xref: /freebsd/sys/kern/kern_descrip.c (revision 833a452e9f082a7982a31c21f0da437dbbe0a39d)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/poll.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sbuf.h>
71 #include <sys/signalvar.h>
72 #include <sys/kdb.h>
73 #include <sys/smr.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
80 #include <sys/user.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97     "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
100 
101 MALLOC_DECLARE(M_FADVISE);
102 
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
106 VFS_SMR_DECLARE;
107 
108 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
109 		    struct thread *td, bool holdleaders, bool audit);
110 static void	export_file_to_kinfo(struct file *fp, int fd,
111 		    cap_rights_t *rightsp, struct kinfo_file *kif,
112 		    struct filedesc *fdp, int flags);
113 static int	fd_first_free(struct filedesc *fdp, int low, int size);
114 static void	fdgrowtable(struct filedesc *fdp, int nfd);
115 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
116 static void	fdunused(struct filedesc *fdp, int fd);
117 static void	fdused(struct filedesc *fdp, int fd);
118 static int	fget_unlocked_seq(struct filedesc *fdp, int fd,
119 		    cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp);
120 static int	getmaxfd(struct thread *td);
121 static u_long	*filecaps_copy_prep(const struct filecaps *src);
122 static void	filecaps_copy_finish(const struct filecaps *src,
123 		    struct filecaps *dst, u_long *ioctls);
124 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
125 static void	filecaps_free_finish(u_long *ioctls);
126 
127 static struct pwd *pwd_alloc(void);
128 
129 /*
130  * Each process has:
131  *
132  * - An array of open file descriptors (fd_ofiles)
133  * - An array of file flags (fd_ofileflags)
134  * - A bitmap recording which descriptors are in use (fd_map)
135  *
136  * A process starts out with NDFILE descriptors.  The value of NDFILE has
137  * been selected based the historical limit of 20 open files, and an
138  * assumption that the majority of processes, especially short-lived
139  * processes like shells, will never need more.
140  *
141  * If this initial allocation is exhausted, a larger descriptor table and
142  * map are allocated dynamically, and the pointers in the process's struct
143  * filedesc are updated to point to those.  This is repeated every time
144  * the process runs out of file descriptors (provided it hasn't hit its
145  * resource limit).
146  *
147  * Since threads may hold references to individual descriptor table
148  * entries, the tables are never freed.  Instead, they are placed on a
149  * linked list and freed only when the struct filedesc is released.
150  */
151 #define NDFILE		20
152 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
153 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
154 #define NDSLOT(x)	((x) / NDENTRIES)
155 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
156 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
157 
158 /*
159  * SLIST entry used to keep track of ofiles which must be reclaimed when
160  * the process exits.
161  */
162 struct freetable {
163 	struct fdescenttbl *ft_table;
164 	SLIST_ENTRY(freetable) ft_next;
165 };
166 
167 /*
168  * Initial allocation: a filedesc structure + the head of SLIST used to
169  * keep track of old ofiles + enough space for NDFILE descriptors.
170  */
171 
172 struct fdescenttbl0 {
173 	int	fdt_nfiles;
174 	struct	filedescent fdt_ofiles[NDFILE];
175 };
176 
177 struct filedesc0 {
178 	struct filedesc fd_fd;
179 	SLIST_HEAD(, freetable) fd_free;
180 	struct	fdescenttbl0 fd_dfiles;
181 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
182 };
183 
184 /*
185  * Descriptor management.
186  */
187 static int __exclusive_cache_line openfiles; /* actual number of open files */
188 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
189 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
190 
191 /*
192  * If low >= size, just return low. Otherwise find the first zero bit in the
193  * given bitmap, starting at low and not exceeding size - 1. Return size if
194  * not found.
195  */
196 static int
197 fd_first_free(struct filedesc *fdp, int low, int size)
198 {
199 	NDSLOTTYPE *map = fdp->fd_map;
200 	NDSLOTTYPE mask;
201 	int off, maxoff;
202 
203 	if (low >= size)
204 		return (low);
205 
206 	off = NDSLOT(low);
207 	if (low % NDENTRIES) {
208 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
209 		if ((mask &= ~map[off]) != 0UL)
210 			return (off * NDENTRIES + ffsl(mask) - 1);
211 		++off;
212 	}
213 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
214 		if (map[off] != ~0UL)
215 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
216 	return (size);
217 }
218 
219 /*
220  * Find the last used fd.
221  *
222  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
223  * Otherwise use fdlastfile.
224  */
225 int
226 fdlastfile_single(struct filedesc *fdp)
227 {
228 	NDSLOTTYPE *map = fdp->fd_map;
229 	int off, minoff;
230 
231 	off = NDSLOT(fdp->fd_nfiles - 1);
232 	for (minoff = NDSLOT(0); off >= minoff; --off)
233 		if (map[off] != 0)
234 			return (off * NDENTRIES + flsl(map[off]) - 1);
235 	return (-1);
236 }
237 
238 int
239 fdlastfile(struct filedesc *fdp)
240 {
241 
242 	FILEDESC_LOCK_ASSERT(fdp);
243 	return (fdlastfile_single(fdp));
244 }
245 
246 static int
247 fdisused(struct filedesc *fdp, int fd)
248 {
249 
250 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
251 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
252 
253 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
254 }
255 
256 /*
257  * Mark a file descriptor as used.
258  */
259 static void
260 fdused_init(struct filedesc *fdp, int fd)
261 {
262 
263 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
264 
265 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
266 }
267 
268 static void
269 fdused(struct filedesc *fdp, int fd)
270 {
271 
272 	FILEDESC_XLOCK_ASSERT(fdp);
273 
274 	fdused_init(fdp, fd);
275 	if (fd == fdp->fd_freefile)
276 		fdp->fd_freefile++;
277 }
278 
279 /*
280  * Mark a file descriptor as unused.
281  */
282 static void
283 fdunused(struct filedesc *fdp, int fd)
284 {
285 
286 	FILEDESC_XLOCK_ASSERT(fdp);
287 
288 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
289 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
290 	    ("fd=%d is still in use", fd));
291 
292 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
293 	if (fd < fdp->fd_freefile)
294 		fdp->fd_freefile = fd;
295 }
296 
297 /*
298  * Free a file descriptor.
299  *
300  * Avoid some work if fdp is about to be destroyed.
301  */
302 static inline void
303 fdefree_last(struct filedescent *fde)
304 {
305 
306 	filecaps_free(&fde->fde_caps);
307 }
308 
309 static inline void
310 fdfree(struct filedesc *fdp, int fd)
311 {
312 	struct filedescent *fde;
313 
314 	FILEDESC_XLOCK_ASSERT(fdp);
315 	fde = &fdp->fd_ofiles[fd];
316 #ifdef CAPABILITIES
317 	seqc_write_begin(&fde->fde_seqc);
318 #endif
319 	fde->fde_file = NULL;
320 #ifdef CAPABILITIES
321 	seqc_write_end(&fde->fde_seqc);
322 #endif
323 	fdefree_last(fde);
324 	fdunused(fdp, fd);
325 }
326 
327 /*
328  * System calls on descriptors.
329  */
330 #ifndef _SYS_SYSPROTO_H_
331 struct getdtablesize_args {
332 	int	dummy;
333 };
334 #endif
335 /* ARGSUSED */
336 int
337 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
338 {
339 #ifdef	RACCT
340 	uint64_t lim;
341 #endif
342 
343 	td->td_retval[0] = getmaxfd(td);
344 #ifdef	RACCT
345 	PROC_LOCK(td->td_proc);
346 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
347 	PROC_UNLOCK(td->td_proc);
348 	if (lim < td->td_retval[0])
349 		td->td_retval[0] = lim;
350 #endif
351 	return (0);
352 }
353 
354 /*
355  * Duplicate a file descriptor to a particular value.
356  *
357  * Note: keep in mind that a potential race condition exists when closing
358  * descriptors from a shared descriptor table (via rfork).
359  */
360 #ifndef _SYS_SYSPROTO_H_
361 struct dup2_args {
362 	u_int	from;
363 	u_int	to;
364 };
365 #endif
366 /* ARGSUSED */
367 int
368 sys_dup2(struct thread *td, struct dup2_args *uap)
369 {
370 
371 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
372 }
373 
374 /*
375  * Duplicate a file descriptor.
376  */
377 #ifndef _SYS_SYSPROTO_H_
378 struct dup_args {
379 	u_int	fd;
380 };
381 #endif
382 /* ARGSUSED */
383 int
384 sys_dup(struct thread *td, struct dup_args *uap)
385 {
386 
387 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
388 }
389 
390 /*
391  * The file control system call.
392  */
393 #ifndef _SYS_SYSPROTO_H_
394 struct fcntl_args {
395 	int	fd;
396 	int	cmd;
397 	long	arg;
398 };
399 #endif
400 /* ARGSUSED */
401 int
402 sys_fcntl(struct thread *td, struct fcntl_args *uap)
403 {
404 
405 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
406 }
407 
408 int
409 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
410 {
411 	struct flock fl;
412 	struct __oflock ofl;
413 	intptr_t arg1;
414 	int error, newcmd;
415 
416 	error = 0;
417 	newcmd = cmd;
418 	switch (cmd) {
419 	case F_OGETLK:
420 	case F_OSETLK:
421 	case F_OSETLKW:
422 		/*
423 		 * Convert old flock structure to new.
424 		 */
425 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
426 		fl.l_start = ofl.l_start;
427 		fl.l_len = ofl.l_len;
428 		fl.l_pid = ofl.l_pid;
429 		fl.l_type = ofl.l_type;
430 		fl.l_whence = ofl.l_whence;
431 		fl.l_sysid = 0;
432 
433 		switch (cmd) {
434 		case F_OGETLK:
435 			newcmd = F_GETLK;
436 			break;
437 		case F_OSETLK:
438 			newcmd = F_SETLK;
439 			break;
440 		case F_OSETLKW:
441 			newcmd = F_SETLKW;
442 			break;
443 		}
444 		arg1 = (intptr_t)&fl;
445 		break;
446 	case F_GETLK:
447 	case F_SETLK:
448 	case F_SETLKW:
449 	case F_SETLK_REMOTE:
450 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
451 		arg1 = (intptr_t)&fl;
452 		break;
453 	default:
454 		arg1 = arg;
455 		break;
456 	}
457 	if (error)
458 		return (error);
459 	error = kern_fcntl(td, fd, newcmd, arg1);
460 	if (error)
461 		return (error);
462 	if (cmd == F_OGETLK) {
463 		ofl.l_start = fl.l_start;
464 		ofl.l_len = fl.l_len;
465 		ofl.l_pid = fl.l_pid;
466 		ofl.l_type = fl.l_type;
467 		ofl.l_whence = fl.l_whence;
468 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
469 	} else if (cmd == F_GETLK) {
470 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
471 	}
472 	return (error);
473 }
474 
475 int
476 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
477 {
478 	struct filedesc *fdp;
479 	struct flock *flp;
480 	struct file *fp, *fp2;
481 	struct filedescent *fde;
482 	struct proc *p;
483 	struct vnode *vp;
484 	struct mount *mp;
485 	struct kinfo_file *kif;
486 	int error, flg, kif_sz, seals, tmp;
487 	uint64_t bsize;
488 	off_t foffset;
489 
490 	error = 0;
491 	flg = F_POSIX;
492 	p = td->td_proc;
493 	fdp = p->p_fd;
494 
495 	AUDIT_ARG_FD(cmd);
496 	AUDIT_ARG_CMD(cmd);
497 	switch (cmd) {
498 	case F_DUPFD:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
501 		break;
502 
503 	case F_DUPFD_CLOEXEC:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
511 		break;
512 
513 	case F_DUP2FD_CLOEXEC:
514 		tmp = arg;
515 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
516 		break;
517 
518 	case F_GETFD:
519 		error = EBADF;
520 		FILEDESC_SLOCK(fdp);
521 		fde = fdeget_locked(fdp, fd);
522 		if (fde != NULL) {
523 			td->td_retval[0] =
524 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
525 			error = 0;
526 		}
527 		FILEDESC_SUNLOCK(fdp);
528 		break;
529 
530 	case F_SETFD:
531 		error = EBADF;
532 		FILEDESC_XLOCK(fdp);
533 		fde = fdeget_locked(fdp, fd);
534 		if (fde != NULL) {
535 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
536 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
537 			error = 0;
538 		}
539 		FILEDESC_XUNLOCK(fdp);
540 		break;
541 
542 	case F_GETFL:
543 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
544 		if (error != 0)
545 			break;
546 		td->td_retval[0] = OFLAGS(fp->f_flag);
547 		fdrop(fp, td);
548 		break;
549 
550 	case F_SETFL:
551 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
552 		if (error != 0)
553 			break;
554 		if (fp->f_ops == &path_fileops) {
555 			fdrop(fp, td);
556 			error = EBADF;
557 			break;
558 		}
559 		do {
560 			tmp = flg = fp->f_flag;
561 			tmp &= ~FCNTLFLAGS;
562 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
563 		} while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
564 		tmp = fp->f_flag & FNONBLOCK;
565 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
566 		if (error != 0) {
567 			fdrop(fp, td);
568 			break;
569 		}
570 		tmp = fp->f_flag & FASYNC;
571 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
572 		if (error == 0) {
573 			fdrop(fp, td);
574 			break;
575 		}
576 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
577 		tmp = 0;
578 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_GETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
587 		if (error == 0)
588 			td->td_retval[0] = tmp;
589 		fdrop(fp, td);
590 		break;
591 
592 	case F_SETOWN:
593 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
594 		if (error != 0)
595 			break;
596 		tmp = arg;
597 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
598 		fdrop(fp, td);
599 		break;
600 
601 	case F_SETLK_REMOTE:
602 		error = priv_check(td, PRIV_NFS_LOCKD);
603 		if (error != 0)
604 			return (error);
605 		flg = F_REMOTE;
606 		goto do_setlk;
607 
608 	case F_SETLKW:
609 		flg |= F_WAIT;
610 		/* FALLTHROUGH F_SETLK */
611 
612 	case F_SETLK:
613 	do_setlk:
614 		flp = (struct flock *)arg;
615 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
616 			error = EINVAL;
617 			break;
618 		}
619 
620 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
621 		if (error != 0)
622 			break;
623 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
624 			error = EBADF;
625 			fdrop(fp, td);
626 			break;
627 		}
628 
629 		if (flp->l_whence == SEEK_CUR) {
630 			foffset = foffset_get(fp);
631 			if (foffset < 0 ||
632 			    (flp->l_start > 0 &&
633 			     foffset > OFF_MAX - flp->l_start)) {
634 				error = EOVERFLOW;
635 				fdrop(fp, td);
636 				break;
637 			}
638 			flp->l_start += foffset;
639 		}
640 
641 		vp = fp->f_vnode;
642 		switch (flp->l_type) {
643 		case F_RDLCK:
644 			if ((fp->f_flag & FREAD) == 0) {
645 				error = EBADF;
646 				break;
647 			}
648 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
649 				PROC_LOCK(p->p_leader);
650 				p->p_leader->p_flag |= P_ADVLOCK;
651 				PROC_UNLOCK(p->p_leader);
652 			}
653 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
654 			    flp, flg);
655 			break;
656 		case F_WRLCK:
657 			if ((fp->f_flag & FWRITE) == 0) {
658 				error = EBADF;
659 				break;
660 			}
661 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
662 				PROC_LOCK(p->p_leader);
663 				p->p_leader->p_flag |= P_ADVLOCK;
664 				PROC_UNLOCK(p->p_leader);
665 			}
666 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
667 			    flp, flg);
668 			break;
669 		case F_UNLCK:
670 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
671 			    flp, flg);
672 			break;
673 		case F_UNLCKSYS:
674 			if (flg != F_REMOTE) {
675 				error = EINVAL;
676 				break;
677 			}
678 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
679 			    F_UNLCKSYS, flp, flg);
680 			break;
681 		default:
682 			error = EINVAL;
683 			break;
684 		}
685 		if (error != 0 || flp->l_type == F_UNLCK ||
686 		    flp->l_type == F_UNLCKSYS) {
687 			fdrop(fp, td);
688 			break;
689 		}
690 
691 		/*
692 		 * Check for a race with close.
693 		 *
694 		 * The vnode is now advisory locked (or unlocked, but this case
695 		 * is not really important) as the caller requested.
696 		 * We had to drop the filedesc lock, so we need to recheck if
697 		 * the descriptor is still valid, because if it was closed
698 		 * in the meantime we need to remove advisory lock from the
699 		 * vnode - close on any descriptor leading to an advisory
700 		 * locked vnode, removes that lock.
701 		 * We will return 0 on purpose in that case, as the result of
702 		 * successful advisory lock might have been externally visible
703 		 * already. This is fine - effectively we pretend to the caller
704 		 * that the closing thread was a bit slower and that the
705 		 * advisory lock succeeded before the close.
706 		 */
707 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
708 		if (error != 0) {
709 			fdrop(fp, td);
710 			break;
711 		}
712 		if (fp != fp2) {
713 			flp->l_whence = SEEK_SET;
714 			flp->l_start = 0;
715 			flp->l_len = 0;
716 			flp->l_type = F_UNLCK;
717 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
718 			    F_UNLCK, flp, F_POSIX);
719 		}
720 		fdrop(fp, td);
721 		fdrop(fp2, td);
722 		break;
723 
724 	case F_GETLK:
725 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
726 		if (error != 0)
727 			break;
728 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
729 			error = EBADF;
730 			fdrop(fp, td);
731 			break;
732 		}
733 		flp = (struct flock *)arg;
734 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
735 		    flp->l_type != F_UNLCK) {
736 			error = EINVAL;
737 			fdrop(fp, td);
738 			break;
739 		}
740 		if (flp->l_whence == SEEK_CUR) {
741 			foffset = foffset_get(fp);
742 			if ((flp->l_start > 0 &&
743 			    foffset > OFF_MAX - flp->l_start) ||
744 			    (flp->l_start < 0 &&
745 			    foffset < OFF_MIN - flp->l_start)) {
746 				error = EOVERFLOW;
747 				fdrop(fp, td);
748 				break;
749 			}
750 			flp->l_start += foffset;
751 		}
752 		vp = fp->f_vnode;
753 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
754 		    F_POSIX);
755 		fdrop(fp, td);
756 		break;
757 
758 	case F_ADD_SEALS:
759 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
760 		if (error != 0)
761 			break;
762 		error = fo_add_seals(fp, arg);
763 		fdrop(fp, td);
764 		break;
765 
766 	case F_GET_SEALS:
767 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
768 		if (error != 0)
769 			break;
770 		if (fo_get_seals(fp, &seals) == 0)
771 			td->td_retval[0] = seals;
772 		else
773 			error = EINVAL;
774 		fdrop(fp, td);
775 		break;
776 
777 	case F_RDAHEAD:
778 		arg = arg ? 128 * 1024: 0;
779 		/* FALLTHROUGH */
780 	case F_READAHEAD:
781 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
782 		if (error != 0)
783 			break;
784 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
785 			fdrop(fp, td);
786 			error = EBADF;
787 			break;
788 		}
789 		vp = fp->f_vnode;
790 		if (vp->v_type != VREG) {
791 			fdrop(fp, td);
792 			error = ENOTTY;
793 			break;
794 		}
795 
796 		/*
797 		 * Exclusive lock synchronizes against f_seqcount reads and
798 		 * writes in sequential_heuristic().
799 		 */
800 		error = vn_lock(vp, LK_EXCLUSIVE);
801 		if (error != 0) {
802 			fdrop(fp, td);
803 			break;
804 		}
805 		if (arg >= 0) {
806 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
807 			arg = MIN(arg, INT_MAX - bsize + 1);
808 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
809 			    (arg + bsize - 1) / bsize);
810 			atomic_set_int(&fp->f_flag, FRDAHEAD);
811 		} else {
812 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
813 		}
814 		VOP_UNLOCK(vp);
815 		fdrop(fp, td);
816 		break;
817 
818 	case F_ISUNIONSTACK:
819 		/*
820 		 * Check if the vnode is part of a union stack (either the
821 		 * "union" flag from mount(2) or unionfs).
822 		 *
823 		 * Prior to introduction of this op libc's readdir would call
824 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
825 		 * data just to check fs name and a mount flag.
826 		 *
827 		 * Fixing the code to handle everything in the kernel instead
828 		 * is a non-trivial endeavor and has low priority, thus this
829 		 * horrible kludge facilitates the current behavior in a much
830 		 * cheaper manner until someone(tm) sorts this out.
831 		 */
832 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
833 		if (error != 0)
834 			break;
835 		if (fp->f_type != DTYPE_VNODE) {
836 			fdrop(fp, td);
837 			error = EBADF;
838 			break;
839 		}
840 		vp = fp->f_vnode;
841 		/*
842 		 * Since we don't prevent dooming the vnode even non-null mp
843 		 * found can become immediately stale. This is tolerable since
844 		 * mount points are type-stable (providing safe memory access)
845 		 * and any vfs op on this vnode going forward will return an
846 		 * error (meaning return value in this case is meaningless).
847 		 */
848 		mp = atomic_load_ptr(&vp->v_mount);
849 		if (__predict_false(mp == NULL)) {
850 			fdrop(fp, td);
851 			error = EBADF;
852 			break;
853 		}
854 		td->td_retval[0] = 0;
855 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
856 		    mp->mnt_flag & MNT_UNION)
857 			td->td_retval[0] = 1;
858 		fdrop(fp, td);
859 		break;
860 
861 	case F_KINFO:
862 #ifdef CAPABILITY_MODE
863 		if (IN_CAPABILITY_MODE(td)) {
864 			error = ECAPMODE;
865 			break;
866 		}
867 #endif
868 		error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
869 		if (error != 0)
870 			break;
871 		if (kif_sz != sizeof(*kif)) {
872 			error = EINVAL;
873 			break;
874 		}
875 		kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
876 		FILEDESC_SLOCK(fdp);
877 		error = fget_cap_locked(fdp, fd, &cap_fcntl_rights, &fp, NULL);
878 		if (error == 0 && fhold(fp)) {
879 			export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
880 			FILEDESC_SUNLOCK(fdp);
881 			fdrop(fp, td);
882 			if ((kif->kf_status & KF_ATTR_VALID) != 0) {
883 				kif->kf_structsize = sizeof(*kif);
884 				error = copyout(kif, (void *)arg, sizeof(*kif));
885 			} else {
886 				error = EBADF;
887 			}
888 		} else {
889 			FILEDESC_SUNLOCK(fdp);
890 			if (error == 0)
891 				error = EBADF;
892 		}
893 		free(kif, M_TEMP);
894 		break;
895 
896 	default:
897 		error = EINVAL;
898 		break;
899 	}
900 	return (error);
901 }
902 
903 static int
904 getmaxfd(struct thread *td)
905 {
906 
907 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
908 }
909 
910 /*
911  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
912  */
913 int
914 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
915 {
916 	struct filedesc *fdp;
917 	struct filedescent *oldfde, *newfde;
918 	struct proc *p;
919 	struct file *delfp, *oldfp;
920 	u_long *oioctls, *nioctls;
921 	int error, maxfd;
922 
923 	p = td->td_proc;
924 	fdp = p->p_fd;
925 	oioctls = NULL;
926 
927 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
928 	MPASS(mode < FDDUP_LASTMODE);
929 
930 	AUDIT_ARG_FD(old);
931 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
932 
933 	/*
934 	 * Verify we have a valid descriptor to dup from and possibly to
935 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
936 	 * return EINVAL when the new descriptor is out of bounds.
937 	 */
938 	if (old < 0)
939 		return (EBADF);
940 	if (new < 0)
941 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
942 	maxfd = getmaxfd(td);
943 	if (new >= maxfd)
944 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
945 
946 	error = EBADF;
947 	FILEDESC_XLOCK(fdp);
948 	if (fget_locked(fdp, old) == NULL)
949 		goto unlock;
950 	if (mode == FDDUP_FIXED && old == new) {
951 		td->td_retval[0] = new;
952 		if (flags & FDDUP_FLAG_CLOEXEC)
953 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
954 		error = 0;
955 		goto unlock;
956 	}
957 
958 	oldfde = &fdp->fd_ofiles[old];
959 	oldfp = oldfde->fde_file;
960 	if (!fhold(oldfp))
961 		goto unlock;
962 
963 	/*
964 	 * If the caller specified a file descriptor, make sure the file
965 	 * table is large enough to hold it, and grab it.  Otherwise, just
966 	 * allocate a new descriptor the usual way.
967 	 */
968 	switch (mode) {
969 	case FDDUP_NORMAL:
970 	case FDDUP_FCNTL:
971 		if ((error = fdalloc(td, new, &new)) != 0) {
972 			fdrop(oldfp, td);
973 			goto unlock;
974 		}
975 		break;
976 	case FDDUP_FIXED:
977 		if (new >= fdp->fd_nfiles) {
978 			/*
979 			 * The resource limits are here instead of e.g.
980 			 * fdalloc(), because the file descriptor table may be
981 			 * shared between processes, so we can't really use
982 			 * racct_add()/racct_sub().  Instead of counting the
983 			 * number of actually allocated descriptors, just put
984 			 * the limit on the size of the file descriptor table.
985 			 */
986 #ifdef RACCT
987 			if (RACCT_ENABLED()) {
988 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
989 				if (error != 0) {
990 					error = EMFILE;
991 					fdrop(oldfp, td);
992 					goto unlock;
993 				}
994 			}
995 #endif
996 			fdgrowtable_exp(fdp, new + 1);
997 		}
998 		if (!fdisused(fdp, new))
999 			fdused(fdp, new);
1000 		break;
1001 	default:
1002 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1003 	}
1004 
1005 	KASSERT(old != new, ("new fd is same as old"));
1006 
1007 	/* Refetch oldfde because the table may have grown and old one freed. */
1008 	oldfde = &fdp->fd_ofiles[old];
1009 	KASSERT(oldfp == oldfde->fde_file,
1010 	    ("fdt_ofiles shift from growth observed at fd %d",
1011 	    old));
1012 
1013 	newfde = &fdp->fd_ofiles[new];
1014 	delfp = newfde->fde_file;
1015 
1016 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1017 
1018 	/*
1019 	 * Duplicate the source descriptor.
1020 	 */
1021 #ifdef CAPABILITIES
1022 	seqc_write_begin(&newfde->fde_seqc);
1023 #endif
1024 	oioctls = filecaps_free_prep(&newfde->fde_caps);
1025 	memcpy(newfde, oldfde, fde_change_size);
1026 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1027 	    nioctls);
1028 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1029 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1030 	else
1031 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1032 #ifdef CAPABILITIES
1033 	seqc_write_end(&newfde->fde_seqc);
1034 #endif
1035 	td->td_retval[0] = new;
1036 
1037 	error = 0;
1038 
1039 	if (delfp != NULL) {
1040 		(void) closefp(fdp, new, delfp, td, true, false);
1041 		FILEDESC_UNLOCK_ASSERT(fdp);
1042 	} else {
1043 unlock:
1044 		FILEDESC_XUNLOCK(fdp);
1045 	}
1046 
1047 	filecaps_free_finish(oioctls);
1048 	return (error);
1049 }
1050 
1051 static void
1052 sigiofree(struct sigio *sigio)
1053 {
1054 	crfree(sigio->sio_ucred);
1055 	free(sigio, M_SIGIO);
1056 }
1057 
1058 static struct sigio *
1059 funsetown_locked(struct sigio *sigio)
1060 {
1061 	struct proc *p;
1062 	struct pgrp *pg;
1063 
1064 	SIGIO_ASSERT_LOCKED();
1065 
1066 	if (sigio == NULL)
1067 		return (NULL);
1068 	*sigio->sio_myref = NULL;
1069 	if (sigio->sio_pgid < 0) {
1070 		pg = sigio->sio_pgrp;
1071 		PGRP_LOCK(pg);
1072 		SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1073 		PGRP_UNLOCK(pg);
1074 	} else {
1075 		p = sigio->sio_proc;
1076 		PROC_LOCK(p);
1077 		SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1078 		PROC_UNLOCK(p);
1079 	}
1080 	return (sigio);
1081 }
1082 
1083 /*
1084  * If sigio is on the list associated with a process or process group,
1085  * disable signalling from the device, remove sigio from the list and
1086  * free sigio.
1087  */
1088 void
1089 funsetown(struct sigio **sigiop)
1090 {
1091 	struct sigio *sigio;
1092 
1093 	/* Racy check, consumers must provide synchronization. */
1094 	if (*sigiop == NULL)
1095 		return;
1096 
1097 	SIGIO_LOCK();
1098 	sigio = funsetown_locked(*sigiop);
1099 	SIGIO_UNLOCK();
1100 	if (sigio != NULL)
1101 		sigiofree(sigio);
1102 }
1103 
1104 /*
1105  * Free a list of sigio structures.  The caller must ensure that new sigio
1106  * structures cannot be added after this point.  For process groups this is
1107  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1108  * as an interlock.
1109  */
1110 void
1111 funsetownlst(struct sigiolst *sigiolst)
1112 {
1113 	struct proc *p;
1114 	struct pgrp *pg;
1115 	struct sigio *sigio, *tmp;
1116 
1117 	/* Racy check. */
1118 	sigio = SLIST_FIRST(sigiolst);
1119 	if (sigio == NULL)
1120 		return;
1121 
1122 	p = NULL;
1123 	pg = NULL;
1124 
1125 	SIGIO_LOCK();
1126 	sigio = SLIST_FIRST(sigiolst);
1127 	if (sigio == NULL) {
1128 		SIGIO_UNLOCK();
1129 		return;
1130 	}
1131 
1132 	/*
1133 	 * Every entry of the list should belong to a single proc or pgrp.
1134 	 */
1135 	if (sigio->sio_pgid < 0) {
1136 		pg = sigio->sio_pgrp;
1137 		sx_assert(&proctree_lock, SX_XLOCKED);
1138 		PGRP_LOCK(pg);
1139 	} else /* if (sigio->sio_pgid > 0) */ {
1140 		p = sigio->sio_proc;
1141 		PROC_LOCK(p);
1142 		KASSERT((p->p_flag & P_WEXIT) != 0,
1143 		    ("%s: process %p is not exiting", __func__, p));
1144 	}
1145 
1146 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1147 		*sigio->sio_myref = NULL;
1148 		if (pg != NULL) {
1149 			KASSERT(sigio->sio_pgid < 0,
1150 			    ("Proc sigio in pgrp sigio list"));
1151 			KASSERT(sigio->sio_pgrp == pg,
1152 			    ("Bogus pgrp in sigio list"));
1153 		} else /* if (p != NULL) */ {
1154 			KASSERT(sigio->sio_pgid > 0,
1155 			    ("Pgrp sigio in proc sigio list"));
1156 			KASSERT(sigio->sio_proc == p,
1157 			    ("Bogus proc in sigio list"));
1158 		}
1159 	}
1160 
1161 	if (pg != NULL)
1162 		PGRP_UNLOCK(pg);
1163 	else
1164 		PROC_UNLOCK(p);
1165 	SIGIO_UNLOCK();
1166 
1167 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1168 		sigiofree(sigio);
1169 }
1170 
1171 /*
1172  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1173  *
1174  * After permission checking, add a sigio structure to the sigio list for
1175  * the process or process group.
1176  */
1177 int
1178 fsetown(pid_t pgid, struct sigio **sigiop)
1179 {
1180 	struct proc *proc;
1181 	struct pgrp *pgrp;
1182 	struct sigio *osigio, *sigio;
1183 	int ret;
1184 
1185 	if (pgid == 0) {
1186 		funsetown(sigiop);
1187 		return (0);
1188 	}
1189 
1190 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1191 	sigio->sio_pgid = pgid;
1192 	sigio->sio_ucred = crhold(curthread->td_ucred);
1193 	sigio->sio_myref = sigiop;
1194 
1195 	ret = 0;
1196 	if (pgid > 0) {
1197 		ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1198 		SIGIO_LOCK();
1199 		osigio = funsetown_locked(*sigiop);
1200 		if (ret == 0) {
1201 			PROC_LOCK(proc);
1202 			_PRELE(proc);
1203 			if ((proc->p_flag & P_WEXIT) != 0) {
1204 				ret = ESRCH;
1205 			} else if (proc->p_session !=
1206 			    curthread->td_proc->p_session) {
1207 				/*
1208 				 * Policy - Don't allow a process to FSETOWN a
1209 				 * process in another session.
1210 				 *
1211 				 * Remove this test to allow maximum flexibility
1212 				 * or restrict FSETOWN to the current process or
1213 				 * process group for maximum safety.
1214 				 */
1215 				ret = EPERM;
1216 			} else {
1217 				sigio->sio_proc = proc;
1218 				SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1219 				    sio_pgsigio);
1220 			}
1221 			PROC_UNLOCK(proc);
1222 		}
1223 	} else /* if (pgid < 0) */ {
1224 		sx_slock(&proctree_lock);
1225 		SIGIO_LOCK();
1226 		osigio = funsetown_locked(*sigiop);
1227 		pgrp = pgfind(-pgid);
1228 		if (pgrp == NULL) {
1229 			ret = ESRCH;
1230 		} else {
1231 			if (pgrp->pg_session != curthread->td_proc->p_session) {
1232 				/*
1233 				 * Policy - Don't allow a process to FSETOWN a
1234 				 * process in another session.
1235 				 *
1236 				 * Remove this test to allow maximum flexibility
1237 				 * or restrict FSETOWN to the current process or
1238 				 * process group for maximum safety.
1239 				 */
1240 				ret = EPERM;
1241 			} else {
1242 				sigio->sio_pgrp = pgrp;
1243 				SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1244 				    sio_pgsigio);
1245 			}
1246 			PGRP_UNLOCK(pgrp);
1247 		}
1248 		sx_sunlock(&proctree_lock);
1249 	}
1250 	if (ret == 0)
1251 		*sigiop = sigio;
1252 	SIGIO_UNLOCK();
1253 	if (osigio != NULL)
1254 		sigiofree(osigio);
1255 	return (ret);
1256 }
1257 
1258 /*
1259  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1260  */
1261 pid_t
1262 fgetown(struct sigio **sigiop)
1263 {
1264 	pid_t pgid;
1265 
1266 	SIGIO_LOCK();
1267 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1268 	SIGIO_UNLOCK();
1269 	return (pgid);
1270 }
1271 
1272 static int
1273 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1274     bool audit)
1275 {
1276 	int error;
1277 
1278 	FILEDESC_XLOCK_ASSERT(fdp);
1279 
1280 	/*
1281 	 * We now hold the fp reference that used to be owned by the
1282 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1283 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1284 	 * added, and deleteing a knote for the new fd.
1285 	 */
1286 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1287 		knote_fdclose(td, fd);
1288 
1289 	/*
1290 	 * We need to notify mqueue if the object is of type mqueue.
1291 	 */
1292 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1293 		mq_fdclose(td, fd, fp);
1294 	FILEDESC_XUNLOCK(fdp);
1295 
1296 #ifdef AUDIT
1297 	if (AUDITING_TD(td) && audit)
1298 		audit_sysclose(td, fd, fp);
1299 #endif
1300 	error = closef(fp, td);
1301 
1302 	/*
1303 	 * All paths leading up to closefp() will have already removed or
1304 	 * replaced the fd in the filedesc table, so a restart would not
1305 	 * operate on the same file.
1306 	 */
1307 	if (error == ERESTART)
1308 		error = EINTR;
1309 
1310 	return (error);
1311 }
1312 
1313 static int
1314 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1315     bool holdleaders, bool audit)
1316 {
1317 	int error;
1318 
1319 	FILEDESC_XLOCK_ASSERT(fdp);
1320 
1321 	if (holdleaders) {
1322 		if (td->td_proc->p_fdtol != NULL) {
1323 			/*
1324 			 * Ask fdfree() to sleep to ensure that all relevant
1325 			 * process leaders can be traversed in closef().
1326 			 */
1327 			fdp->fd_holdleaderscount++;
1328 		} else {
1329 			holdleaders = false;
1330 		}
1331 	}
1332 
1333 	error = closefp_impl(fdp, fd, fp, td, audit);
1334 	if (holdleaders) {
1335 		FILEDESC_XLOCK(fdp);
1336 		fdp->fd_holdleaderscount--;
1337 		if (fdp->fd_holdleaderscount == 0 &&
1338 		    fdp->fd_holdleaderswakeup != 0) {
1339 			fdp->fd_holdleaderswakeup = 0;
1340 			wakeup(&fdp->fd_holdleaderscount);
1341 		}
1342 		FILEDESC_XUNLOCK(fdp);
1343 	}
1344 	return (error);
1345 }
1346 
1347 static int
1348 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1349     bool holdleaders, bool audit)
1350 {
1351 
1352 	FILEDESC_XLOCK_ASSERT(fdp);
1353 
1354 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1355 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1356 	} else {
1357 		return (closefp_impl(fdp, fd, fp, td, audit));
1358 	}
1359 }
1360 
1361 /*
1362  * Close a file descriptor.
1363  */
1364 #ifndef _SYS_SYSPROTO_H_
1365 struct close_args {
1366 	int     fd;
1367 };
1368 #endif
1369 /* ARGSUSED */
1370 int
1371 sys_close(struct thread *td, struct close_args *uap)
1372 {
1373 
1374 	return (kern_close(td, uap->fd));
1375 }
1376 
1377 int
1378 kern_close(struct thread *td, int fd)
1379 {
1380 	struct filedesc *fdp;
1381 	struct file *fp;
1382 
1383 	fdp = td->td_proc->p_fd;
1384 
1385 	FILEDESC_XLOCK(fdp);
1386 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1387 		FILEDESC_XUNLOCK(fdp);
1388 		return (EBADF);
1389 	}
1390 	fdfree(fdp, fd);
1391 
1392 	/* closefp() drops the FILEDESC lock for us. */
1393 	return (closefp(fdp, fd, fp, td, true, true));
1394 }
1395 
1396 int
1397 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1398 {
1399 	struct filedesc *fdp;
1400 	const struct fdescenttbl *fdt;
1401 	struct file *fp;
1402 	int fd;
1403 
1404 	/*
1405 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1406 	 * open should not be a usage error.  From a close_range() perspective,
1407 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1408 	 * be a usage error as all fd above 3 are in-fact already closed.
1409 	 */
1410 	if (highfd < lowfd) {
1411 		return (EINVAL);
1412 	}
1413 
1414 	fdp = td->td_proc->p_fd;
1415 	FILEDESC_XLOCK(fdp);
1416 	fdt = atomic_load_ptr(&fdp->fd_files);
1417 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1418 	fd = lowfd;
1419 	if (__predict_false(fd > highfd)) {
1420 		goto out_locked;
1421 	}
1422 	for (;;) {
1423 		fp = fdt->fdt_ofiles[fd].fde_file;
1424 		if (fp == NULL) {
1425 			if (fd == highfd)
1426 				goto out_locked;
1427 		} else {
1428 			fdfree(fdp, fd);
1429 			(void) closefp(fdp, fd, fp, td, true, true);
1430 			if (fd == highfd)
1431 				goto out_unlocked;
1432 			FILEDESC_XLOCK(fdp);
1433 			fdt = atomic_load_ptr(&fdp->fd_files);
1434 		}
1435 		fd++;
1436 	}
1437 out_locked:
1438 	FILEDESC_XUNLOCK(fdp);
1439 out_unlocked:
1440 	return (0);
1441 }
1442 
1443 #ifndef _SYS_SYSPROTO_H_
1444 struct close_range_args {
1445 	u_int	lowfd;
1446 	u_int	highfd;
1447 	int	flags;
1448 };
1449 #endif
1450 int
1451 sys_close_range(struct thread *td, struct close_range_args *uap)
1452 {
1453 
1454 	AUDIT_ARG_FD(uap->lowfd);
1455 	AUDIT_ARG_CMD(uap->highfd);
1456 	AUDIT_ARG_FFLAGS(uap->flags);
1457 
1458 	/* No flags currently defined */
1459 	if (uap->flags != 0)
1460 		return (EINVAL);
1461 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1462 }
1463 
1464 #ifdef COMPAT_FREEBSD12
1465 /*
1466  * Close open file descriptors.
1467  */
1468 #ifndef _SYS_SYSPROTO_H_
1469 struct freebsd12_closefrom_args {
1470 	int	lowfd;
1471 };
1472 #endif
1473 /* ARGSUSED */
1474 int
1475 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1476 {
1477 	u_int lowfd;
1478 
1479 	AUDIT_ARG_FD(uap->lowfd);
1480 
1481 	/*
1482 	 * Treat negative starting file descriptor values identical to
1483 	 * closefrom(0) which closes all files.
1484 	 */
1485 	lowfd = MAX(0, uap->lowfd);
1486 	return (kern_close_range(td, lowfd, ~0U));
1487 }
1488 #endif	/* COMPAT_FREEBSD12 */
1489 
1490 #if defined(COMPAT_43)
1491 /*
1492  * Return status information about a file descriptor.
1493  */
1494 #ifndef _SYS_SYSPROTO_H_
1495 struct ofstat_args {
1496 	int	fd;
1497 	struct	ostat *sb;
1498 };
1499 #endif
1500 /* ARGSUSED */
1501 int
1502 ofstat(struct thread *td, struct ofstat_args *uap)
1503 {
1504 	struct ostat oub;
1505 	struct stat ub;
1506 	int error;
1507 
1508 	error = kern_fstat(td, uap->fd, &ub);
1509 	if (error == 0) {
1510 		cvtstat(&ub, &oub);
1511 		error = copyout(&oub, uap->sb, sizeof(oub));
1512 	}
1513 	return (error);
1514 }
1515 #endif /* COMPAT_43 */
1516 
1517 #if defined(COMPAT_FREEBSD11)
1518 int
1519 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1520 {
1521 	struct stat sb;
1522 	struct freebsd11_stat osb;
1523 	int error;
1524 
1525 	error = kern_fstat(td, uap->fd, &sb);
1526 	if (error != 0)
1527 		return (error);
1528 	error = freebsd11_cvtstat(&sb, &osb);
1529 	if (error == 0)
1530 		error = copyout(&osb, uap->sb, sizeof(osb));
1531 	return (error);
1532 }
1533 #endif	/* COMPAT_FREEBSD11 */
1534 
1535 /*
1536  * Return status information about a file descriptor.
1537  */
1538 #ifndef _SYS_SYSPROTO_H_
1539 struct fstat_args {
1540 	int	fd;
1541 	struct	stat *sb;
1542 };
1543 #endif
1544 /* ARGSUSED */
1545 int
1546 sys_fstat(struct thread *td, struct fstat_args *uap)
1547 {
1548 	struct stat ub;
1549 	int error;
1550 
1551 	error = kern_fstat(td, uap->fd, &ub);
1552 	if (error == 0)
1553 		error = copyout(&ub, uap->sb, sizeof(ub));
1554 	return (error);
1555 }
1556 
1557 int
1558 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1559 {
1560 	struct file *fp;
1561 	int error;
1562 
1563 	AUDIT_ARG_FD(fd);
1564 
1565 	error = fget(td, fd, &cap_fstat_rights, &fp);
1566 	if (__predict_false(error != 0))
1567 		return (error);
1568 
1569 	AUDIT_ARG_FILE(td->td_proc, fp);
1570 
1571 	error = fo_stat(fp, sbp, td->td_ucred);
1572 	fdrop(fp, td);
1573 #ifdef __STAT_TIME_T_EXT
1574 	sbp->st_atim_ext = 0;
1575 	sbp->st_mtim_ext = 0;
1576 	sbp->st_ctim_ext = 0;
1577 	sbp->st_btim_ext = 0;
1578 #endif
1579 #ifdef KTRACE
1580 	if (KTRPOINT(td, KTR_STRUCT))
1581 		ktrstat_error(sbp, error);
1582 #endif
1583 	return (error);
1584 }
1585 
1586 #if defined(COMPAT_FREEBSD11)
1587 /*
1588  * Return status information about a file descriptor.
1589  */
1590 #ifndef _SYS_SYSPROTO_H_
1591 struct freebsd11_nfstat_args {
1592 	int	fd;
1593 	struct	nstat *sb;
1594 };
1595 #endif
1596 /* ARGSUSED */
1597 int
1598 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1599 {
1600 	struct nstat nub;
1601 	struct stat ub;
1602 	int error;
1603 
1604 	error = kern_fstat(td, uap->fd, &ub);
1605 	if (error != 0)
1606 		return (error);
1607 	error = freebsd11_cvtnstat(&ub, &nub);
1608 	if (error != 0)
1609 		error = copyout(&nub, uap->sb, sizeof(nub));
1610 	return (error);
1611 }
1612 #endif /* COMPAT_FREEBSD11 */
1613 
1614 /*
1615  * Return pathconf information about a file descriptor.
1616  */
1617 #ifndef _SYS_SYSPROTO_H_
1618 struct fpathconf_args {
1619 	int	fd;
1620 	int	name;
1621 };
1622 #endif
1623 /* ARGSUSED */
1624 int
1625 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1626 {
1627 	long value;
1628 	int error;
1629 
1630 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1631 	if (error == 0)
1632 		td->td_retval[0] = value;
1633 	return (error);
1634 }
1635 
1636 int
1637 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1638 {
1639 	struct file *fp;
1640 	struct vnode *vp;
1641 	int error;
1642 
1643 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1644 	if (error != 0)
1645 		return (error);
1646 
1647 	if (name == _PC_ASYNC_IO) {
1648 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1649 		goto out;
1650 	}
1651 	vp = fp->f_vnode;
1652 	if (vp != NULL) {
1653 		vn_lock(vp, LK_SHARED | LK_RETRY);
1654 		error = VOP_PATHCONF(vp, name, valuep);
1655 		VOP_UNLOCK(vp);
1656 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1657 		if (name != _PC_PIPE_BUF) {
1658 			error = EINVAL;
1659 		} else {
1660 			*valuep = PIPE_BUF;
1661 			error = 0;
1662 		}
1663 	} else {
1664 		error = EOPNOTSUPP;
1665 	}
1666 out:
1667 	fdrop(fp, td);
1668 	return (error);
1669 }
1670 
1671 /*
1672  * Copy filecaps structure allocating memory for ioctls array if needed.
1673  *
1674  * The last parameter indicates whether the fdtable is locked. If it is not and
1675  * ioctls are encountered, copying fails and the caller must lock the table.
1676  *
1677  * Note that if the table was not locked, the caller has to check the relevant
1678  * sequence counter to determine whether the operation was successful.
1679  */
1680 bool
1681 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1682 {
1683 	size_t size;
1684 
1685 	if (src->fc_ioctls != NULL && !locked)
1686 		return (false);
1687 	memcpy(dst, src, sizeof(*src));
1688 	if (src->fc_ioctls == NULL)
1689 		return (true);
1690 
1691 	KASSERT(src->fc_nioctls > 0,
1692 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1693 
1694 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1695 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1696 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1697 	return (true);
1698 }
1699 
1700 static u_long *
1701 filecaps_copy_prep(const struct filecaps *src)
1702 {
1703 	u_long *ioctls;
1704 	size_t size;
1705 
1706 	if (__predict_true(src->fc_ioctls == NULL))
1707 		return (NULL);
1708 
1709 	KASSERT(src->fc_nioctls > 0,
1710 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1711 
1712 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1713 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1714 	return (ioctls);
1715 }
1716 
1717 static void
1718 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1719     u_long *ioctls)
1720 {
1721 	size_t size;
1722 
1723 	*dst = *src;
1724 	if (__predict_true(src->fc_ioctls == NULL)) {
1725 		MPASS(ioctls == NULL);
1726 		return;
1727 	}
1728 
1729 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1730 	dst->fc_ioctls = ioctls;
1731 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1732 }
1733 
1734 /*
1735  * Move filecaps structure to the new place and clear the old place.
1736  */
1737 void
1738 filecaps_move(struct filecaps *src, struct filecaps *dst)
1739 {
1740 
1741 	*dst = *src;
1742 	bzero(src, sizeof(*src));
1743 }
1744 
1745 /*
1746  * Fill the given filecaps structure with full rights.
1747  */
1748 static void
1749 filecaps_fill(struct filecaps *fcaps)
1750 {
1751 
1752 	CAP_ALL(&fcaps->fc_rights);
1753 	fcaps->fc_ioctls = NULL;
1754 	fcaps->fc_nioctls = -1;
1755 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1756 }
1757 
1758 /*
1759  * Free memory allocated within filecaps structure.
1760  */
1761 void
1762 filecaps_free(struct filecaps *fcaps)
1763 {
1764 
1765 	free(fcaps->fc_ioctls, M_FILECAPS);
1766 	bzero(fcaps, sizeof(*fcaps));
1767 }
1768 
1769 static u_long *
1770 filecaps_free_prep(struct filecaps *fcaps)
1771 {
1772 	u_long *ioctls;
1773 
1774 	ioctls = fcaps->fc_ioctls;
1775 	bzero(fcaps, sizeof(*fcaps));
1776 	return (ioctls);
1777 }
1778 
1779 static void
1780 filecaps_free_finish(u_long *ioctls)
1781 {
1782 
1783 	free(ioctls, M_FILECAPS);
1784 }
1785 
1786 /*
1787  * Validate the given filecaps structure.
1788  */
1789 static void
1790 filecaps_validate(const struct filecaps *fcaps, const char *func)
1791 {
1792 
1793 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1794 	    ("%s: invalid rights", func));
1795 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1796 	    ("%s: invalid fcntls", func));
1797 	KASSERT(fcaps->fc_fcntls == 0 ||
1798 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1799 	    ("%s: fcntls without CAP_FCNTL", func));
1800 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1801 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1802 	    ("%s: invalid ioctls", func));
1803 	KASSERT(fcaps->fc_nioctls == 0 ||
1804 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1805 	    ("%s: ioctls without CAP_IOCTL", func));
1806 }
1807 
1808 static void
1809 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1810 {
1811 	int nfd1;
1812 
1813 	FILEDESC_XLOCK_ASSERT(fdp);
1814 
1815 	nfd1 = fdp->fd_nfiles * 2;
1816 	if (nfd1 < nfd)
1817 		nfd1 = nfd;
1818 	fdgrowtable(fdp, nfd1);
1819 }
1820 
1821 /*
1822  * Grow the file table to accommodate (at least) nfd descriptors.
1823  */
1824 static void
1825 fdgrowtable(struct filedesc *fdp, int nfd)
1826 {
1827 	struct filedesc0 *fdp0;
1828 	struct freetable *ft;
1829 	struct fdescenttbl *ntable;
1830 	struct fdescenttbl *otable;
1831 	int nnfiles, onfiles;
1832 	NDSLOTTYPE *nmap, *omap;
1833 
1834 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1835 
1836 	/* save old values */
1837 	onfiles = fdp->fd_nfiles;
1838 	otable = fdp->fd_files;
1839 	omap = fdp->fd_map;
1840 
1841 	/* compute the size of the new table */
1842 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1843 	if (nnfiles <= onfiles)
1844 		/* the table is already large enough */
1845 		return;
1846 
1847 	/*
1848 	 * Allocate a new table.  We need enough space for the number of
1849 	 * entries, file entries themselves and the struct freetable we will use
1850 	 * when we decommission the table and place it on the freelist.
1851 	 * We place the struct freetable in the middle so we don't have
1852 	 * to worry about padding.
1853 	 */
1854 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1855 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1856 	    sizeof(struct freetable),
1857 	    M_FILEDESC, M_ZERO | M_WAITOK);
1858 	/* copy the old data */
1859 	ntable->fdt_nfiles = nnfiles;
1860 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1861 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1862 
1863 	/*
1864 	 * Allocate a new map only if the old is not large enough.  It will
1865 	 * grow at a slower rate than the table as it can map more
1866 	 * entries than the table can hold.
1867 	 */
1868 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1869 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1870 		    M_ZERO | M_WAITOK);
1871 		/* copy over the old data and update the pointer */
1872 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1873 		fdp->fd_map = nmap;
1874 	}
1875 
1876 	/*
1877 	 * Make sure that ntable is correctly initialized before we replace
1878 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1879 	 * data.
1880 	 */
1881 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1882 
1883 	/*
1884 	 * Free the old file table when not shared by other threads or processes.
1885 	 * The old file table is considered to be shared when either are true:
1886 	 * - The process has more than one thread.
1887 	 * - The file descriptor table has been shared via fdshare().
1888 	 *
1889 	 * When shared, the old file table will be placed on a freelist
1890 	 * which will be processed when the struct filedesc is released.
1891 	 *
1892 	 * Note that if onfiles == NDFILE, we're dealing with the original
1893 	 * static allocation contained within (struct filedesc0 *)fdp,
1894 	 * which must not be freed.
1895 	 */
1896 	if (onfiles > NDFILE) {
1897 		/*
1898 		 * Note we may be called here from fdinit while allocating a
1899 		 * table for a new process in which case ->p_fd points
1900 		 * elsewhere.
1901 		 */
1902 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1903 			free(otable, M_FILEDESC);
1904 		} else {
1905 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1906 			fdp0 = (struct filedesc0 *)fdp;
1907 			ft->ft_table = otable;
1908 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1909 		}
1910 	}
1911 	/*
1912 	 * The map does not have the same possibility of threads still
1913 	 * holding references to it.  So always free it as long as it
1914 	 * does not reference the original static allocation.
1915 	 */
1916 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1917 		free(omap, M_FILEDESC);
1918 }
1919 
1920 /*
1921  * Allocate a file descriptor for the process.
1922  */
1923 int
1924 fdalloc(struct thread *td, int minfd, int *result)
1925 {
1926 	struct proc *p = td->td_proc;
1927 	struct filedesc *fdp = p->p_fd;
1928 	int fd, maxfd, allocfd;
1929 #ifdef RACCT
1930 	int error;
1931 #endif
1932 
1933 	FILEDESC_XLOCK_ASSERT(fdp);
1934 
1935 	if (fdp->fd_freefile > minfd)
1936 		minfd = fdp->fd_freefile;
1937 
1938 	maxfd = getmaxfd(td);
1939 
1940 	/*
1941 	 * Search the bitmap for a free descriptor starting at minfd.
1942 	 * If none is found, grow the file table.
1943 	 */
1944 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1945 	if (__predict_false(fd >= maxfd))
1946 		return (EMFILE);
1947 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1948 		allocfd = min(fd * 2, maxfd);
1949 #ifdef RACCT
1950 		if (RACCT_ENABLED()) {
1951 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1952 			if (error != 0)
1953 				return (EMFILE);
1954 		}
1955 #endif
1956 		/*
1957 		 * fd is already equal to first free descriptor >= minfd, so
1958 		 * we only need to grow the table and we are done.
1959 		 */
1960 		fdgrowtable_exp(fdp, allocfd);
1961 	}
1962 
1963 	/*
1964 	 * Perform some sanity checks, then mark the file descriptor as
1965 	 * used and return it to the caller.
1966 	 */
1967 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1968 	    ("invalid descriptor %d", fd));
1969 	KASSERT(!fdisused(fdp, fd),
1970 	    ("fd_first_free() returned non-free descriptor"));
1971 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1972 	    ("file descriptor isn't free"));
1973 	fdused(fdp, fd);
1974 	*result = fd;
1975 	return (0);
1976 }
1977 
1978 /*
1979  * Allocate n file descriptors for the process.
1980  */
1981 int
1982 fdallocn(struct thread *td, int minfd, int *fds, int n)
1983 {
1984 	struct proc *p = td->td_proc;
1985 	struct filedesc *fdp = p->p_fd;
1986 	int i;
1987 
1988 	FILEDESC_XLOCK_ASSERT(fdp);
1989 
1990 	for (i = 0; i < n; i++)
1991 		if (fdalloc(td, 0, &fds[i]) != 0)
1992 			break;
1993 
1994 	if (i < n) {
1995 		for (i--; i >= 0; i--)
1996 			fdunused(fdp, fds[i]);
1997 		return (EMFILE);
1998 	}
1999 
2000 	return (0);
2001 }
2002 
2003 /*
2004  * Create a new open file structure and allocate a file descriptor for the
2005  * process that refers to it.  We add one reference to the file for the
2006  * descriptor table and one reference for resultfp. This is to prevent us
2007  * being preempted and the entry in the descriptor table closed after we
2008  * release the FILEDESC lock.
2009  */
2010 int
2011 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2012     struct filecaps *fcaps)
2013 {
2014 	struct file *fp;
2015 	int error, fd;
2016 
2017 	MPASS(resultfp != NULL);
2018 	MPASS(resultfd != NULL);
2019 
2020 	error = _falloc_noinstall(td, &fp, 2);
2021 	if (__predict_false(error != 0)) {
2022 		return (error);
2023 	}
2024 
2025 	error = finstall_refed(td, fp, &fd, flags, fcaps);
2026 	if (__predict_false(error != 0)) {
2027 		falloc_abort(td, fp);
2028 		return (error);
2029 	}
2030 
2031 	*resultfp = fp;
2032 	*resultfd = fd;
2033 
2034 	return (0);
2035 }
2036 
2037 /*
2038  * Create a new open file structure without allocating a file descriptor.
2039  */
2040 int
2041 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2042 {
2043 	struct file *fp;
2044 	int maxuserfiles = maxfiles - (maxfiles / 20);
2045 	int openfiles_new;
2046 	static struct timeval lastfail;
2047 	static int curfail;
2048 
2049 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2050 	MPASS(n > 0);
2051 
2052 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2053 	if ((openfiles_new >= maxuserfiles &&
2054 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2055 	    openfiles_new >= maxfiles) {
2056 		atomic_subtract_int(&openfiles, 1);
2057 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2058 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2059 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2060 		}
2061 		return (ENFILE);
2062 	}
2063 	fp = uma_zalloc(file_zone, M_WAITOK);
2064 	bzero(fp, sizeof(*fp));
2065 	refcount_init(&fp->f_count, n);
2066 	fp->f_cred = crhold(td->td_ucred);
2067 	fp->f_ops = &badfileops;
2068 	*resultfp = fp;
2069 	return (0);
2070 }
2071 
2072 void
2073 falloc_abort(struct thread *td, struct file *fp)
2074 {
2075 
2076 	/*
2077 	 * For assertion purposes.
2078 	 */
2079 	refcount_init(&fp->f_count, 0);
2080 	_fdrop(fp, td);
2081 }
2082 
2083 /*
2084  * Install a file in a file descriptor table.
2085  */
2086 void
2087 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2088     struct filecaps *fcaps)
2089 {
2090 	struct filedescent *fde;
2091 
2092 	MPASS(fp != NULL);
2093 	if (fcaps != NULL)
2094 		filecaps_validate(fcaps, __func__);
2095 	FILEDESC_XLOCK_ASSERT(fdp);
2096 
2097 	fde = &fdp->fd_ofiles[fd];
2098 #ifdef CAPABILITIES
2099 	seqc_write_begin(&fde->fde_seqc);
2100 #endif
2101 	fde->fde_file = fp;
2102 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2103 	if (fcaps != NULL)
2104 		filecaps_move(fcaps, &fde->fde_caps);
2105 	else
2106 		filecaps_fill(&fde->fde_caps);
2107 #ifdef CAPABILITIES
2108 	seqc_write_end(&fde->fde_seqc);
2109 #endif
2110 }
2111 
2112 int
2113 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2114     struct filecaps *fcaps)
2115 {
2116 	struct filedesc *fdp = td->td_proc->p_fd;
2117 	int error;
2118 
2119 	MPASS(fd != NULL);
2120 
2121 	FILEDESC_XLOCK(fdp);
2122 	error = fdalloc(td, 0, fd);
2123 	if (__predict_true(error == 0)) {
2124 		_finstall(fdp, fp, *fd, flags, fcaps);
2125 	}
2126 	FILEDESC_XUNLOCK(fdp);
2127 	return (error);
2128 }
2129 
2130 int
2131 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2132     struct filecaps *fcaps)
2133 {
2134 	int error;
2135 
2136 	MPASS(fd != NULL);
2137 
2138 	if (!fhold(fp))
2139 		return (EBADF);
2140 	error = finstall_refed(td, fp, fd, flags, fcaps);
2141 	if (__predict_false(error != 0)) {
2142 		fdrop(fp, td);
2143 	}
2144 	return (error);
2145 }
2146 
2147 /*
2148  * Build a new filedesc structure from another.
2149  *
2150  * If fdp is not NULL, return with it shared locked.
2151  */
2152 struct filedesc *
2153 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2154 {
2155 	struct filedesc0 *newfdp0;
2156 	struct filedesc *newfdp;
2157 
2158 	if (prepfiles)
2159 		MPASS(lastfile != NULL);
2160 	else
2161 		MPASS(lastfile == NULL);
2162 
2163 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2164 	newfdp = &newfdp0->fd_fd;
2165 
2166 	/* Create the file descriptor table. */
2167 	FILEDESC_LOCK_INIT(newfdp);
2168 	refcount_init(&newfdp->fd_refcnt, 1);
2169 	refcount_init(&newfdp->fd_holdcnt, 1);
2170 	newfdp->fd_map = newfdp0->fd_dmap;
2171 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2172 	newfdp->fd_files->fdt_nfiles = NDFILE;
2173 
2174 	if (fdp == NULL)
2175 		return (newfdp);
2176 
2177 	FILEDESC_SLOCK(fdp);
2178 	if (!prepfiles) {
2179 		FILEDESC_SUNLOCK(fdp);
2180 		return (newfdp);
2181 	}
2182 
2183 	for (;;) {
2184 		*lastfile = fdlastfile(fdp);
2185 		if (*lastfile < newfdp->fd_nfiles)
2186 			break;
2187 		FILEDESC_SUNLOCK(fdp);
2188 		fdgrowtable(newfdp, *lastfile + 1);
2189 		FILEDESC_SLOCK(fdp);
2190 	}
2191 
2192 	return (newfdp);
2193 }
2194 
2195 /*
2196  * Build a pwddesc structure from another.
2197  * Copy the current, root, and jail root vnode references.
2198  *
2199  * If pdp is not NULL, return with it shared locked.
2200  */
2201 struct pwddesc *
2202 pdinit(struct pwddesc *pdp, bool keeplock)
2203 {
2204 	struct pwddesc *newpdp;
2205 	struct pwd *newpwd;
2206 
2207 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2208 
2209 	PWDDESC_LOCK_INIT(newpdp);
2210 	refcount_init(&newpdp->pd_refcount, 1);
2211 	newpdp->pd_cmask = CMASK;
2212 
2213 	if (pdp == NULL) {
2214 		newpwd = pwd_alloc();
2215 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2216 		return (newpdp);
2217 	}
2218 
2219 	PWDDESC_XLOCK(pdp);
2220 	newpwd = pwd_hold_pwddesc(pdp);
2221 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2222 	if (!keeplock)
2223 		PWDDESC_XUNLOCK(pdp);
2224 	return (newpdp);
2225 }
2226 
2227 /*
2228  * Hold either filedesc or pwddesc of the passed process.
2229  *
2230  * The process lock is used to synchronize against the target exiting and
2231  * freeing the data.
2232  *
2233  * Clearing can be ilustrated in 3 steps:
2234  * 1. set the pointer to NULL. Either routine can race against it, hence
2235  *   atomic_load_ptr.
2236  * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2237  *   race to either still see the pointer or find NULL. It is still safe to
2238  *   grab a reference as clearing is stalled.
2239  * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2240  *   guaranteed to see NULL, making it safe to finish clearing
2241  */
2242 static struct filedesc *
2243 fdhold(struct proc *p)
2244 {
2245 	struct filedesc *fdp;
2246 
2247 	PROC_LOCK_ASSERT(p, MA_OWNED);
2248 	fdp = atomic_load_ptr(&p->p_fd);
2249 	if (fdp != NULL)
2250 		refcount_acquire(&fdp->fd_holdcnt);
2251 	return (fdp);
2252 }
2253 
2254 static struct pwddesc *
2255 pdhold(struct proc *p)
2256 {
2257 	struct pwddesc *pdp;
2258 
2259 	PROC_LOCK_ASSERT(p, MA_OWNED);
2260 	pdp = atomic_load_ptr(&p->p_pd);
2261 	if (pdp != NULL)
2262 		refcount_acquire(&pdp->pd_refcount);
2263 	return (pdp);
2264 }
2265 
2266 static void
2267 fddrop(struct filedesc *fdp)
2268 {
2269 
2270 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2271 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2272 			return;
2273 	}
2274 
2275 	FILEDESC_LOCK_DESTROY(fdp);
2276 	uma_zfree(filedesc0_zone, fdp);
2277 }
2278 
2279 static void
2280 pddrop(struct pwddesc *pdp)
2281 {
2282 	struct pwd *pwd;
2283 
2284 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2285 		return;
2286 
2287 	PWDDESC_XLOCK(pdp);
2288 	if (refcount_release(&pdp->pd_refcount) == 0) {
2289 		PWDDESC_XUNLOCK(pdp);
2290 		return;
2291 	}
2292 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2293 	pwd_set(pdp, NULL);
2294 	PWDDESC_XUNLOCK(pdp);
2295 	pwd_drop(pwd);
2296 
2297 	PWDDESC_LOCK_DESTROY(pdp);
2298 	free(pdp, M_PWDDESC);
2299 }
2300 
2301 /*
2302  * Share a filedesc structure.
2303  */
2304 struct filedesc *
2305 fdshare(struct filedesc *fdp)
2306 {
2307 
2308 	refcount_acquire(&fdp->fd_refcnt);
2309 	return (fdp);
2310 }
2311 
2312 /*
2313  * Share a pwddesc structure.
2314  */
2315 struct pwddesc *
2316 pdshare(struct pwddesc *pdp)
2317 {
2318 	refcount_acquire(&pdp->pd_refcount);
2319 	return (pdp);
2320 }
2321 
2322 /*
2323  * Unshare a filedesc structure, if necessary by making a copy
2324  */
2325 void
2326 fdunshare(struct thread *td)
2327 {
2328 	struct filedesc *tmp;
2329 	struct proc *p = td->td_proc;
2330 
2331 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2332 		return;
2333 
2334 	tmp = fdcopy(p->p_fd);
2335 	fdescfree(td);
2336 	p->p_fd = tmp;
2337 }
2338 
2339 /*
2340  * Unshare a pwddesc structure.
2341  */
2342 void
2343 pdunshare(struct thread *td)
2344 {
2345 	struct pwddesc *pdp;
2346 	struct proc *p;
2347 
2348 	p = td->td_proc;
2349 	/* Not shared. */
2350 	if (p->p_pd->pd_refcount == 1)
2351 		return;
2352 
2353 	pdp = pdcopy(p->p_pd);
2354 	pdescfree(td);
2355 	p->p_pd = pdp;
2356 }
2357 
2358 /*
2359  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2360  * this is to ease callers, not catch errors.
2361  */
2362 struct filedesc *
2363 fdcopy(struct filedesc *fdp)
2364 {
2365 	struct filedesc *newfdp;
2366 	struct filedescent *nfde, *ofde;
2367 	int i, lastfile;
2368 
2369 	MPASS(fdp != NULL);
2370 
2371 	newfdp = fdinit(fdp, true, &lastfile);
2372 	/* copy all passable descriptors (i.e. not kqueue) */
2373 	newfdp->fd_freefile = -1;
2374 	for (i = 0; i <= lastfile; ++i) {
2375 		ofde = &fdp->fd_ofiles[i];
2376 		if (ofde->fde_file == NULL ||
2377 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2378 		    !fhold(ofde->fde_file)) {
2379 			if (newfdp->fd_freefile == -1)
2380 				newfdp->fd_freefile = i;
2381 			continue;
2382 		}
2383 		nfde = &newfdp->fd_ofiles[i];
2384 		*nfde = *ofde;
2385 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2386 		fdused_init(newfdp, i);
2387 	}
2388 	if (newfdp->fd_freefile == -1)
2389 		newfdp->fd_freefile = i;
2390 	FILEDESC_SUNLOCK(fdp);
2391 	return (newfdp);
2392 }
2393 
2394 /*
2395  * Copy a pwddesc structure.
2396  */
2397 struct pwddesc *
2398 pdcopy(struct pwddesc *pdp)
2399 {
2400 	struct pwddesc *newpdp;
2401 
2402 	MPASS(pdp != NULL);
2403 
2404 	newpdp = pdinit(pdp, true);
2405 	newpdp->pd_cmask = pdp->pd_cmask;
2406 	PWDDESC_XUNLOCK(pdp);
2407 	return (newpdp);
2408 }
2409 
2410 /*
2411  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2412  * one of processes using it exits) and the table used to be shared.
2413  */
2414 static void
2415 fdclearlocks(struct thread *td)
2416 {
2417 	struct filedesc *fdp;
2418 	struct filedesc_to_leader *fdtol;
2419 	struct flock lf;
2420 	struct file *fp;
2421 	struct proc *p;
2422 	struct vnode *vp;
2423 	int i, lastfile;
2424 
2425 	p = td->td_proc;
2426 	fdp = p->p_fd;
2427 	fdtol = p->p_fdtol;
2428 	MPASS(fdtol != NULL);
2429 
2430 	FILEDESC_XLOCK(fdp);
2431 	KASSERT(fdtol->fdl_refcount > 0,
2432 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2433 	    fdtol->fdl_refcount));
2434 	if (fdtol->fdl_refcount == 1 &&
2435 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2436 		lastfile = fdlastfile(fdp);
2437 		for (i = 0; i <= lastfile; i++) {
2438 			fp = fdp->fd_ofiles[i].fde_file;
2439 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2440 			    !fhold(fp))
2441 				continue;
2442 			FILEDESC_XUNLOCK(fdp);
2443 			lf.l_whence = SEEK_SET;
2444 			lf.l_start = 0;
2445 			lf.l_len = 0;
2446 			lf.l_type = F_UNLCK;
2447 			vp = fp->f_vnode;
2448 			(void) VOP_ADVLOCK(vp,
2449 			    (caddr_t)p->p_leader, F_UNLCK,
2450 			    &lf, F_POSIX);
2451 			FILEDESC_XLOCK(fdp);
2452 			fdrop(fp, td);
2453 		}
2454 	}
2455 retry:
2456 	if (fdtol->fdl_refcount == 1) {
2457 		if (fdp->fd_holdleaderscount > 0 &&
2458 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2459 			/*
2460 			 * close() or kern_dup() has cleared a reference
2461 			 * in a shared file descriptor table.
2462 			 */
2463 			fdp->fd_holdleaderswakeup = 1;
2464 			sx_sleep(&fdp->fd_holdleaderscount,
2465 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2466 			goto retry;
2467 		}
2468 		if (fdtol->fdl_holdcount > 0) {
2469 			/*
2470 			 * Ensure that fdtol->fdl_leader remains
2471 			 * valid in closef().
2472 			 */
2473 			fdtol->fdl_wakeup = 1;
2474 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2475 			    "fdlhold", 0);
2476 			goto retry;
2477 		}
2478 	}
2479 	fdtol->fdl_refcount--;
2480 	if (fdtol->fdl_refcount == 0 &&
2481 	    fdtol->fdl_holdcount == 0) {
2482 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2483 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2484 	} else
2485 		fdtol = NULL;
2486 	p->p_fdtol = NULL;
2487 	FILEDESC_XUNLOCK(fdp);
2488 	if (fdtol != NULL)
2489 		free(fdtol, M_FILEDESC_TO_LEADER);
2490 }
2491 
2492 /*
2493  * Release a filedesc structure.
2494  */
2495 static void
2496 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2497 {
2498 	struct filedesc0 *fdp0;
2499 	struct freetable *ft, *tft;
2500 	struct filedescent *fde;
2501 	struct file *fp;
2502 	int i, lastfile;
2503 
2504 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2505 	    ("%s: fd table %p carries references", __func__, fdp));
2506 
2507 	/*
2508 	 * Serialize with threads iterating over the table, if any.
2509 	 */
2510 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2511 		FILEDESC_XLOCK(fdp);
2512 		FILEDESC_XUNLOCK(fdp);
2513 	}
2514 
2515 	lastfile = fdlastfile_single(fdp);
2516 	for (i = 0; i <= lastfile; i++) {
2517 		fde = &fdp->fd_ofiles[i];
2518 		fp = fde->fde_file;
2519 		if (fp != NULL) {
2520 			fdefree_last(fde);
2521 			(void) closef(fp, td);
2522 		}
2523 	}
2524 
2525 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2526 		free(fdp->fd_map, M_FILEDESC);
2527 	if (fdp->fd_nfiles > NDFILE)
2528 		free(fdp->fd_files, M_FILEDESC);
2529 
2530 	fdp0 = (struct filedesc0 *)fdp;
2531 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2532 		free(ft->ft_table, M_FILEDESC);
2533 
2534 	fddrop(fdp);
2535 }
2536 
2537 void
2538 fdescfree(struct thread *td)
2539 {
2540 	struct proc *p;
2541 	struct filedesc *fdp;
2542 
2543 	p = td->td_proc;
2544 	fdp = p->p_fd;
2545 	MPASS(fdp != NULL);
2546 
2547 #ifdef RACCT
2548 	if (RACCT_ENABLED())
2549 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2550 #endif
2551 
2552 	if (p->p_fdtol != NULL)
2553 		fdclearlocks(td);
2554 
2555 	/*
2556 	 * Check fdhold for an explanation.
2557 	 */
2558 	atomic_store_ptr(&p->p_fd, NULL);
2559 	atomic_thread_fence_seq_cst();
2560 	PROC_WAIT_UNLOCKED(p);
2561 
2562 	if (refcount_release(&fdp->fd_refcnt) == 0)
2563 		return;
2564 
2565 	fdescfree_fds(td, fdp);
2566 }
2567 
2568 void
2569 pdescfree(struct thread *td)
2570 {
2571 	struct proc *p;
2572 	struct pwddesc *pdp;
2573 
2574 	p = td->td_proc;
2575 	pdp = p->p_pd;
2576 	MPASS(pdp != NULL);
2577 
2578 	/*
2579 	 * Check pdhold for an explanation.
2580 	 */
2581 	atomic_store_ptr(&p->p_pd, NULL);
2582 	atomic_thread_fence_seq_cst();
2583 	PROC_WAIT_UNLOCKED(p);
2584 
2585 	pddrop(pdp);
2586 }
2587 
2588 /*
2589  * For setugid programs, we don't want to people to use that setugidness
2590  * to generate error messages which write to a file which otherwise would
2591  * otherwise be off-limits to the process.  We check for filesystems where
2592  * the vnode can change out from under us after execve (like [lin]procfs).
2593  *
2594  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2595  * sufficient.  We also don't check for setugidness since we know we are.
2596  */
2597 static bool
2598 is_unsafe(struct file *fp)
2599 {
2600 	struct vnode *vp;
2601 
2602 	if (fp->f_type != DTYPE_VNODE)
2603 		return (false);
2604 
2605 	vp = fp->f_vnode;
2606 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2607 }
2608 
2609 /*
2610  * Make this setguid thing safe, if at all possible.
2611  */
2612 void
2613 fdsetugidsafety(struct thread *td)
2614 {
2615 	struct filedesc *fdp;
2616 	struct file *fp;
2617 	int i;
2618 
2619 	fdp = td->td_proc->p_fd;
2620 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2621 	    ("the fdtable should not be shared"));
2622 	MPASS(fdp->fd_nfiles >= 3);
2623 	for (i = 0; i <= 2; i++) {
2624 		fp = fdp->fd_ofiles[i].fde_file;
2625 		if (fp != NULL && is_unsafe(fp)) {
2626 			FILEDESC_XLOCK(fdp);
2627 			knote_fdclose(td, i);
2628 			/*
2629 			 * NULL-out descriptor prior to close to avoid
2630 			 * a race while close blocks.
2631 			 */
2632 			fdfree(fdp, i);
2633 			FILEDESC_XUNLOCK(fdp);
2634 			(void) closef(fp, td);
2635 		}
2636 	}
2637 }
2638 
2639 /*
2640  * If a specific file object occupies a specific file descriptor, close the
2641  * file descriptor entry and drop a reference on the file object.  This is a
2642  * convenience function to handle a subsequent error in a function that calls
2643  * falloc() that handles the race that another thread might have closed the
2644  * file descriptor out from under the thread creating the file object.
2645  */
2646 void
2647 fdclose(struct thread *td, struct file *fp, int idx)
2648 {
2649 	struct filedesc *fdp = td->td_proc->p_fd;
2650 
2651 	FILEDESC_XLOCK(fdp);
2652 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2653 		fdfree(fdp, idx);
2654 		FILEDESC_XUNLOCK(fdp);
2655 		fdrop(fp, td);
2656 	} else
2657 		FILEDESC_XUNLOCK(fdp);
2658 }
2659 
2660 /*
2661  * Close any files on exec?
2662  */
2663 void
2664 fdcloseexec(struct thread *td)
2665 {
2666 	struct filedesc *fdp;
2667 	struct filedescent *fde;
2668 	struct file *fp;
2669 	int i, lastfile;
2670 
2671 	fdp = td->td_proc->p_fd;
2672 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2673 	    ("the fdtable should not be shared"));
2674 	lastfile = fdlastfile_single(fdp);
2675 	for (i = 0; i <= lastfile; i++) {
2676 		fde = &fdp->fd_ofiles[i];
2677 		fp = fde->fde_file;
2678 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2679 		    (fde->fde_flags & UF_EXCLOSE))) {
2680 			FILEDESC_XLOCK(fdp);
2681 			fdfree(fdp, i);
2682 			(void) closefp(fdp, i, fp, td, false, false);
2683 			FILEDESC_UNLOCK_ASSERT(fdp);
2684 		}
2685 	}
2686 }
2687 
2688 /*
2689  * It is unsafe for set[ug]id processes to be started with file
2690  * descriptors 0..2 closed, as these descriptors are given implicit
2691  * significance in the Standard C library.  fdcheckstd() will create a
2692  * descriptor referencing /dev/null for each of stdin, stdout, and
2693  * stderr that is not already open.
2694  */
2695 int
2696 fdcheckstd(struct thread *td)
2697 {
2698 	struct filedesc *fdp;
2699 	register_t save;
2700 	int i, error, devnull;
2701 
2702 	fdp = td->td_proc->p_fd;
2703 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2704 	    ("the fdtable should not be shared"));
2705 	MPASS(fdp->fd_nfiles >= 3);
2706 	devnull = -1;
2707 	for (i = 0; i <= 2; i++) {
2708 		if (fdp->fd_ofiles[i].fde_file != NULL)
2709 			continue;
2710 
2711 		save = td->td_retval[0];
2712 		if (devnull != -1) {
2713 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2714 		} else {
2715 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2716 			    UIO_SYSSPACE, O_RDWR, 0);
2717 			if (error == 0) {
2718 				devnull = td->td_retval[0];
2719 				KASSERT(devnull == i, ("we didn't get our fd"));
2720 			}
2721 		}
2722 		td->td_retval[0] = save;
2723 		if (error != 0)
2724 			return (error);
2725 	}
2726 	return (0);
2727 }
2728 
2729 /*
2730  * Internal form of close.  Decrement reference count on file structure.
2731  * Note: td may be NULL when closing a file that was being passed in a
2732  * message.
2733  */
2734 int
2735 closef(struct file *fp, struct thread *td)
2736 {
2737 	struct vnode *vp;
2738 	struct flock lf;
2739 	struct filedesc_to_leader *fdtol;
2740 	struct filedesc *fdp;
2741 
2742 	MPASS(td != NULL);
2743 
2744 	/*
2745 	 * POSIX record locking dictates that any close releases ALL
2746 	 * locks owned by this process.  This is handled by setting
2747 	 * a flag in the unlock to free ONLY locks obeying POSIX
2748 	 * semantics, and not to free BSD-style file locks.
2749 	 * If the descriptor was in a message, POSIX-style locks
2750 	 * aren't passed with the descriptor, and the thread pointer
2751 	 * will be NULL.  Callers should be careful only to pass a
2752 	 * NULL thread pointer when there really is no owning
2753 	 * context that might have locks, or the locks will be
2754 	 * leaked.
2755 	 */
2756 	if (fp->f_type == DTYPE_VNODE) {
2757 		vp = fp->f_vnode;
2758 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2759 			lf.l_whence = SEEK_SET;
2760 			lf.l_start = 0;
2761 			lf.l_len = 0;
2762 			lf.l_type = F_UNLCK;
2763 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2764 			    F_UNLCK, &lf, F_POSIX);
2765 		}
2766 		fdtol = td->td_proc->p_fdtol;
2767 		if (fdtol != NULL) {
2768 			/*
2769 			 * Handle special case where file descriptor table is
2770 			 * shared between multiple process leaders.
2771 			 */
2772 			fdp = td->td_proc->p_fd;
2773 			FILEDESC_XLOCK(fdp);
2774 			for (fdtol = fdtol->fdl_next;
2775 			    fdtol != td->td_proc->p_fdtol;
2776 			    fdtol = fdtol->fdl_next) {
2777 				if ((fdtol->fdl_leader->p_flag &
2778 				    P_ADVLOCK) == 0)
2779 					continue;
2780 				fdtol->fdl_holdcount++;
2781 				FILEDESC_XUNLOCK(fdp);
2782 				lf.l_whence = SEEK_SET;
2783 				lf.l_start = 0;
2784 				lf.l_len = 0;
2785 				lf.l_type = F_UNLCK;
2786 				vp = fp->f_vnode;
2787 				(void) VOP_ADVLOCK(vp,
2788 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2789 				    F_POSIX);
2790 				FILEDESC_XLOCK(fdp);
2791 				fdtol->fdl_holdcount--;
2792 				if (fdtol->fdl_holdcount == 0 &&
2793 				    fdtol->fdl_wakeup != 0) {
2794 					fdtol->fdl_wakeup = 0;
2795 					wakeup(fdtol);
2796 				}
2797 			}
2798 			FILEDESC_XUNLOCK(fdp);
2799 		}
2800 	}
2801 	return (fdrop_close(fp, td));
2802 }
2803 
2804 /*
2805  * Hack for file descriptor passing code.
2806  */
2807 void
2808 closef_nothread(struct file *fp)
2809 {
2810 
2811 	fdrop(fp, NULL);
2812 }
2813 
2814 /*
2815  * Initialize the file pointer with the specified properties.
2816  *
2817  * The ops are set with release semantics to be certain that the flags, type,
2818  * and data are visible when ops is.  This is to prevent ops methods from being
2819  * called with bad data.
2820  */
2821 void
2822 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2823 {
2824 	fp->f_data = data;
2825 	fp->f_flag = flag;
2826 	fp->f_type = type;
2827 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2828 }
2829 
2830 void
2831 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2832 {
2833 	fp->f_seqcount[UIO_READ] = 1;
2834 	fp->f_seqcount[UIO_WRITE] = 1;
2835 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2836 	    data, ops);
2837 }
2838 
2839 int
2840 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2841     struct file **fpp, struct filecaps *havecapsp)
2842 {
2843 	struct filedescent *fde;
2844 	int error;
2845 
2846 	FILEDESC_LOCK_ASSERT(fdp);
2847 
2848 	*fpp = NULL;
2849 	fde = fdeget_locked(fdp, fd);
2850 	if (fde == NULL) {
2851 		error = EBADF;
2852 		goto out;
2853 	}
2854 
2855 #ifdef CAPABILITIES
2856 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2857 	if (error != 0)
2858 		goto out;
2859 #endif
2860 
2861 	if (havecapsp != NULL)
2862 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2863 
2864 	*fpp = fde->fde_file;
2865 
2866 	error = 0;
2867 out:
2868 	return (error);
2869 }
2870 
2871 int
2872 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2873     struct file **fpp, struct filecaps *havecapsp)
2874 {
2875 	struct filedesc *fdp = td->td_proc->p_fd;
2876 	int error;
2877 #ifndef CAPABILITIES
2878 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2879 	if (havecapsp != NULL && error == 0)
2880 		filecaps_fill(havecapsp);
2881 #else
2882 	struct file *fp;
2883 	seqc_t seq;
2884 
2885 	*fpp = NULL;
2886 	for (;;) {
2887 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2888 		if (error != 0)
2889 			return (error);
2890 
2891 		if (havecapsp != NULL) {
2892 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2893 			    havecapsp, false)) {
2894 				fdrop(fp, td);
2895 				goto get_locked;
2896 			}
2897 		}
2898 
2899 		if (!fd_modified(fdp, fd, seq))
2900 			break;
2901 		fdrop(fp, td);
2902 	}
2903 
2904 	*fpp = fp;
2905 	return (0);
2906 
2907 get_locked:
2908 	FILEDESC_SLOCK(fdp);
2909 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2910 	if (error == 0 && !fhold(*fpp))
2911 		error = EBADF;
2912 	FILEDESC_SUNLOCK(fdp);
2913 #endif
2914 	return (error);
2915 }
2916 
2917 #ifdef CAPABILITIES
2918 int
2919 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2920 {
2921 	const struct filedescent *fde;
2922 	const struct fdescenttbl *fdt;
2923 	struct filedesc *fdp;
2924 	struct file *fp;
2925 	struct vnode *vp;
2926 	const cap_rights_t *haverights;
2927 	cap_rights_t rights;
2928 	seqc_t seq;
2929 
2930 	VFS_SMR_ASSERT_ENTERED();
2931 
2932 	rights = *ndp->ni_rightsneeded;
2933 	cap_rights_set_one(&rights, CAP_LOOKUP);
2934 
2935 	fdp = curproc->p_fd;
2936 	fdt = fdp->fd_files;
2937 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2938 		return (EBADF);
2939 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2940 	fde = &fdt->fdt_ofiles[fd];
2941 	haverights = cap_rights_fde_inline(fde);
2942 	fp = fde->fde_file;
2943 	if (__predict_false(fp == NULL))
2944 		return (EAGAIN);
2945 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2946 		return (EAGAIN);
2947 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2948 	vp = fp->f_vnode;
2949 	if (__predict_false(vp == NULL)) {
2950 		return (EAGAIN);
2951 	}
2952 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2953 		return (EAGAIN);
2954 	}
2955 	/*
2956 	 * Use an acquire barrier to force re-reading of fdt so it is
2957 	 * refreshed for verification.
2958 	 */
2959 	atomic_thread_fence_acq();
2960 	fdt = fdp->fd_files;
2961 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2962 		return (EAGAIN);
2963 	/*
2964 	 * If file descriptor doesn't have all rights,
2965 	 * all lookups relative to it must also be
2966 	 * strictly relative.
2967 	 *
2968 	 * Not yet supported by fast path.
2969 	 */
2970 	CAP_ALL(&rights);
2971 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2972 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2973 	    ndp->ni_filecaps.fc_nioctls != -1) {
2974 #ifdef notyet
2975 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2976 #else
2977 		return (EAGAIN);
2978 #endif
2979 	}
2980 	*vpp = vp;
2981 	return (0);
2982 }
2983 #else
2984 int
2985 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2986 {
2987 	const struct fdescenttbl *fdt;
2988 	struct filedesc *fdp;
2989 	struct file *fp;
2990 	struct vnode *vp;
2991 
2992 	VFS_SMR_ASSERT_ENTERED();
2993 
2994 	fdp = curproc->p_fd;
2995 	fdt = fdp->fd_files;
2996 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2997 		return (EBADF);
2998 	fp = fdt->fdt_ofiles[fd].fde_file;
2999 	if (__predict_false(fp == NULL))
3000 		return (EAGAIN);
3001 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
3002 	vp = fp->f_vnode;
3003 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3004 		return (EAGAIN);
3005 	}
3006 	/*
3007 	 * Use an acquire barrier to force re-reading of fdt so it is
3008 	 * refreshed for verification.
3009 	 */
3010 	atomic_thread_fence_acq();
3011 	fdt = fdp->fd_files;
3012 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3013 		return (EAGAIN);
3014 	filecaps_fill(&ndp->ni_filecaps);
3015 	*vpp = vp;
3016 	return (0);
3017 }
3018 #endif
3019 
3020 static int
3021 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3022     struct file **fpp, seqc_t *seqp)
3023 {
3024 #ifdef CAPABILITIES
3025 	const struct filedescent *fde;
3026 #endif
3027 	const struct fdescenttbl *fdt;
3028 	struct file *fp;
3029 #ifdef CAPABILITIES
3030 	seqc_t seq;
3031 	cap_rights_t haverights;
3032 	int error;
3033 #endif
3034 
3035 	fdt = fdp->fd_files;
3036 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3037 		return (EBADF);
3038 	/*
3039 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
3040 	 * never raising a refcount above 0.  To accomplish this we have
3041 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3042 	 * must be re-verified once we acquire a reference to be certain
3043 	 * that the identity is still correct and we did not lose a race
3044 	 * due to preemption.
3045 	 */
3046 	for (;;) {
3047 #ifdef CAPABILITIES
3048 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3049 		fde = &fdt->fdt_ofiles[fd];
3050 		haverights = *cap_rights_fde_inline(fde);
3051 		fp = fde->fde_file;
3052 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3053 			continue;
3054 #else
3055 		fp = fdt->fdt_ofiles[fd].fde_file;
3056 #endif
3057 		if (fp == NULL)
3058 			return (EBADF);
3059 #ifdef CAPABILITIES
3060 		error = cap_check_inline(&haverights, needrightsp);
3061 		if (error != 0)
3062 			return (error);
3063 #endif
3064 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3065 			/*
3066 			 * Force a reload. Other thread could reallocate the
3067 			 * table before this fd was closed, so it is possible
3068 			 * that there is a stale fp pointer in cached version.
3069 			 */
3070 			fdt = atomic_load_ptr(&fdp->fd_files);
3071 			continue;
3072 		}
3073 		/*
3074 		 * Use an acquire barrier to force re-reading of fdt so it is
3075 		 * refreshed for verification.
3076 		 */
3077 		atomic_thread_fence_acq();
3078 		fdt = fdp->fd_files;
3079 #ifdef	CAPABILITIES
3080 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3081 #else
3082 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3083 #endif
3084 			break;
3085 		fdrop(fp, curthread);
3086 	}
3087 	*fpp = fp;
3088 	if (seqp != NULL) {
3089 #ifdef CAPABILITIES
3090 		*seqp = seq;
3091 #endif
3092 	}
3093 	return (0);
3094 }
3095 
3096 /*
3097  * See the comments in fget_unlocked_seq for an explanation of how this works.
3098  *
3099  * This is a simplified variant which bails out to the aforementioned routine
3100  * if anything goes wrong. In practice this only happens when userspace is
3101  * racing with itself.
3102  */
3103 int
3104 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3105     struct file **fpp)
3106 {
3107 #ifdef CAPABILITIES
3108 	const struct filedescent *fde;
3109 #endif
3110 	const struct fdescenttbl *fdt;
3111 	struct file *fp;
3112 #ifdef CAPABILITIES
3113 	seqc_t seq;
3114 	const cap_rights_t *haverights;
3115 #endif
3116 
3117 	fdt = fdp->fd_files;
3118 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3119 		*fpp = NULL;
3120 		return (EBADF);
3121 	}
3122 #ifdef CAPABILITIES
3123 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3124 	fde = &fdt->fdt_ofiles[fd];
3125 	haverights = cap_rights_fde_inline(fde);
3126 	fp = fde->fde_file;
3127 #else
3128 	fp = fdt->fdt_ofiles[fd].fde_file;
3129 #endif
3130 	if (__predict_false(fp == NULL))
3131 		goto out_fallback;
3132 #ifdef CAPABILITIES
3133 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3134 		goto out_fallback;
3135 #endif
3136 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3137 		goto out_fallback;
3138 
3139 	/*
3140 	 * Use an acquire barrier to force re-reading of fdt so it is
3141 	 * refreshed for verification.
3142 	 */
3143 	atomic_thread_fence_acq();
3144 	fdt = fdp->fd_files;
3145 #ifdef	CAPABILITIES
3146 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3147 #else
3148 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3149 #endif
3150 		goto out_fdrop;
3151 	*fpp = fp;
3152 	return (0);
3153 out_fdrop:
3154 	fdrop(fp, curthread);
3155 out_fallback:
3156 	*fpp = NULL;
3157 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3158 }
3159 
3160 /*
3161  * Translate fd -> file when the caller guarantees the file descriptor table
3162  * can't be changed by others.
3163  *
3164  * Note this does not mean the file object itself is only visible to the caller,
3165  * merely that it wont disappear without having to be referenced.
3166  *
3167  * Must be paired with fput_only_user.
3168  */
3169 #ifdef	CAPABILITIES
3170 int
3171 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3172     struct file **fpp)
3173 {
3174 	const struct filedescent *fde;
3175 	const struct fdescenttbl *fdt;
3176 	const cap_rights_t *haverights;
3177 	struct file *fp;
3178 	int error;
3179 
3180 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3181 
3182 	*fpp = NULL;
3183 	if (__predict_false(fd >= fdp->fd_nfiles))
3184 		return (EBADF);
3185 
3186 	fdt = fdp->fd_files;
3187 	fde = &fdt->fdt_ofiles[fd];
3188 	fp = fde->fde_file;
3189 	if (__predict_false(fp == NULL))
3190 		return (EBADF);
3191 	MPASS(refcount_load(&fp->f_count) > 0);
3192 	haverights = cap_rights_fde_inline(fde);
3193 	error = cap_check_inline(haverights, needrightsp);
3194 	if (__predict_false(error != 0))
3195 		return (error);
3196 	*fpp = fp;
3197 	return (0);
3198 }
3199 #else
3200 int
3201 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3202     struct file **fpp)
3203 {
3204 	struct file *fp;
3205 
3206 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3207 
3208 	*fpp = NULL;
3209 	if (__predict_false(fd >= fdp->fd_nfiles))
3210 		return (EBADF);
3211 
3212 	fp = fdp->fd_ofiles[fd].fde_file;
3213 	if (__predict_false(fp == NULL))
3214 		return (EBADF);
3215 
3216 	MPASS(refcount_load(&fp->f_count) > 0);
3217 	*fpp = fp;
3218 	return (0);
3219 }
3220 #endif
3221 
3222 /*
3223  * Extract the file pointer associated with the specified descriptor for the
3224  * current user process.
3225  *
3226  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3227  * returned.
3228  *
3229  * File's rights will be checked against the capability rights mask.
3230  *
3231  * If an error occurred the non-zero error is returned and *fpp is set to
3232  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3233  * responsible for fdrop().
3234  */
3235 static __inline int
3236 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3237     cap_rights_t *needrightsp)
3238 {
3239 	struct filedesc *fdp;
3240 	struct file *fp;
3241 	int error;
3242 
3243 	*fpp = NULL;
3244 	fdp = td->td_proc->p_fd;
3245 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3246 	if (__predict_false(error != 0))
3247 		return (error);
3248 	if (__predict_false(fp->f_ops == &badfileops)) {
3249 		fdrop(fp, td);
3250 		return (EBADF);
3251 	}
3252 
3253 	/*
3254 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3255 	 */
3256 	error = 0;
3257 	switch (flags) {
3258 	case FREAD:
3259 	case FWRITE:
3260 		if ((fp->f_flag & flags) == 0)
3261 			error = EBADF;
3262 		break;
3263 	case FEXEC:
3264 		if (fp->f_ops != &path_fileops &&
3265 		    ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3266 		    (fp->f_flag & FWRITE) != 0))
3267 			error = EBADF;
3268 		break;
3269 	case 0:
3270 		break;
3271 	default:
3272 		KASSERT(0, ("wrong flags"));
3273 	}
3274 
3275 	if (error != 0) {
3276 		fdrop(fp, td);
3277 		return (error);
3278 	}
3279 
3280 	*fpp = fp;
3281 	return (0);
3282 }
3283 
3284 int
3285 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3286 {
3287 
3288 	return (_fget(td, fd, fpp, 0, rightsp));
3289 }
3290 
3291 int
3292 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3293     struct file **fpp)
3294 {
3295 	int error;
3296 #ifndef CAPABILITIES
3297 	error = _fget(td, fd, fpp, 0, rightsp);
3298 	if (maxprotp != NULL)
3299 		*maxprotp = VM_PROT_ALL;
3300 	return (error);
3301 #else
3302 	cap_rights_t fdrights;
3303 	struct filedesc *fdp;
3304 	struct file *fp;
3305 	seqc_t seq;
3306 
3307 	*fpp = NULL;
3308 	fdp = td->td_proc->p_fd;
3309 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3310 	for (;;) {
3311 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3312 		if (__predict_false(error != 0))
3313 			return (error);
3314 		if (__predict_false(fp->f_ops == &badfileops)) {
3315 			fdrop(fp, td);
3316 			return (EBADF);
3317 		}
3318 		if (maxprotp != NULL)
3319 			fdrights = *cap_rights(fdp, fd);
3320 		if (!fd_modified(fdp, fd, seq))
3321 			break;
3322 		fdrop(fp, td);
3323 	}
3324 
3325 	/*
3326 	 * If requested, convert capability rights to access flags.
3327 	 */
3328 	if (maxprotp != NULL)
3329 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3330 	*fpp = fp;
3331 	return (0);
3332 #endif
3333 }
3334 
3335 int
3336 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3337 {
3338 
3339 	return (_fget(td, fd, fpp, FREAD, rightsp));
3340 }
3341 
3342 int
3343 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3344 {
3345 
3346 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3347 }
3348 
3349 int
3350 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3351     struct file **fpp)
3352 {
3353 	struct filedesc *fdp = td->td_proc->p_fd;
3354 #ifndef CAPABILITIES
3355 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3356 #else
3357 	struct file *fp;
3358 	int error;
3359 	seqc_t seq;
3360 
3361 	*fpp = NULL;
3362 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3363 	for (;;) {
3364 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3365 		if (error != 0)
3366 			return (error);
3367 		error = cap_fcntl_check(fdp, fd, needfcntl);
3368 		if (!fd_modified(fdp, fd, seq))
3369 			break;
3370 		fdrop(fp, td);
3371 	}
3372 	if (error != 0) {
3373 		fdrop(fp, td);
3374 		return (error);
3375 	}
3376 	*fpp = fp;
3377 	return (0);
3378 #endif
3379 }
3380 
3381 /*
3382  * Like fget() but loads the underlying vnode, or returns an error if the
3383  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3384  * never have VM objects.  The returned vnode will be vref()'d.
3385  *
3386  * XXX: what about the unused flags ?
3387  */
3388 static __inline int
3389 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3390     struct vnode **vpp)
3391 {
3392 	struct file *fp;
3393 	int error;
3394 
3395 	*vpp = NULL;
3396 	error = _fget(td, fd, &fp, flags, needrightsp);
3397 	if (error != 0)
3398 		return (error);
3399 	if (fp->f_vnode == NULL) {
3400 		error = EINVAL;
3401 	} else {
3402 		*vpp = fp->f_vnode;
3403 		vref(*vpp);
3404 	}
3405 	fdrop(fp, td);
3406 
3407 	return (error);
3408 }
3409 
3410 int
3411 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3412 {
3413 
3414 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3415 }
3416 
3417 int
3418 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3419     struct filecaps *havecaps, struct vnode **vpp)
3420 {
3421 	struct filecaps caps;
3422 	struct file *fp;
3423 	int error;
3424 
3425 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3426 	if (error != 0)
3427 		return (error);
3428 	if (fp->f_ops == &badfileops) {
3429 		error = EBADF;
3430 		goto out;
3431 	}
3432 	if (fp->f_vnode == NULL) {
3433 		error = EINVAL;
3434 		goto out;
3435 	}
3436 
3437 	*havecaps = caps;
3438 	*vpp = fp->f_vnode;
3439 	vref(*vpp);
3440 	fdrop(fp, td);
3441 
3442 	return (0);
3443 out:
3444 	filecaps_free(&caps);
3445 	fdrop(fp, td);
3446 	return (error);
3447 }
3448 
3449 int
3450 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3451 {
3452 
3453 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3454 }
3455 
3456 int
3457 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3458 {
3459 
3460 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3461 }
3462 
3463 #ifdef notyet
3464 int
3465 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3466     struct vnode **vpp)
3467 {
3468 
3469 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3470 }
3471 #endif
3472 
3473 /*
3474  * Handle the last reference to a file being closed.
3475  *
3476  * Without the noinline attribute clang keeps inlining the func thorough this
3477  * file when fdrop is used.
3478  */
3479 int __noinline
3480 _fdrop(struct file *fp, struct thread *td)
3481 {
3482 	int error;
3483 #ifdef INVARIANTS
3484 	int count;
3485 
3486 	count = refcount_load(&fp->f_count);
3487 	if (count != 0)
3488 		panic("fdrop: fp %p count %d", fp, count);
3489 #endif
3490 	error = fo_close(fp, td);
3491 	atomic_subtract_int(&openfiles, 1);
3492 	crfree(fp->f_cred);
3493 	free(fp->f_advice, M_FADVISE);
3494 	uma_zfree(file_zone, fp);
3495 
3496 	return (error);
3497 }
3498 
3499 /*
3500  * Apply an advisory lock on a file descriptor.
3501  *
3502  * Just attempt to get a record lock of the requested type on the entire file
3503  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3504  */
3505 #ifndef _SYS_SYSPROTO_H_
3506 struct flock_args {
3507 	int	fd;
3508 	int	how;
3509 };
3510 #endif
3511 /* ARGSUSED */
3512 int
3513 sys_flock(struct thread *td, struct flock_args *uap)
3514 {
3515 	struct file *fp;
3516 	struct vnode *vp;
3517 	struct flock lf;
3518 	int error;
3519 
3520 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3521 	if (error != 0)
3522 		return (error);
3523 	error = EOPNOTSUPP;
3524 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3525 		goto done;
3526 	}
3527 	if (fp->f_ops == &path_fileops) {
3528 		goto done;
3529 	}
3530 
3531 	error = 0;
3532 	vp = fp->f_vnode;
3533 	lf.l_whence = SEEK_SET;
3534 	lf.l_start = 0;
3535 	lf.l_len = 0;
3536 	if (uap->how & LOCK_UN) {
3537 		lf.l_type = F_UNLCK;
3538 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3539 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3540 		goto done;
3541 	}
3542 	if (uap->how & LOCK_EX)
3543 		lf.l_type = F_WRLCK;
3544 	else if (uap->how & LOCK_SH)
3545 		lf.l_type = F_RDLCK;
3546 	else {
3547 		error = EBADF;
3548 		goto done;
3549 	}
3550 	atomic_set_int(&fp->f_flag, FHASLOCK);
3551 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3552 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3553 done:
3554 	fdrop(fp, td);
3555 	return (error);
3556 }
3557 /*
3558  * Duplicate the specified descriptor to a free descriptor.
3559  */
3560 int
3561 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3562     int openerror, int *indxp)
3563 {
3564 	struct filedescent *newfde, *oldfde;
3565 	struct file *fp;
3566 	u_long *ioctls;
3567 	int error, indx;
3568 
3569 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3570 	    ("unexpected error %d in %s", openerror, __func__));
3571 
3572 	/*
3573 	 * If the to-be-dup'd fd number is greater than the allowed number
3574 	 * of file descriptors, or the fd to be dup'd has already been
3575 	 * closed, then reject.
3576 	 */
3577 	FILEDESC_XLOCK(fdp);
3578 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3579 		FILEDESC_XUNLOCK(fdp);
3580 		return (EBADF);
3581 	}
3582 
3583 	error = fdalloc(td, 0, &indx);
3584 	if (error != 0) {
3585 		FILEDESC_XUNLOCK(fdp);
3586 		return (error);
3587 	}
3588 
3589 	/*
3590 	 * There are two cases of interest here.
3591 	 *
3592 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3593 	 *
3594 	 * For ENXIO steal away the file structure from (dfd) and store it in
3595 	 * (indx).  (dfd) is effectively closed by this operation.
3596 	 */
3597 	switch (openerror) {
3598 	case ENODEV:
3599 		/*
3600 		 * Check that the mode the file is being opened for is a
3601 		 * subset of the mode of the existing descriptor.
3602 		 */
3603 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3604 			fdunused(fdp, indx);
3605 			FILEDESC_XUNLOCK(fdp);
3606 			return (EACCES);
3607 		}
3608 		if (!fhold(fp)) {
3609 			fdunused(fdp, indx);
3610 			FILEDESC_XUNLOCK(fdp);
3611 			return (EBADF);
3612 		}
3613 		newfde = &fdp->fd_ofiles[indx];
3614 		oldfde = &fdp->fd_ofiles[dfd];
3615 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3616 #ifdef CAPABILITIES
3617 		seqc_write_begin(&newfde->fde_seqc);
3618 #endif
3619 		memcpy(newfde, oldfde, fde_change_size);
3620 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3621 		    ioctls);
3622 #ifdef CAPABILITIES
3623 		seqc_write_end(&newfde->fde_seqc);
3624 #endif
3625 		break;
3626 	case ENXIO:
3627 		/*
3628 		 * Steal away the file pointer from dfd and stuff it into indx.
3629 		 */
3630 		newfde = &fdp->fd_ofiles[indx];
3631 		oldfde = &fdp->fd_ofiles[dfd];
3632 #ifdef CAPABILITIES
3633 		seqc_write_begin(&newfde->fde_seqc);
3634 #endif
3635 		memcpy(newfde, oldfde, fde_change_size);
3636 		oldfde->fde_file = NULL;
3637 		fdunused(fdp, dfd);
3638 #ifdef CAPABILITIES
3639 		seqc_write_end(&newfde->fde_seqc);
3640 #endif
3641 		break;
3642 	}
3643 	FILEDESC_XUNLOCK(fdp);
3644 	*indxp = indx;
3645 	return (0);
3646 }
3647 
3648 /*
3649  * This sysctl determines if we will allow a process to chroot(2) if it
3650  * has a directory open:
3651  *	0: disallowed for all processes.
3652  *	1: allowed for processes that were not already chroot(2)'ed.
3653  *	2: allowed for all processes.
3654  */
3655 
3656 static int chroot_allow_open_directories = 1;
3657 
3658 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3659     &chroot_allow_open_directories, 0,
3660     "Allow a process to chroot(2) if it has a directory open");
3661 
3662 /*
3663  * Helper function for raised chroot(2) security function:  Refuse if
3664  * any filedescriptors are open directories.
3665  */
3666 static int
3667 chroot_refuse_vdir_fds(struct filedesc *fdp)
3668 {
3669 	struct vnode *vp;
3670 	struct file *fp;
3671 	int fd, lastfile;
3672 
3673 	FILEDESC_LOCK_ASSERT(fdp);
3674 
3675 	lastfile = fdlastfile(fdp);
3676 	for (fd = 0; fd <= lastfile; fd++) {
3677 		fp = fget_locked(fdp, fd);
3678 		if (fp == NULL)
3679 			continue;
3680 		if (fp->f_type == DTYPE_VNODE) {
3681 			vp = fp->f_vnode;
3682 			if (vp->v_type == VDIR)
3683 				return (EPERM);
3684 		}
3685 	}
3686 	return (0);
3687 }
3688 
3689 static void
3690 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3691 {
3692 
3693 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3694 		vrefact(oldpwd->pwd_cdir);
3695 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3696 	}
3697 
3698 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3699 		vrefact(oldpwd->pwd_rdir);
3700 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3701 	}
3702 
3703 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3704 		vrefact(oldpwd->pwd_jdir);
3705 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3706 	}
3707 }
3708 
3709 struct pwd *
3710 pwd_hold_pwddesc(struct pwddesc *pdp)
3711 {
3712 	struct pwd *pwd;
3713 
3714 	PWDDESC_ASSERT_XLOCKED(pdp);
3715 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3716 	if (pwd != NULL)
3717 		refcount_acquire(&pwd->pwd_refcount);
3718 	return (pwd);
3719 }
3720 
3721 bool
3722 pwd_hold_smr(struct pwd *pwd)
3723 {
3724 
3725 	MPASS(pwd != NULL);
3726 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3727 		return (true);
3728 	}
3729 	return (false);
3730 }
3731 
3732 struct pwd *
3733 pwd_hold(struct thread *td)
3734 {
3735 	struct pwddesc *pdp;
3736 	struct pwd *pwd;
3737 
3738 	pdp = td->td_proc->p_pd;
3739 
3740 	vfs_smr_enter();
3741 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3742 	if (pwd_hold_smr(pwd)) {
3743 		vfs_smr_exit();
3744 		return (pwd);
3745 	}
3746 	vfs_smr_exit();
3747 	PWDDESC_XLOCK(pdp);
3748 	pwd = pwd_hold_pwddesc(pdp);
3749 	MPASS(pwd != NULL);
3750 	PWDDESC_XUNLOCK(pdp);
3751 	return (pwd);
3752 }
3753 
3754 struct pwd *
3755 pwd_hold_proc(struct proc *p)
3756 {
3757 	struct pwddesc *pdp;
3758 	struct pwd *pwd;
3759 
3760 	PROC_ASSERT_HELD(p);
3761 	PROC_LOCK(p);
3762 	pdp = pdhold(p);
3763 	MPASS(pdp != NULL);
3764 	PROC_UNLOCK(p);
3765 
3766 	PWDDESC_XLOCK(pdp);
3767 	pwd = pwd_hold_pwddesc(pdp);
3768 	MPASS(pwd != NULL);
3769 	PWDDESC_XUNLOCK(pdp);
3770 	pddrop(pdp);
3771 	return (pwd);
3772 }
3773 
3774 static struct pwd *
3775 pwd_alloc(void)
3776 {
3777 	struct pwd *pwd;
3778 
3779 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3780 	bzero(pwd, sizeof(*pwd));
3781 	refcount_init(&pwd->pwd_refcount, 1);
3782 	return (pwd);
3783 }
3784 
3785 void
3786 pwd_drop(struct pwd *pwd)
3787 {
3788 
3789 	if (!refcount_release(&pwd->pwd_refcount))
3790 		return;
3791 
3792 	if (pwd->pwd_cdir != NULL)
3793 		vrele(pwd->pwd_cdir);
3794 	if (pwd->pwd_rdir != NULL)
3795 		vrele(pwd->pwd_rdir);
3796 	if (pwd->pwd_jdir != NULL)
3797 		vrele(pwd->pwd_jdir);
3798 	uma_zfree_smr(pwd_zone, pwd);
3799 }
3800 
3801 /*
3802 * The caller is responsible for invoking priv_check() and
3803 * mac_vnode_check_chroot() to authorize this operation.
3804 */
3805 int
3806 pwd_chroot(struct thread *td, struct vnode *vp)
3807 {
3808 	struct pwddesc *pdp;
3809 	struct filedesc *fdp;
3810 	struct pwd *newpwd, *oldpwd;
3811 	int error;
3812 
3813 	fdp = td->td_proc->p_fd;
3814 	pdp = td->td_proc->p_pd;
3815 	newpwd = pwd_alloc();
3816 	FILEDESC_SLOCK(fdp);
3817 	PWDDESC_XLOCK(pdp);
3818 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3819 	if (chroot_allow_open_directories == 0 ||
3820 	    (chroot_allow_open_directories == 1 &&
3821 	    oldpwd->pwd_rdir != rootvnode)) {
3822 		error = chroot_refuse_vdir_fds(fdp);
3823 		FILEDESC_SUNLOCK(fdp);
3824 		if (error != 0) {
3825 			PWDDESC_XUNLOCK(pdp);
3826 			pwd_drop(newpwd);
3827 			return (error);
3828 		}
3829 	} else {
3830 		FILEDESC_SUNLOCK(fdp);
3831 	}
3832 
3833 	vrefact(vp);
3834 	newpwd->pwd_rdir = vp;
3835 	if (oldpwd->pwd_jdir == NULL) {
3836 		vrefact(vp);
3837 		newpwd->pwd_jdir = vp;
3838 	}
3839 	pwd_fill(oldpwd, newpwd);
3840 	pwd_set(pdp, newpwd);
3841 	PWDDESC_XUNLOCK(pdp);
3842 	pwd_drop(oldpwd);
3843 	return (0);
3844 }
3845 
3846 void
3847 pwd_chdir(struct thread *td, struct vnode *vp)
3848 {
3849 	struct pwddesc *pdp;
3850 	struct pwd *newpwd, *oldpwd;
3851 
3852 	VNPASS(vp->v_usecount > 0, vp);
3853 
3854 	newpwd = pwd_alloc();
3855 	pdp = td->td_proc->p_pd;
3856 	PWDDESC_XLOCK(pdp);
3857 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3858 	newpwd->pwd_cdir = vp;
3859 	pwd_fill(oldpwd, newpwd);
3860 	pwd_set(pdp, newpwd);
3861 	PWDDESC_XUNLOCK(pdp);
3862 	pwd_drop(oldpwd);
3863 }
3864 
3865 /*
3866  * jail_attach(2) changes both root and working directories.
3867  */
3868 int
3869 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
3870 {
3871 	struct pwddesc *pdp;
3872 	struct filedesc *fdp;
3873 	struct pwd *newpwd, *oldpwd;
3874 	int error;
3875 
3876 	fdp = td->td_proc->p_fd;
3877 	pdp = td->td_proc->p_pd;
3878 	newpwd = pwd_alloc();
3879 	FILEDESC_SLOCK(fdp);
3880 	PWDDESC_XLOCK(pdp);
3881 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3882 	error = chroot_refuse_vdir_fds(fdp);
3883 	FILEDESC_SUNLOCK(fdp);
3884 	if (error != 0) {
3885 		PWDDESC_XUNLOCK(pdp);
3886 		pwd_drop(newpwd);
3887 		return (error);
3888 	}
3889 
3890 	vrefact(vp);
3891 	newpwd->pwd_rdir = vp;
3892 	vrefact(vp);
3893 	newpwd->pwd_cdir = vp;
3894 	if (oldpwd->pwd_jdir == NULL) {
3895 		vrefact(vp);
3896 		newpwd->pwd_jdir = vp;
3897 	}
3898 	pwd_fill(oldpwd, newpwd);
3899 	pwd_set(pdp, newpwd);
3900 	PWDDESC_XUNLOCK(pdp);
3901 	pwd_drop(oldpwd);
3902 	return (0);
3903 }
3904 
3905 void
3906 pwd_ensure_dirs(void)
3907 {
3908 	struct pwddesc *pdp;
3909 	struct pwd *oldpwd, *newpwd;
3910 
3911 	pdp = curproc->p_pd;
3912 	PWDDESC_XLOCK(pdp);
3913 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3914 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3915 		PWDDESC_XUNLOCK(pdp);
3916 		return;
3917 	}
3918 	PWDDESC_XUNLOCK(pdp);
3919 
3920 	newpwd = pwd_alloc();
3921 	PWDDESC_XLOCK(pdp);
3922 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3923 	pwd_fill(oldpwd, newpwd);
3924 	if (newpwd->pwd_cdir == NULL) {
3925 		vrefact(rootvnode);
3926 		newpwd->pwd_cdir = rootvnode;
3927 	}
3928 	if (newpwd->pwd_rdir == NULL) {
3929 		vrefact(rootvnode);
3930 		newpwd->pwd_rdir = rootvnode;
3931 	}
3932 	pwd_set(pdp, newpwd);
3933 	PWDDESC_XUNLOCK(pdp);
3934 	pwd_drop(oldpwd);
3935 }
3936 
3937 void
3938 pwd_set_rootvnode(void)
3939 {
3940 	struct pwddesc *pdp;
3941 	struct pwd *oldpwd, *newpwd;
3942 
3943 	pdp = curproc->p_pd;
3944 
3945 	newpwd = pwd_alloc();
3946 	PWDDESC_XLOCK(pdp);
3947 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3948 	vrefact(rootvnode);
3949 	newpwd->pwd_cdir = rootvnode;
3950 	vrefact(rootvnode);
3951 	newpwd->pwd_rdir = rootvnode;
3952 	pwd_fill(oldpwd, newpwd);
3953 	pwd_set(pdp, newpwd);
3954 	PWDDESC_XUNLOCK(pdp);
3955 	pwd_drop(oldpwd);
3956 }
3957 
3958 /*
3959  * Scan all active processes and prisons to see if any of them have a current
3960  * or root directory of `olddp'. If so, replace them with the new mount point.
3961  */
3962 void
3963 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3964 {
3965 	struct pwddesc *pdp;
3966 	struct pwd *newpwd, *oldpwd;
3967 	struct prison *pr;
3968 	struct proc *p;
3969 	int nrele;
3970 
3971 	if (vrefcnt(olddp) == 1)
3972 		return;
3973 	nrele = 0;
3974 	newpwd = pwd_alloc();
3975 	sx_slock(&allproc_lock);
3976 	FOREACH_PROC_IN_SYSTEM(p) {
3977 		PROC_LOCK(p);
3978 		pdp = pdhold(p);
3979 		PROC_UNLOCK(p);
3980 		if (pdp == NULL)
3981 			continue;
3982 		PWDDESC_XLOCK(pdp);
3983 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3984 		if (oldpwd == NULL ||
3985 		    (oldpwd->pwd_cdir != olddp &&
3986 		    oldpwd->pwd_rdir != olddp &&
3987 		    oldpwd->pwd_jdir != olddp)) {
3988 			PWDDESC_XUNLOCK(pdp);
3989 			pddrop(pdp);
3990 			continue;
3991 		}
3992 		if (oldpwd->pwd_cdir == olddp) {
3993 			vrefact(newdp);
3994 			newpwd->pwd_cdir = newdp;
3995 		}
3996 		if (oldpwd->pwd_rdir == olddp) {
3997 			vrefact(newdp);
3998 			newpwd->pwd_rdir = newdp;
3999 		}
4000 		if (oldpwd->pwd_jdir == olddp) {
4001 			vrefact(newdp);
4002 			newpwd->pwd_jdir = newdp;
4003 		}
4004 		pwd_fill(oldpwd, newpwd);
4005 		pwd_set(pdp, newpwd);
4006 		PWDDESC_XUNLOCK(pdp);
4007 		pwd_drop(oldpwd);
4008 		pddrop(pdp);
4009 		newpwd = pwd_alloc();
4010 	}
4011 	sx_sunlock(&allproc_lock);
4012 	pwd_drop(newpwd);
4013 	if (rootvnode == olddp) {
4014 		vrefact(newdp);
4015 		rootvnode = newdp;
4016 		nrele++;
4017 	}
4018 	mtx_lock(&prison0.pr_mtx);
4019 	if (prison0.pr_root == olddp) {
4020 		vrefact(newdp);
4021 		prison0.pr_root = newdp;
4022 		nrele++;
4023 	}
4024 	mtx_unlock(&prison0.pr_mtx);
4025 	sx_slock(&allprison_lock);
4026 	TAILQ_FOREACH(pr, &allprison, pr_list) {
4027 		mtx_lock(&pr->pr_mtx);
4028 		if (pr->pr_root == olddp) {
4029 			vrefact(newdp);
4030 			pr->pr_root = newdp;
4031 			nrele++;
4032 		}
4033 		mtx_unlock(&pr->pr_mtx);
4034 	}
4035 	sx_sunlock(&allprison_lock);
4036 	while (nrele--)
4037 		vrele(olddp);
4038 }
4039 
4040 struct filedesc_to_leader *
4041 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
4042 {
4043 	struct filedesc_to_leader *fdtol;
4044 
4045 	fdtol = malloc(sizeof(struct filedesc_to_leader),
4046 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4047 	fdtol->fdl_refcount = 1;
4048 	fdtol->fdl_holdcount = 0;
4049 	fdtol->fdl_wakeup = 0;
4050 	fdtol->fdl_leader = leader;
4051 	if (old != NULL) {
4052 		FILEDESC_XLOCK(fdp);
4053 		fdtol->fdl_next = old->fdl_next;
4054 		fdtol->fdl_prev = old;
4055 		old->fdl_next = fdtol;
4056 		fdtol->fdl_next->fdl_prev = fdtol;
4057 		FILEDESC_XUNLOCK(fdp);
4058 	} else {
4059 		fdtol->fdl_next = fdtol;
4060 		fdtol->fdl_prev = fdtol;
4061 	}
4062 	return (fdtol);
4063 }
4064 
4065 static int
4066 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4067 {
4068 	NDSLOTTYPE *map;
4069 	struct filedesc *fdp;
4070 	u_int namelen;
4071 	int count, off, minoff;
4072 
4073 	namelen = arg2;
4074 	if (namelen != 1)
4075 		return (EINVAL);
4076 
4077 	if (*(int *)arg1 != 0)
4078 		return (EINVAL);
4079 
4080 	fdp = curproc->p_fd;
4081 	count = 0;
4082 	FILEDESC_SLOCK(fdp);
4083 	map = fdp->fd_map;
4084 	off = NDSLOT(fdp->fd_nfiles - 1);
4085 	for (minoff = NDSLOT(0); off >= minoff; --off)
4086 		count += bitcountl(map[off]);
4087 	FILEDESC_SUNLOCK(fdp);
4088 
4089 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4090 }
4091 
4092 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4093     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4094     "Number of open file descriptors");
4095 
4096 /*
4097  * Get file structures globally.
4098  */
4099 static int
4100 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4101 {
4102 	struct xfile xf;
4103 	struct filedesc *fdp;
4104 	struct file *fp;
4105 	struct proc *p;
4106 	int error, n, lastfile;
4107 
4108 	error = sysctl_wire_old_buffer(req, 0);
4109 	if (error != 0)
4110 		return (error);
4111 	if (req->oldptr == NULL) {
4112 		n = 0;
4113 		sx_slock(&allproc_lock);
4114 		FOREACH_PROC_IN_SYSTEM(p) {
4115 			PROC_LOCK(p);
4116 			if (p->p_state == PRS_NEW) {
4117 				PROC_UNLOCK(p);
4118 				continue;
4119 			}
4120 			fdp = fdhold(p);
4121 			PROC_UNLOCK(p);
4122 			if (fdp == NULL)
4123 				continue;
4124 			/* overestimates sparse tables. */
4125 			n += fdp->fd_nfiles;
4126 			fddrop(fdp);
4127 		}
4128 		sx_sunlock(&allproc_lock);
4129 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4130 	}
4131 	error = 0;
4132 	bzero(&xf, sizeof(xf));
4133 	xf.xf_size = sizeof(xf);
4134 	sx_slock(&allproc_lock);
4135 	FOREACH_PROC_IN_SYSTEM(p) {
4136 		PROC_LOCK(p);
4137 		if (p->p_state == PRS_NEW) {
4138 			PROC_UNLOCK(p);
4139 			continue;
4140 		}
4141 		if (p_cansee(req->td, p) != 0) {
4142 			PROC_UNLOCK(p);
4143 			continue;
4144 		}
4145 		xf.xf_pid = p->p_pid;
4146 		xf.xf_uid = p->p_ucred->cr_uid;
4147 		fdp = fdhold(p);
4148 		PROC_UNLOCK(p);
4149 		if (fdp == NULL)
4150 			continue;
4151 		FILEDESC_SLOCK(fdp);
4152 		lastfile = fdlastfile(fdp);
4153 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4154 		    n++) {
4155 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4156 				continue;
4157 			xf.xf_fd = n;
4158 			xf.xf_file = (uintptr_t)fp;
4159 			xf.xf_data = (uintptr_t)fp->f_data;
4160 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4161 			xf.xf_type = (uintptr_t)fp->f_type;
4162 			xf.xf_count = refcount_load(&fp->f_count);
4163 			xf.xf_msgcount = 0;
4164 			xf.xf_offset = foffset_get(fp);
4165 			xf.xf_flag = fp->f_flag;
4166 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4167 			if (error)
4168 				break;
4169 		}
4170 		FILEDESC_SUNLOCK(fdp);
4171 		fddrop(fdp);
4172 		if (error)
4173 			break;
4174 	}
4175 	sx_sunlock(&allproc_lock);
4176 	return (error);
4177 }
4178 
4179 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4180     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4181 
4182 #ifdef KINFO_FILE_SIZE
4183 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4184 #endif
4185 
4186 static int
4187 xlate_fflags(int fflags)
4188 {
4189 	static const struct {
4190 		int	fflag;
4191 		int	kf_fflag;
4192 	} fflags_table[] = {
4193 		{ FAPPEND, KF_FLAG_APPEND },
4194 		{ FASYNC, KF_FLAG_ASYNC },
4195 		{ FFSYNC, KF_FLAG_FSYNC },
4196 		{ FHASLOCK, KF_FLAG_HASLOCK },
4197 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4198 		{ FREAD, KF_FLAG_READ },
4199 		{ FWRITE, KF_FLAG_WRITE },
4200 		{ O_CREAT, KF_FLAG_CREAT },
4201 		{ O_DIRECT, KF_FLAG_DIRECT },
4202 		{ O_EXCL, KF_FLAG_EXCL },
4203 		{ O_EXEC, KF_FLAG_EXEC },
4204 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4205 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4206 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4207 		{ O_TRUNC, KF_FLAG_TRUNC }
4208 	};
4209 	unsigned int i;
4210 	int kflags;
4211 
4212 	kflags = 0;
4213 	for (i = 0; i < nitems(fflags_table); i++)
4214 		if (fflags & fflags_table[i].fflag)
4215 			kflags |=  fflags_table[i].kf_fflag;
4216 	return (kflags);
4217 }
4218 
4219 /* Trim unused data from kf_path by truncating the structure size. */
4220 void
4221 pack_kinfo(struct kinfo_file *kif)
4222 {
4223 
4224 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4225 	    strlen(kif->kf_path) + 1;
4226 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4227 }
4228 
4229 static void
4230 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4231     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4232 {
4233 	int error;
4234 
4235 	bzero(kif, sizeof(*kif));
4236 
4237 	/* Set a default type to allow for empty fill_kinfo() methods. */
4238 	kif->kf_type = KF_TYPE_UNKNOWN;
4239 	kif->kf_flags = xlate_fflags(fp->f_flag);
4240 	if (rightsp != NULL)
4241 		kif->kf_cap_rights = *rightsp;
4242 	else
4243 		cap_rights_init_zero(&kif->kf_cap_rights);
4244 	kif->kf_fd = fd;
4245 	kif->kf_ref_count = refcount_load(&fp->f_count);
4246 	kif->kf_offset = foffset_get(fp);
4247 
4248 	/*
4249 	 * This may drop the filedesc lock, so the 'fp' cannot be
4250 	 * accessed after this call.
4251 	 */
4252 	error = fo_fill_kinfo(fp, kif, fdp);
4253 	if (error == 0)
4254 		kif->kf_status |= KF_ATTR_VALID;
4255 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4256 		pack_kinfo(kif);
4257 	else
4258 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4259 }
4260 
4261 static void
4262 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4263     struct kinfo_file *kif, int flags)
4264 {
4265 	int error;
4266 
4267 	bzero(kif, sizeof(*kif));
4268 
4269 	kif->kf_type = KF_TYPE_VNODE;
4270 	error = vn_fill_kinfo_vnode(vp, kif);
4271 	if (error == 0)
4272 		kif->kf_status |= KF_ATTR_VALID;
4273 	kif->kf_flags = xlate_fflags(fflags);
4274 	cap_rights_init_zero(&kif->kf_cap_rights);
4275 	kif->kf_fd = fd;
4276 	kif->kf_ref_count = -1;
4277 	kif->kf_offset = -1;
4278 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4279 		pack_kinfo(kif);
4280 	else
4281 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4282 	vrele(vp);
4283 }
4284 
4285 struct export_fd_buf {
4286 	struct filedesc		*fdp;
4287 	struct pwddesc	*pdp;
4288 	struct sbuf 		*sb;
4289 	ssize_t			remainder;
4290 	struct kinfo_file	kif;
4291 	int			flags;
4292 };
4293 
4294 static int
4295 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4296 {
4297 	struct kinfo_file *kif;
4298 
4299 	kif = &efbuf->kif;
4300 	if (efbuf->remainder != -1) {
4301 		if (efbuf->remainder < kif->kf_structsize)
4302 			return (ENOMEM);
4303 		efbuf->remainder -= kif->kf_structsize;
4304 	}
4305 	if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4306 		return (sbuf_error(efbuf->sb));
4307 	return (0);
4308 }
4309 
4310 static int
4311 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4312     struct export_fd_buf *efbuf)
4313 {
4314 	int error;
4315 
4316 	if (efbuf->remainder == 0)
4317 		return (ENOMEM);
4318 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4319 	    efbuf->flags);
4320 	FILEDESC_SUNLOCK(efbuf->fdp);
4321 	error = export_kinfo_to_sb(efbuf);
4322 	FILEDESC_SLOCK(efbuf->fdp);
4323 	return (error);
4324 }
4325 
4326 static int
4327 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4328     struct export_fd_buf *efbuf)
4329 {
4330 	int error;
4331 
4332 	if (efbuf->remainder == 0)
4333 		return (ENOMEM);
4334 	if (efbuf->pdp != NULL)
4335 		PWDDESC_XUNLOCK(efbuf->pdp);
4336 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4337 	error = export_kinfo_to_sb(efbuf);
4338 	if (efbuf->pdp != NULL)
4339 		PWDDESC_XLOCK(efbuf->pdp);
4340 	return (error);
4341 }
4342 
4343 /*
4344  * Store a process file descriptor information to sbuf.
4345  *
4346  * Takes a locked proc as argument, and returns with the proc unlocked.
4347  */
4348 int
4349 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4350     int flags)
4351 {
4352 	struct file *fp;
4353 	struct filedesc *fdp;
4354 	struct pwddesc *pdp;
4355 	struct export_fd_buf *efbuf;
4356 	struct vnode *cttyvp, *textvp, *tracevp;
4357 	struct pwd *pwd;
4358 	int error, i, lastfile;
4359 	cap_rights_t rights;
4360 
4361 	PROC_LOCK_ASSERT(p, MA_OWNED);
4362 
4363 	/* ktrace vnode */
4364 	tracevp = ktr_get_tracevp(p, true);
4365 	/* text vnode */
4366 	textvp = p->p_textvp;
4367 	if (textvp != NULL)
4368 		vrefact(textvp);
4369 	/* Controlling tty. */
4370 	cttyvp = NULL;
4371 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4372 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4373 		if (cttyvp != NULL)
4374 			vrefact(cttyvp);
4375 	}
4376 	fdp = fdhold(p);
4377 	pdp = pdhold(p);
4378 	PROC_UNLOCK(p);
4379 
4380 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4381 	efbuf->fdp = NULL;
4382 	efbuf->pdp = NULL;
4383 	efbuf->sb = sb;
4384 	efbuf->remainder = maxlen;
4385 	efbuf->flags = flags;
4386 
4387 	error = 0;
4388 	if (tracevp != NULL)
4389 		error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4390 		    FREAD | FWRITE, efbuf);
4391 	if (error == 0 && textvp != NULL)
4392 		error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4393 		    efbuf);
4394 	if (error == 0 && cttyvp != NULL)
4395 		error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4396 		    FREAD | FWRITE, efbuf);
4397 	if (error != 0 || pdp == NULL || fdp == NULL)
4398 		goto fail;
4399 	efbuf->fdp = fdp;
4400 	efbuf->pdp = pdp;
4401 	PWDDESC_XLOCK(pdp);
4402 	pwd = pwd_hold_pwddesc(pdp);
4403 	if (pwd != NULL) {
4404 		/* working directory */
4405 		if (pwd->pwd_cdir != NULL) {
4406 			vrefact(pwd->pwd_cdir);
4407 			error = export_vnode_to_sb(pwd->pwd_cdir,
4408 			    KF_FD_TYPE_CWD, FREAD, efbuf);
4409 		}
4410 		/* root directory */
4411 		if (error == 0 && pwd->pwd_rdir != NULL) {
4412 			vrefact(pwd->pwd_rdir);
4413 			error = export_vnode_to_sb(pwd->pwd_rdir,
4414 			    KF_FD_TYPE_ROOT, FREAD, efbuf);
4415 		}
4416 		/* jail directory */
4417 		if (error == 0 && pwd->pwd_jdir != NULL) {
4418 			vrefact(pwd->pwd_jdir);
4419 			error = export_vnode_to_sb(pwd->pwd_jdir,
4420 			    KF_FD_TYPE_JAIL, FREAD, efbuf);
4421 		}
4422 	}
4423 	PWDDESC_XUNLOCK(pdp);
4424 	if (error != 0)
4425 		goto fail;
4426 	if (pwd != NULL)
4427 		pwd_drop(pwd);
4428 	FILEDESC_SLOCK(fdp);
4429 	lastfile = fdlastfile(fdp);
4430 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4431 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4432 			continue;
4433 #ifdef CAPABILITIES
4434 		rights = *cap_rights(fdp, i);
4435 #else /* !CAPABILITIES */
4436 		rights = cap_no_rights;
4437 #endif
4438 		/*
4439 		 * Create sysctl entry.  It is OK to drop the filedesc
4440 		 * lock inside of export_file_to_sb() as we will
4441 		 * re-validate and re-evaluate its properties when the
4442 		 * loop continues.
4443 		 */
4444 		error = export_file_to_sb(fp, i, &rights, efbuf);
4445 		if (error != 0)
4446 			break;
4447 	}
4448 	FILEDESC_SUNLOCK(fdp);
4449 fail:
4450 	if (fdp != NULL)
4451 		fddrop(fdp);
4452 	if (pdp != NULL)
4453 		pddrop(pdp);
4454 	free(efbuf, M_TEMP);
4455 	return (error);
4456 }
4457 
4458 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4459 
4460 /*
4461  * Get per-process file descriptors for use by procstat(1), et al.
4462  */
4463 static int
4464 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4465 {
4466 	struct sbuf sb;
4467 	struct proc *p;
4468 	ssize_t maxlen;
4469 	u_int namelen;
4470 	int error, error2, *name;
4471 
4472 	namelen = arg2;
4473 	if (namelen != 1)
4474 		return (EINVAL);
4475 
4476 	name = (int *)arg1;
4477 
4478 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4479 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4480 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4481 	if (error != 0) {
4482 		sbuf_delete(&sb);
4483 		return (error);
4484 	}
4485 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4486 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4487 	    KERN_FILEDESC_PACK_KINFO);
4488 	error2 = sbuf_finish(&sb);
4489 	sbuf_delete(&sb);
4490 	return (error != 0 ? error : error2);
4491 }
4492 
4493 #ifdef COMPAT_FREEBSD7
4494 #ifdef KINFO_OFILE_SIZE
4495 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4496 #endif
4497 
4498 static void
4499 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4500 {
4501 
4502 	okif->kf_structsize = sizeof(*okif);
4503 	okif->kf_type = kif->kf_type;
4504 	okif->kf_fd = kif->kf_fd;
4505 	okif->kf_ref_count = kif->kf_ref_count;
4506 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4507 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4508 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4509 	okif->kf_offset = kif->kf_offset;
4510 	if (kif->kf_type == KF_TYPE_VNODE)
4511 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4512 	else
4513 		okif->kf_vnode_type = KF_VTYPE_VNON;
4514 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4515 	if (kif->kf_type == KF_TYPE_SOCKET) {
4516 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4517 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4518 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4519 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4520 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4521 	} else {
4522 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4523 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4524 	}
4525 }
4526 
4527 static int
4528 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4529     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4530 {
4531 	int error;
4532 
4533 	vrefact(vp);
4534 	PWDDESC_XUNLOCK(pdp);
4535 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4536 	kinfo_to_okinfo(kif, okif);
4537 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4538 	PWDDESC_XLOCK(pdp);
4539 	return (error);
4540 }
4541 
4542 /*
4543  * Get per-process file descriptors for use by procstat(1), et al.
4544  */
4545 static int
4546 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4547 {
4548 	struct kinfo_ofile *okif;
4549 	struct kinfo_file *kif;
4550 	struct filedesc *fdp;
4551 	struct pwddesc *pdp;
4552 	struct pwd *pwd;
4553 	u_int namelen;
4554 	int error, i, lastfile, *name;
4555 	struct file *fp;
4556 	struct proc *p;
4557 
4558 	namelen = arg2;
4559 	if (namelen != 1)
4560 		return (EINVAL);
4561 
4562 	name = (int *)arg1;
4563 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4564 	if (error != 0)
4565 		return (error);
4566 	fdp = fdhold(p);
4567 	if (fdp != NULL)
4568 		pdp = pdhold(p);
4569 	PROC_UNLOCK(p);
4570 	if (fdp == NULL || pdp == NULL) {
4571 		if (fdp != NULL)
4572 			fddrop(fdp);
4573 		return (ENOENT);
4574 	}
4575 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4576 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4577 	PWDDESC_XLOCK(pdp);
4578 	pwd = pwd_hold_pwddesc(pdp);
4579 	if (pwd != NULL) {
4580 		if (pwd->pwd_cdir != NULL)
4581 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4582 			    okif, pdp, req);
4583 		if (pwd->pwd_rdir != NULL)
4584 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4585 			    okif, pdp, req);
4586 		if (pwd->pwd_jdir != NULL)
4587 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4588 			    okif, pdp, req);
4589 	}
4590 	PWDDESC_XUNLOCK(pdp);
4591 	if (pwd != NULL)
4592 		pwd_drop(pwd);
4593 	FILEDESC_SLOCK(fdp);
4594 	lastfile = fdlastfile(fdp);
4595 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4596 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4597 			continue;
4598 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4599 		    KERN_FILEDESC_PACK_KINFO);
4600 		FILEDESC_SUNLOCK(fdp);
4601 		kinfo_to_okinfo(kif, okif);
4602 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4603 		FILEDESC_SLOCK(fdp);
4604 		if (error)
4605 			break;
4606 	}
4607 	FILEDESC_SUNLOCK(fdp);
4608 	fddrop(fdp);
4609 	pddrop(pdp);
4610 	free(kif, M_TEMP);
4611 	free(okif, M_TEMP);
4612 	return (0);
4613 }
4614 
4615 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4616     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4617     "Process ofiledesc entries");
4618 #endif	/* COMPAT_FREEBSD7 */
4619 
4620 int
4621 vntype_to_kinfo(int vtype)
4622 {
4623 	struct {
4624 		int	vtype;
4625 		int	kf_vtype;
4626 	} vtypes_table[] = {
4627 		{ VBAD, KF_VTYPE_VBAD },
4628 		{ VBLK, KF_VTYPE_VBLK },
4629 		{ VCHR, KF_VTYPE_VCHR },
4630 		{ VDIR, KF_VTYPE_VDIR },
4631 		{ VFIFO, KF_VTYPE_VFIFO },
4632 		{ VLNK, KF_VTYPE_VLNK },
4633 		{ VNON, KF_VTYPE_VNON },
4634 		{ VREG, KF_VTYPE_VREG },
4635 		{ VSOCK, KF_VTYPE_VSOCK }
4636 	};
4637 	unsigned int i;
4638 
4639 	/*
4640 	 * Perform vtype translation.
4641 	 */
4642 	for (i = 0; i < nitems(vtypes_table); i++)
4643 		if (vtypes_table[i].vtype == vtype)
4644 			return (vtypes_table[i].kf_vtype);
4645 
4646 	return (KF_VTYPE_UNKNOWN);
4647 }
4648 
4649 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4650     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4651     "Process filedesc entries");
4652 
4653 /*
4654  * Store a process current working directory information to sbuf.
4655  *
4656  * Takes a locked proc as argument, and returns with the proc unlocked.
4657  */
4658 int
4659 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4660 {
4661 	struct pwddesc *pdp;
4662 	struct pwd *pwd;
4663 	struct export_fd_buf *efbuf;
4664 	struct vnode *cdir;
4665 	int error;
4666 
4667 	PROC_LOCK_ASSERT(p, MA_OWNED);
4668 
4669 	pdp = pdhold(p);
4670 	PROC_UNLOCK(p);
4671 	if (pdp == NULL)
4672 		return (EINVAL);
4673 
4674 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4675 	efbuf->fdp = NULL;
4676 	efbuf->pdp = pdp;
4677 	efbuf->sb = sb;
4678 	efbuf->remainder = maxlen;
4679 	efbuf->flags = 0;
4680 
4681 	PWDDESC_XLOCK(pdp);
4682 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4683 	cdir = pwd->pwd_cdir;
4684 	if (cdir == NULL) {
4685 		error = EINVAL;
4686 	} else {
4687 		vrefact(cdir);
4688 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4689 	}
4690 	PWDDESC_XUNLOCK(pdp);
4691 	pddrop(pdp);
4692 	free(efbuf, M_TEMP);
4693 	return (error);
4694 }
4695 
4696 /*
4697  * Get per-process current working directory.
4698  */
4699 static int
4700 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4701 {
4702 	struct sbuf sb;
4703 	struct proc *p;
4704 	ssize_t maxlen;
4705 	u_int namelen;
4706 	int error, error2, *name;
4707 
4708 	namelen = arg2;
4709 	if (namelen != 1)
4710 		return (EINVAL);
4711 
4712 	name = (int *)arg1;
4713 
4714 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4715 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4716 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4717 	if (error != 0) {
4718 		sbuf_delete(&sb);
4719 		return (error);
4720 	}
4721 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4722 	error = kern_proc_cwd_out(p, &sb, maxlen);
4723 	error2 = sbuf_finish(&sb);
4724 	sbuf_delete(&sb);
4725 	return (error != 0 ? error : error2);
4726 }
4727 
4728 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4729     sysctl_kern_proc_cwd, "Process current working directory");
4730 
4731 #ifdef DDB
4732 /*
4733  * For the purposes of debugging, generate a human-readable string for the
4734  * file type.
4735  */
4736 static const char *
4737 file_type_to_name(short type)
4738 {
4739 
4740 	switch (type) {
4741 	case 0:
4742 		return ("zero");
4743 	case DTYPE_VNODE:
4744 		return ("vnode");
4745 	case DTYPE_SOCKET:
4746 		return ("socket");
4747 	case DTYPE_PIPE:
4748 		return ("pipe");
4749 	case DTYPE_FIFO:
4750 		return ("fifo");
4751 	case DTYPE_KQUEUE:
4752 		return ("kqueue");
4753 	case DTYPE_CRYPTO:
4754 		return ("crypto");
4755 	case DTYPE_MQUEUE:
4756 		return ("mqueue");
4757 	case DTYPE_SHM:
4758 		return ("shm");
4759 	case DTYPE_SEM:
4760 		return ("ksem");
4761 	case DTYPE_PTS:
4762 		return ("pts");
4763 	case DTYPE_DEV:
4764 		return ("dev");
4765 	case DTYPE_PROCDESC:
4766 		return ("proc");
4767 	case DTYPE_EVENTFD:
4768 		return ("eventfd");
4769 	case DTYPE_LINUXTFD:
4770 		return ("ltimer");
4771 	default:
4772 		return ("unkn");
4773 	}
4774 }
4775 
4776 /*
4777  * For the purposes of debugging, identify a process (if any, perhaps one of
4778  * many) that references the passed file in its file descriptor array. Return
4779  * NULL if none.
4780  */
4781 static struct proc *
4782 file_to_first_proc(struct file *fp)
4783 {
4784 	struct filedesc *fdp;
4785 	struct proc *p;
4786 	int n;
4787 
4788 	FOREACH_PROC_IN_SYSTEM(p) {
4789 		if (p->p_state == PRS_NEW)
4790 			continue;
4791 		fdp = p->p_fd;
4792 		if (fdp == NULL)
4793 			continue;
4794 		for (n = 0; n < fdp->fd_nfiles; n++) {
4795 			if (fp == fdp->fd_ofiles[n].fde_file)
4796 				return (p);
4797 		}
4798 	}
4799 	return (NULL);
4800 }
4801 
4802 static void
4803 db_print_file(struct file *fp, int header)
4804 {
4805 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4806 	struct proc *p;
4807 
4808 	if (header)
4809 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4810 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4811 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4812 		    "FCmd");
4813 	p = file_to_first_proc(fp);
4814 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4815 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4816 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4817 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4818 
4819 #undef XPTRWIDTH
4820 }
4821 
4822 DB_SHOW_COMMAND(file, db_show_file)
4823 {
4824 	struct file *fp;
4825 
4826 	if (!have_addr) {
4827 		db_printf("usage: show file <addr>\n");
4828 		return;
4829 	}
4830 	fp = (struct file *)addr;
4831 	db_print_file(fp, 1);
4832 }
4833 
4834 DB_SHOW_COMMAND(files, db_show_files)
4835 {
4836 	struct filedesc *fdp;
4837 	struct file *fp;
4838 	struct proc *p;
4839 	int header;
4840 	int n;
4841 
4842 	header = 1;
4843 	FOREACH_PROC_IN_SYSTEM(p) {
4844 		if (p->p_state == PRS_NEW)
4845 			continue;
4846 		if ((fdp = p->p_fd) == NULL)
4847 			continue;
4848 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4849 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4850 				continue;
4851 			db_print_file(fp, header);
4852 			header = 0;
4853 		}
4854 	}
4855 }
4856 #endif
4857 
4858 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4859     &maxfilesperproc, 0, "Maximum files allowed open per process");
4860 
4861 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4862     &maxfiles, 0, "Maximum number of files");
4863 
4864 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4865     &openfiles, 0, "System-wide number of open files");
4866 
4867 /* ARGSUSED*/
4868 static void
4869 filelistinit(void *dummy)
4870 {
4871 
4872 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4873 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4874 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4875 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4876 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4877 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4878 	/*
4879 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4880 	 * the vnode zone.
4881 	 */
4882 	vfs_smr = uma_zone_get_smr(pwd_zone);
4883 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4884 }
4885 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4886 
4887 /*-------------------------------------------------------------------*/
4888 
4889 static int
4890 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4891     int flags, struct thread *td)
4892 {
4893 
4894 	return (EBADF);
4895 }
4896 
4897 static int
4898 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4899     struct thread *td)
4900 {
4901 
4902 	return (EINVAL);
4903 }
4904 
4905 static int
4906 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4907     struct thread *td)
4908 {
4909 
4910 	return (EBADF);
4911 }
4912 
4913 static int
4914 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4915     struct thread *td)
4916 {
4917 
4918 	return (0);
4919 }
4920 
4921 static int
4922 badfo_kqfilter(struct file *fp, struct knote *kn)
4923 {
4924 
4925 	return (EBADF);
4926 }
4927 
4928 static int
4929 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
4930 {
4931 
4932 	return (EBADF);
4933 }
4934 
4935 static int
4936 badfo_close(struct file *fp, struct thread *td)
4937 {
4938 
4939 	return (0);
4940 }
4941 
4942 static int
4943 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4944     struct thread *td)
4945 {
4946 
4947 	return (EBADF);
4948 }
4949 
4950 static int
4951 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4952     struct thread *td)
4953 {
4954 
4955 	return (EBADF);
4956 }
4957 
4958 static int
4959 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4960     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4961     struct thread *td)
4962 {
4963 
4964 	return (EBADF);
4965 }
4966 
4967 static int
4968 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4969 {
4970 
4971 	return (0);
4972 }
4973 
4974 struct fileops badfileops = {
4975 	.fo_read = badfo_readwrite,
4976 	.fo_write = badfo_readwrite,
4977 	.fo_truncate = badfo_truncate,
4978 	.fo_ioctl = badfo_ioctl,
4979 	.fo_poll = badfo_poll,
4980 	.fo_kqfilter = badfo_kqfilter,
4981 	.fo_stat = badfo_stat,
4982 	.fo_close = badfo_close,
4983 	.fo_chmod = badfo_chmod,
4984 	.fo_chown = badfo_chown,
4985 	.fo_sendfile = badfo_sendfile,
4986 	.fo_fill_kinfo = badfo_fill_kinfo,
4987 };
4988 
4989 static int
4990 path_poll(struct file *fp, int events, struct ucred *active_cred,
4991     struct thread *td)
4992 {
4993 	return (POLLNVAL);
4994 }
4995 
4996 static int
4997 path_close(struct file *fp, struct thread *td)
4998 {
4999 	MPASS(fp->f_type == DTYPE_VNODE);
5000 	fp->f_ops = &badfileops;
5001 	vdrop(fp->f_vnode);
5002 	return (0);
5003 }
5004 
5005 struct fileops path_fileops = {
5006 	.fo_read = badfo_readwrite,
5007 	.fo_write = badfo_readwrite,
5008 	.fo_truncate = badfo_truncate,
5009 	.fo_ioctl = badfo_ioctl,
5010 	.fo_poll = path_poll,
5011 	.fo_kqfilter = vn_kqfilter_opath,
5012 	.fo_stat = vn_statfile,
5013 	.fo_close = path_close,
5014 	.fo_chmod = badfo_chmod,
5015 	.fo_chown = badfo_chown,
5016 	.fo_sendfile = badfo_sendfile,
5017 	.fo_fill_kinfo = vn_fill_kinfo,
5018 	.fo_flags = DFLAG_PASSABLE,
5019 };
5020 
5021 int
5022 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5023     int flags, struct thread *td)
5024 {
5025 
5026 	return (EOPNOTSUPP);
5027 }
5028 
5029 int
5030 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5031     struct thread *td)
5032 {
5033 
5034 	return (EINVAL);
5035 }
5036 
5037 int
5038 invfo_ioctl(struct file *fp, u_long com, void *data,
5039     struct ucred *active_cred, struct thread *td)
5040 {
5041 
5042 	return (ENOTTY);
5043 }
5044 
5045 int
5046 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5047     struct thread *td)
5048 {
5049 
5050 	return (poll_no_poll(events));
5051 }
5052 
5053 int
5054 invfo_kqfilter(struct file *fp, struct knote *kn)
5055 {
5056 
5057 	return (EINVAL);
5058 }
5059 
5060 int
5061 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5062     struct thread *td)
5063 {
5064 
5065 	return (EINVAL);
5066 }
5067 
5068 int
5069 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5070     struct thread *td)
5071 {
5072 
5073 	return (EINVAL);
5074 }
5075 
5076 int
5077 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5078     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5079     struct thread *td)
5080 {
5081 
5082 	return (EINVAL);
5083 }
5084 
5085 /*-------------------------------------------------------------------*/
5086 
5087 /*
5088  * File Descriptor pseudo-device driver (/dev/fd/).
5089  *
5090  * Opening minor device N dup()s the file (if any) connected to file
5091  * descriptor N belonging to the calling process.  Note that this driver
5092  * consists of only the ``open()'' routine, because all subsequent
5093  * references to this file will be direct to the other driver.
5094  *
5095  * XXX: we could give this one a cloning event handler if necessary.
5096  */
5097 
5098 /* ARGSUSED */
5099 static int
5100 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5101 {
5102 
5103 	/*
5104 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5105 	 * the file descriptor being sought for duplication. The error
5106 	 * return ensures that the vnode for this device will be released
5107 	 * by vn_open. Open will detect this special error and take the
5108 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5109 	 * will simply report the error.
5110 	 */
5111 	td->td_dupfd = dev2unit(dev);
5112 	return (ENODEV);
5113 }
5114 
5115 static struct cdevsw fildesc_cdevsw = {
5116 	.d_version =	D_VERSION,
5117 	.d_open =	fdopen,
5118 	.d_name =	"FD",
5119 };
5120 
5121 static void
5122 fildesc_drvinit(void *unused)
5123 {
5124 	struct cdev *dev;
5125 
5126 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5127 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5128 	make_dev_alias(dev, "stdin");
5129 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5130 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5131 	make_dev_alias(dev, "stdout");
5132 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5133 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5134 	make_dev_alias(dev, "stderr");
5135 }
5136 
5137 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5138