efbuf->flags);
4172 	FILEDESC_SUNLOCK(efbuf->fdp);
4173 	error = export_kinfo_to_sb(efbuf);
4174 	FILEDESC_SLOCK(efbuf->fdp);
4175 	return (error);
4176 }
4177 
4178 static int
4179 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4180     struct export_fd_buf *efbuf)
4181 {
4182 	int error;
4183 
4184 	if (efbuf->remainder == 0)
4185 		return (0);
4186 	if (efbuf->pdp != NULL)
4187 		PWDDESC_XUNLOCK(efbuf->pdp);
4188 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4189 	error = export_kinfo_to_sb(efbuf);
4190 	if (efbuf->pdp != NULL)
4191 		PWDDESC_XLOCK(efbuf->pdp);
4192 	return (error);
4193 }
4194 
4195 /*
4196  * Store a process file descriptor information to sbuf.
4197  *
4198  * Takes a locked proc as argument, and returns with the proc unlocked.
4199  */
4200 int
4201 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4202     int flags)
4203 {
4204 	struct file *fp;
4205 	struct filedesc *fdp;
4206 	struct pwddesc *pdp;
4207 	struct export_fd_buf *efbuf;
4208 	struct vnode *cttyvp, *textvp, *tracevp;
4209 	struct pwd *pwd;
4210 	int error, i, lastfile;
4211 	cap_rights_t rights;
4212 
4213 	PROC_LOCK_ASSERT(p, MA_OWNED);
4214 
4215 	/* ktrace vnode */
4216 	tracevp = p->p_tracevp;
4217 	if (tracevp != NULL)
4218 		vrefact(tracevp);
4219 	/* text vnode */
4220 	textvp = p->p_textvp;
4221 	if (textvp != NULL)
4222 		vrefact(textvp);
4223 	/* Controlling tty. */
4224 	cttyvp = NULL;
4225 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4226 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4227 		if (cttyvp != NULL)
4228 			vrefact(cttyvp);
4229 	}
4230 	fdp = fdhold(p);
4231 	pdp = pdhold(p);
4232 	PROC_UNLOCK(p);
4233 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4234 	efbuf->fdp = NULL;
4235 	efbuf->pdp = NULL;
4236 	efbuf->sb = sb;
4237 	efbuf->remainder = maxlen;
4238 	efbuf->flags = flags;
4239 	if (tracevp != NULL)
4240 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4241 		    efbuf);
4242 	if (textvp != NULL)
4243 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4244 	if (cttyvp != NULL)
4245 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4246 		    efbuf);
4247 	error = 0;
4248 	if (pdp == NULL || fdp == NULL)
4249 		goto fail;
4250 	efbuf->fdp = fdp;
4251 	efbuf->pdp = pdp;
4252 	PWDDESC_XLOCK(pdp);
4253 	pwd = pwd_hold_pwddesc(pdp);
4254 	if (pwd != NULL) {
4255 		/* working directory */
4256 		if (pwd->pwd_cdir != NULL) {
4257 			vrefact(pwd->pwd_cdir);
4258 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4259 		}
4260 		/* root directory */
4261 		if (pwd->pwd_rdir != NULL) {
4262 			vrefact(pwd->pwd_rdir);
4263 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4264 		}
4265 		/* jail directory */
4266 		if (pwd->pwd_jdir != NULL) {
4267 			vrefact(pwd->pwd_jdir);
4268 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4269 		}
4270 	}
4271 	PWDDESC_XUNLOCK(pdp);
4272 	if (pwd != NULL)
4273 		pwd_drop(pwd);
4274 	FILEDESC_SLOCK(fdp);
4275 	lastfile = fdlastfile(fdp);
4276 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4277 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4278 			continue;
4279 #ifdef CAPABILITIES
4280 		rights = *cap_rights(fdp, i);
4281 #else /* !CAPABILITIES */
4282 		rights = cap_no_rights;
4283 #endif
4284 		/*
4285 		 * Create sysctl entry.  It is OK to drop the filedesc
4286 		 * lock inside of export_file_to_sb() as we will
4287 		 * re-validate and re-evaluate its properties when the
4288 		 * loop continues.
4289 		 */
4290 		error = export_file_to_sb(fp, i, &rights, efbuf);
4291 		if (error != 0 || efbuf->remainder == 0)
4292 			break;
4293 	}
4294 	FILEDESC_SUNLOCK(fdp);
4295 fail:
4296 	if (fdp != NULL)
4297 		fddrop(fdp);
4298 	if (pdp != NULL)
4299 		pddrop(pdp);
4300 	free(efbuf, M_TEMP);
4301 	return (error);
4302 }
4303 
4304 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4305 
4306 /*
4307  * Get per-process file descriptors for use by procstat(1), et al.
4308  */
4309 static int
4310 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4311 {
4312 	struct sbuf sb;
4313 	struct proc *p;
4314 	ssize_t maxlen;
4315 	int error, error2, *name;
4316 
4317 	name = (int *)arg1;
4318 
4319 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4320 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4321 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4322 	if (error != 0) {
4323 		sbuf_delete(&sb);
4324 		return (error);
4325 	}
4326 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4327 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4328 	    KERN_FILEDESC_PACK_KINFO);
4329 	error2 = sbuf_finish(&sb);
4330 	sbuf_delete(&sb);
4331 	return (error != 0 ? error : error2);
4332 }
4333 
4334 #ifdef COMPAT_FREEBSD7
4335 #ifdef KINFO_OFILE_SIZE
4336 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4337 #endif
4338 
4339 static void
4340 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4341 {
4342 
4343 	okif->kf_structsize = sizeof(*okif);
4344 	okif->kf_type = kif->kf_type;
4345 	okif->kf_fd = kif->kf_fd;
4346 	okif->kf_ref_count = kif->kf_ref_count;
4347 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4348 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4349 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4350 	okif->kf_offset = kif->kf_offset;
4351 	if (kif->kf_type == KF_TYPE_VNODE)
4352 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4353 	else
4354 		okif->kf_vnode_type = KF_VTYPE_VNON;
4355 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4356 	if (kif->kf_type == KF_TYPE_SOCKET) {
4357 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4358 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4359 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4360 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4361 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4362 	} else {
4363 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4364 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4365 	}
4366 }
4367 
4368 static int
4369 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4370     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4371 {
4372 	int error;
4373 
4374 	vrefact(vp);
4375 	PWDDESC_XUNLOCK(pdp);
4376 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4377 	kinfo_to_okinfo(kif, okif);
4378 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4379 	PWDDESC_XLOCK(pdp);
4380 	return (error);
4381 }
4382 
4383 /*
4384  * Get per-process file descriptors for use by procstat(1), et al.
4385  */
4386 static int
4387 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4388 {
4389 	struct kinfo_ofile *okif;
4390 	struct kinfo_file *kif;
4391 	struct filedesc *fdp;
4392 	struct pwddesc *pdp;
4393 	struct pwd *pwd;
4394 	int error, i, lastfile, *name;
4395 	struct file *fp;
4396 	struct proc *p;
4397 
4398 	name = (int *)arg1;
4399 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4400 	if (error != 0)
4401 		return (error);
4402 	fdp = fdhold(p);
4403 	if (fdp != NULL)
4404 		pdp = pdhold(p);
4405 	PROC_UNLOCK(p);
4406 	if (fdp == NULL || pdp == NULL) {
4407 		if (fdp != NULL)
4408 			fddrop(fdp);
4409 		return (ENOENT);
4410 	}
4411 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4412 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4413 	PWDDESC_XLOCK(pdp);
4414 	pwd = pwd_hold_pwddesc(pdp);
4415 	if (pwd != NULL) {
4416 		if (pwd->pwd_cdir != NULL)
4417 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4418 			    okif, pdp, req);
4419 		if (pwd->pwd_rdir != NULL)
4420 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4421 			    okif, pdp, req);
4422 		if (pwd->pwd_jdir != NULL)
4423 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4424 			    okif, pdp, req);
4425 	}
4426 	PWDDESC_XUNLOCK(pdp);
4427 	if (pwd != NULL)
4428 		pwd_drop(pwd);
4429 	FILEDESC_SLOCK(fdp);
4430 	lastfile = fdlastfile(fdp);
4431 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4432 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4433 			continue;
4434 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4435 		    KERN_FILEDESC_PACK_KINFO);
4436 		FILEDESC_SUNLOCK(fdp);
4437 		kinfo_to_okinfo(kif, okif);
4438 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4439 		FILEDESC_SLOCK(fdp);
4440 		if (error)
4441 			break;
4442 	}
4443 	FILEDESC_SUNLOCK(fdp);
4444 	fddrop(fdp);
4445 	pddrop(pdp);
4446 	free(kif, M_TEMP);
4447 	free(okif, M_TEMP);
4448 	return (0);
4449 }
4450 
4451 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4452     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4453     "Process ofiledesc entries");
4454 #endif	/* COMPAT_FREEBSD7 */
4455 
4456 int
4457 vntype_to_kinfo(int vtype)
4458 {
4459 	struct {
4460 		int	vtype;
4461 		int	kf_vtype;
4462 	} vtypes_table[] = {
4463 		{ VBAD, KF_VTYPE_VBAD },
4464 		{ VBLK, KF_VTYPE_VBLK },
4465 		{ VCHR, KF_VTYPE_VCHR },
4466 		{ VDIR, KF_VTYPE_VDIR },
4467 		{ VFIFO, KF_VTYPE_VFIFO },
4468 		{ VLNK, KF_VTYPE_VLNK },
4469 		{ VNON, KF_VTYPE_VNON },
4470 		{ VREG, KF_VTYPE_VREG },
4471 		{ VSOCK, KF_VTYPE_VSOCK }
4472 	};
4473 	unsigned int i;
4474 
4475 	/*
4476 	 * Perform vtype translation.
4477 	 */
4478 	for (i = 0; i < nitems(vtypes_table); i++)
4479 		if (vtypes_table[i].vtype == vtype)
4480 			return (vtypes_table[i].kf_vtype);
4481 
4482 	return (KF_VTYPE_UNKNOWN);
4483 }
4484 
4485 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4486     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4487     "Process filedesc entries");
4488 
4489 /*
4490  * Store a process current working directory information to sbuf.
4491  *
4492  * Takes a locked proc as argument, and returns with the proc unlocked.
4493  */
4494 int
4495 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4496 {
4497 	struct pwddesc *pdp;
4498 	struct pwd *pwd;
4499 	struct export_fd_buf *efbuf;
4500 	struct vnode *cdir;
4501 	int error;
4502 
4503 	PROC_LOCK_ASSERT(p, MA_OWNED);
4504 
4505 	pdp = pdhold(p);
4506 	PROC_UNLOCK(p);
4507 	if (pdp == NULL)
4508 		return (EINVAL);
4509 
4510 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4511 	efbuf->pdp = pdp;
4512 	efbuf->sb = sb;
4513 	efbuf->remainder = maxlen;
4514 
4515 	PWDDESC_XLOCK(pdp);
4516 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4517 	cdir = pwd->pwd_cdir;
4518 	if (cdir == NULL) {
4519 		error = EINVAL;
4520 	} else {
4521 		vrefact(cdir);
4522 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4523 	}
4524 	PWDDESC_XUNLOCK(pdp);
4525 	pddrop(pdp);
4526 	free(efbuf, M_TEMP);
4527 	return (error);
4528 }
4529 
4530 /*
4531  * Get per-process current working directory.
4532  */
4533 static int
4534 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4535 {
4536 	struct sbuf sb;
4537 	struct proc *p;
4538 	ssize_t maxlen;
4539 	int error, error2, *name;
4540 
4541 	name = (int *)arg1;
4542 
4543 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4544 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4545 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4546 	if (error != 0) {
4547 		sbuf_delete(&sb);
4548 		return (error);
4549 	}
4550 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4551 	error = kern_proc_cwd_out(p, &sb, maxlen);
4552 	error2 = sbuf_finish(&sb);
4553 	sbuf_delete(&sb);
4554 	return (error != 0 ? error : error2);
4555 }
4556 
4557 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4558     sysctl_kern_proc_cwd, "Process current working directory");
4559 
4560 #ifdef DDB
4561 /*
4562  * For the purposes of debugging, generate a human-readable string for the
4563  * file type.
4564  */
4565 static const char *
4566 file_type_to_name(short type)
4567 {
4568 
4569 	switch (type) {
4570 	case 0:
4571 		return ("zero");
4572 	case DTYPE_VNODE:
4573 		return ("vnode");
4574 	case DTYPE_SOCKET:
4575 		return ("socket");
4576 	case DTYPE_PIPE:
4577 		return ("pipe");
4578 	case DTYPE_FIFO:
4579 		return ("fifo");
4580 	case DTYPE_KQUEUE:
4581 		return ("kqueue");
4582 	case DTYPE_CRYPTO:
4583 		return ("crypto");
4584 	case DTYPE_MQUEUE:
4585 		return ("mqueue");
4586 	case DTYPE_SHM:
4587 		return ("shm");
4588 	case DTYPE_SEM:
4589 		return ("ksem");
4590 	case DTYPE_PTS:
4591 		return ("pts");
4592 	case DTYPE_DEV:
4593 		return ("dev");
4594 	case DTYPE_PROCDESC:
4595 		return ("proc");
4596 	case DTYPE_EVENTFD:
4597 		return ("eventfd");
4598 	case DTYPE_LINUXTFD:
4599 		return ("ltimer");
4600 	default:
4601 		return ("unkn");
4602 	}
4603 }
4604 
4605 /*
4606  * For the purposes of debugging, identify a process (if any, perhaps one of
4607  * many) that references the passed file in its file descriptor array. Return
4608  * NULL if none.
4609  */
4610 static struct proc *
4611 file_to_first_proc(struct file *fp)
4612 {
4613 	struct filedesc *fdp;
4614 	struct proc *p;
4615 	int n;
4616 
4617 	FOREACH_PROC_IN_SYSTEM(p) {
4618 		if (p->p_state == PRS_NEW)
4619 			continue;
4620 		fdp = p->p_fd;
4621 		if (fdp == NULL)
4622 			continue;
4623 		for (n = 0; n < fdp->fd_nfiles; n++) {
4624 			if (fp == fdp->fd_ofiles[n].fde_file)
4625 				return (p);
4626 		}
4627 	}
4628 	return (NULL);
4629 }
4630 
4631 static void
4632 db_print_file(struct file *fp, int header)
4633 {
4634 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4635 	struct proc *p;
4636 
4637 	if (header)
4638 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4639 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4640 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4641 		    "FCmd");
4642 	p = file_to_first_proc(fp);
4643 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4644 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4645 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4646 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4647 
4648 #undef XPTRWIDTH
4649 }
4650 
4651 DB_SHOW_COMMAND(file, db_show_file)
4652 {
4653 	struct file *fp;
4654 
4655 	if (!have_addr) {
4656 		db_printf("usage: show file <addr>\n");
4657 		return;
4658 	}
4659 	fp = (struct file *)addr;
4660 	db_print_file(fp, 1);
4661 }
4662 
4663 DB_SHOW_COMMAND(files, db_show_files)
4664 {
4665 	struct filedesc *fdp;
4666 	struct file *fp;
4667 	struct proc *p;
4668 	int header;
4669 	int n;
4670 
4671 	header = 1;
4672 	FOREACH_PROC_IN_SYSTEM(p) {
4673 		if (p->p_state == PRS_NEW)
4674 			continue;
4675 		if ((fdp = p->p_fd) == NULL)
4676 			continue;
4677 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4678 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4679 				continue;
4680 			db_print_file(fp, header);
4681 			header = 0;
4682 		}
4683 	}
4684 }
4685 #endif
4686 
4687 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4688     &maxfilesperproc, 0, "Maximum files allowed open per process");
4689 
4690 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4691     &maxfiles, 0, "Maximum number of files");
4692 
4693 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4694     &openfiles, 0, "System-wide number of open files");
4695 
4696 /* ARGSUSED*/
4697 static void
4698 filelistinit(void *dummy)
4699 {
4700 
4701 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4702 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4703 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4704 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4705 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4706 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4707 	/*
4708 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4709 	 * the vnode zone.
4710 	 */
4711 	vfs_smr = uma_zone_get_smr(pwd_zone);
4712 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4713 }
4714 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4715 
4716 /*-------------------------------------------------------------------*/
4717 
4718 static int
4719 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4720     int flags, struct thread *td)
4721 {
4722 
4723 	return (EBADF);
4724 }
4725 
4726 static int
4727 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4728     struct thread *td)
4729 {
4730 
4731 	return (EINVAL);
4732 }
4733 
4734 static int
4735 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4736     struct thread *td)
4737 {
4738 
4739 	return (EBADF);
4740 }
4741 
4742 static int
4743 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4744     struct thread *td)
4745 {
4746 
4747 	return (0);
4748 }
4749 
4750 static int
4751 badfo_kqfilter(struct file *fp, struct knote *kn)
4752 {
4753 
4754 	return (EBADF);
4755 }
4756 
4757 static int
4758 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4759     struct thread *td)
4760 {
4761 
4762 	return (EBADF);
4763 }
4764 
4765 static int
4766 badfo_close(struct file *fp, struct thread *td)
4767 {
4768 
4769 	return (0);
4770 }
4771 
4772 static int
4773 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4774     struct thread *td)
4775 {
4776 
4777 	return (EBADF);
4778 }
4779 
4780 static int
4781 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4782     struct thread *td)
4783 {
4784 
4785 	return (EBADF);
4786 }
4787 
4788 static int
4789 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4790     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4791     struct thread *td)
4792 {
4793 
4794 	return (EBADF);
4795 }
4796 
4797 static int
4798 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4799 {
4800 
4801 	return (0);
4802 }
4803 
4804 struct fileops badfileops = {
4805 	.fo_read = badfo_readwrite,
4806 	.fo_write = badfo_readwrite,
4807 	.fo_truncate = badfo_truncate,
4808 	.fo_ioctl = badfo_ioctl,
4809 	.fo_poll = badfo_poll,
4810 	.fo_kqfilter = badfo_kqfilter,
4811 	.fo_stat = badfo_stat,
4812 	.fo_close = badfo_close,
4813 	.fo_chmod = badfo_chmod,
4814 	.fo_chown = badfo_chown,
4815 	.fo_sendfile = badfo_sendfile,
4816 	.fo_fill_kinfo = badfo_fill_kinfo,
4817 };
4818 
4819 int
4820 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4821     int flags, struct thread *td)
4822 {
4823 
4824 	return (EOPNOTSUPP);
4825 }
4826 
4827 int
4828 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4829     struct thread *td)
4830 {
4831 
4832 	return (EINVAL);
4833 }
4834 
4835 int
4836 invfo_ioctl(struct file *fp, u_long com, void *data,
4837     struct ucred *active_cred, struct thread *td)
4838 {
4839 
4840 	return (ENOTTY);
4841 }
4842 
4843 int
4844 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4845     struct thread *td)
4846 {
4847 
4848 	return (poll_no_poll(events));
4849 }
4850 
4851 int
4852 invfo_kqfilter(struct file *fp, struct knote *kn)
4853 {
4854 
4855 	return (EINVAL);
4856 }
4857 
4858 int
4859 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4860     struct thread *td)
4861 {
4862 
4863 	return (EINVAL);
4864 }
4865 
4866 int
4867 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4868     struct thread *td)
4869 {
4870 
4871 	return (EINVAL);
4872 }
4873 
4874 int
4875 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4876     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4877     struct thread *td)
4878 {
4879 
4880 	return (EINVAL);
4881 }
4882 
4883 /*-------------------------------------------------------------------*/
4884 
4885 /*
4886  * File Descriptor pseudo-device driver (/dev/fd/).
4887  *
4888  * Opening minor device N dup()s the file (if any) connected to file
4889  * descriptor N belonging to the calling process.  Note that this driver
4890  * consists of only the ``open()'' routine, because all subsequent
4891  * references to this file will be direct to the other driver.
4892  *
4893  * XXX: we could give this one a cloning event handler if necessary.
4894  */
4895 
4896 /* ARGSUSED */
4897 static int
4898 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4899 {
4900 
4901 	/*
4902 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4903 	 * the file descriptor being sought for duplication. The error
4904 	 * return ensures that the vnode for this device will be released
4905 	 * by vn_open. Open will detect this special error and take the
4906 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4907 	 * will simply report the error.
4908 	 */
4909 	td->td_dupfd = dev2unit(dev);
4910 	return (ENODEV);
4911 }
4912 
4913 static struct cdevsw fildesc_cdevsw = {
4914 	.d_version =	D_VERSION,
4915 	.d_open =	fdopen,
4916 	.d_name =	"FD",
4917 };
4918 
4919 static void
4920 fildesc_drvinit(void *unused)
4921 {
4922 	struct cdev *dev;
4923 
4924 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4925 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4926 	make_dev_alias(dev, "stdin");
4927 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4928 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4929 	make_dev_alias(dev, "stdout");
4930 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4931 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4932 	make_dev_alias(dev, "stderr");
4933 }
4934 
4935 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4936