xref: /freebsd/sys/kern/kern_descrip.c (revision c66ec88fed842fbaad62c30d510644ceb7bd2d71)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
97 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
98     "file desc to leader structures");
99 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
100 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
101 
102 MALLOC_DECLARE(M_FADVISE);
103 
104 static __read_mostly uma_zone_t file_zone;
105 static __read_mostly uma_zone_t filedesc0_zone;
106 __read_mostly uma_zone_t pwd_zone;
107 VFS_SMR_DECLARE;
108 
109 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
110 		    struct thread *td, bool holdleaders, bool audit);
111 static int	fd_first_free(struct filedesc *fdp, int low, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the last used fd.
217  *
218  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
219  * Otherwise use fdlastfile.
220  */
221 int
222 fdlastfile_single(struct filedesc *fdp)
223 {
224 	NDSLOTTYPE *map = fdp->fd_map;
225 	int off, minoff;
226 
227 	off = NDSLOT(fdp->fd_nfiles - 1);
228 	for (minoff = NDSLOT(0); off >= minoff; --off)
229 		if (map[off] != 0)
230 			return (off * NDENTRIES + flsl(map[off]) - 1);
231 	return (-1);
232 }
233 
234 int
235 fdlastfile(struct filedesc *fdp)
236 {
237 
238 	FILEDESC_LOCK_ASSERT(fdp);
239 	return (fdlastfile_single(fdp));
240 }
241 
242 static int
243 fdisused(struct filedesc *fdp, int fd)
244 {
245 
246 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
247 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
248 
249 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
250 }
251 
252 /*
253  * Mark a file descriptor as used.
254  */
255 static void
256 fdused_init(struct filedesc *fdp, int fd)
257 {
258 
259 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
260 
261 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
262 }
263 
264 static void
265 fdused(struct filedesc *fdp, int fd)
266 {
267 
268 	FILEDESC_XLOCK_ASSERT(fdp);
269 
270 	fdused_init(fdp, fd);
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile++;
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	FILEDESC_XLOCK_ASSERT(fdp);
311 	fde = &fdp->fd_ofiles[fd];
312 #ifdef CAPABILITIES
313 	seqc_write_begin(&fde->fde_seqc);
314 #endif
315 	fde->fde_file = NULL;
316 #ifdef CAPABILITIES
317 	seqc_write_end(&fde->fde_seqc);
318 #endif
319 	fdefree_last(fde);
320 	fdunused(fdp, fd);
321 }
322 
323 /*
324  * System calls on descriptors.
325  */
326 #ifndef _SYS_SYSPROTO_H_
327 struct getdtablesize_args {
328 	int	dummy;
329 };
330 #endif
331 /* ARGSUSED */
332 int
333 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
334 {
335 #ifdef	RACCT
336 	uint64_t lim;
337 #endif
338 
339 	td->td_retval[0] = getmaxfd(td);
340 #ifdef	RACCT
341 	PROC_LOCK(td->td_proc);
342 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
343 	PROC_UNLOCK(td->td_proc);
344 	if (lim < td->td_retval[0])
345 		td->td_retval[0] = lim;
346 #endif
347 	return (0);
348 }
349 
350 /*
351  * Duplicate a file descriptor to a particular value.
352  *
353  * Note: keep in mind that a potential race condition exists when closing
354  * descriptors from a shared descriptor table (via rfork).
355  */
356 #ifndef _SYS_SYSPROTO_H_
357 struct dup2_args {
358 	u_int	from;
359 	u_int	to;
360 };
361 #endif
362 /* ARGSUSED */
363 int
364 sys_dup2(struct thread *td, struct dup2_args *uap)
365 {
366 
367 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
368 }
369 
370 /*
371  * Duplicate a file descriptor.
372  */
373 #ifndef _SYS_SYSPROTO_H_
374 struct dup_args {
375 	u_int	fd;
376 };
377 #endif
378 /* ARGSUSED */
379 int
380 sys_dup(struct thread *td, struct dup_args *uap)
381 {
382 
383 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
384 }
385 
386 /*
387  * The file control system call.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct fcntl_args {
391 	int	fd;
392 	int	cmd;
393 	long	arg;
394 };
395 #endif
396 /* ARGSUSED */
397 int
398 sys_fcntl(struct thread *td, struct fcntl_args *uap)
399 {
400 
401 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
402 }
403 
404 int
405 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
406 {
407 	struct flock fl;
408 	struct __oflock ofl;
409 	intptr_t arg1;
410 	int error, newcmd;
411 
412 	error = 0;
413 	newcmd = cmd;
414 	switch (cmd) {
415 	case F_OGETLK:
416 	case F_OSETLK:
417 	case F_OSETLKW:
418 		/*
419 		 * Convert old flock structure to new.
420 		 */
421 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
422 		fl.l_start = ofl.l_start;
423 		fl.l_len = ofl.l_len;
424 		fl.l_pid = ofl.l_pid;
425 		fl.l_type = ofl.l_type;
426 		fl.l_whence = ofl.l_whence;
427 		fl.l_sysid = 0;
428 
429 		switch (cmd) {
430 		case F_OGETLK:
431 			newcmd = F_GETLK;
432 			break;
433 		case F_OSETLK:
434 			newcmd = F_SETLK;
435 			break;
436 		case F_OSETLKW:
437 			newcmd = F_SETLKW;
438 			break;
439 		}
440 		arg1 = (intptr_t)&fl;
441 		break;
442 	case F_GETLK:
443 	case F_SETLK:
444 	case F_SETLKW:
445 	case F_SETLK_REMOTE:
446 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
447 		arg1 = (intptr_t)&fl;
448 		break;
449 	default:
450 		arg1 = arg;
451 		break;
452 	}
453 	if (error)
454 		return (error);
455 	error = kern_fcntl(td, fd, newcmd, arg1);
456 	if (error)
457 		return (error);
458 	if (cmd == F_OGETLK) {
459 		ofl.l_start = fl.l_start;
460 		ofl.l_len = fl.l_len;
461 		ofl.l_pid = fl.l_pid;
462 		ofl.l_type = fl.l_type;
463 		ofl.l_whence = fl.l_whence;
464 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
465 	} else if (cmd == F_GETLK) {
466 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
467 	}
468 	return (error);
469 }
470 
471 int
472 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 {
474 	struct filedesc *fdp;
475 	struct flock *flp;
476 	struct file *fp, *fp2;
477 	struct filedescent *fde;
478 	struct proc *p;
479 	struct vnode *vp;
480 	struct mount *mp;
481 	int error, flg, seals, tmp;
482 	uint64_t bsize;
483 	off_t foffset;
484 
485 	error = 0;
486 	flg = F_POSIX;
487 	p = td->td_proc;
488 	fdp = p->p_fd;
489 
490 	AUDIT_ARG_FD(cmd);
491 	AUDIT_ARG_CMD(cmd);
492 	switch (cmd) {
493 	case F_DUPFD:
494 		tmp = arg;
495 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
496 		break;
497 
498 	case F_DUPFD_CLOEXEC:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
501 		break;
502 
503 	case F_DUP2FD:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD_CLOEXEC:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
511 		break;
512 
513 	case F_GETFD:
514 		error = EBADF;
515 		FILEDESC_SLOCK(fdp);
516 		fde = fdeget_locked(fdp, fd);
517 		if (fde != NULL) {
518 			td->td_retval[0] =
519 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
520 			error = 0;
521 		}
522 		FILEDESC_SUNLOCK(fdp);
523 		break;
524 
525 	case F_SETFD:
526 		error = EBADF;
527 		FILEDESC_XLOCK(fdp);
528 		fde = fdeget_locked(fdp, fd);
529 		if (fde != NULL) {
530 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
531 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
532 			error = 0;
533 		}
534 		FILEDESC_XUNLOCK(fdp);
535 		break;
536 
537 	case F_GETFL:
538 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
539 		if (error != 0)
540 			break;
541 		td->td_retval[0] = OFLAGS(fp->f_flag);
542 		fdrop(fp, td);
543 		break;
544 
545 	case F_SETFL:
546 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
547 		if (error != 0)
548 			break;
549 		do {
550 			tmp = flg = fp->f_flag;
551 			tmp &= ~FCNTLFLAGS;
552 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
553 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
554 		tmp = fp->f_flag & FNONBLOCK;
555 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
556 		if (error != 0) {
557 			fdrop(fp, td);
558 			break;
559 		}
560 		tmp = fp->f_flag & FASYNC;
561 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
562 		if (error == 0) {
563 			fdrop(fp, td);
564 			break;
565 		}
566 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
567 		tmp = 0;
568 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
569 		fdrop(fp, td);
570 		break;
571 
572 	case F_GETOWN:
573 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
574 		if (error != 0)
575 			break;
576 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 		if (error == 0)
578 			td->td_retval[0] = tmp;
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_SETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		tmp = arg;
587 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
588 		fdrop(fp, td);
589 		break;
590 
591 	case F_SETLK_REMOTE:
592 		error = priv_check(td, PRIV_NFS_LOCKD);
593 		if (error != 0)
594 			return (error);
595 		flg = F_REMOTE;
596 		goto do_setlk;
597 
598 	case F_SETLKW:
599 		flg |= F_WAIT;
600 		/* FALLTHROUGH F_SETLK */
601 
602 	case F_SETLK:
603 	do_setlk:
604 		flp = (struct flock *)arg;
605 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
606 			error = EINVAL;
607 			break;
608 		}
609 
610 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
611 		if (error != 0)
612 			break;
613 		if (fp->f_type != DTYPE_VNODE) {
614 			error = EBADF;
615 			fdrop(fp, td);
616 			break;
617 		}
618 
619 		if (flp->l_whence == SEEK_CUR) {
620 			foffset = foffset_get(fp);
621 			if (foffset < 0 ||
622 			    (flp->l_start > 0 &&
623 			     foffset > OFF_MAX - flp->l_start)) {
624 				error = EOVERFLOW;
625 				fdrop(fp, td);
626 				break;
627 			}
628 			flp->l_start += foffset;
629 		}
630 
631 		vp = fp->f_vnode;
632 		switch (flp->l_type) {
633 		case F_RDLCK:
634 			if ((fp->f_flag & FREAD) == 0) {
635 				error = EBADF;
636 				break;
637 			}
638 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
639 				PROC_LOCK(p->p_leader);
640 				p->p_leader->p_flag |= P_ADVLOCK;
641 				PROC_UNLOCK(p->p_leader);
642 			}
643 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
644 			    flp, flg);
645 			break;
646 		case F_WRLCK:
647 			if ((fp->f_flag & FWRITE) == 0) {
648 				error = EBADF;
649 				break;
650 			}
651 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
652 				PROC_LOCK(p->p_leader);
653 				p->p_leader->p_flag |= P_ADVLOCK;
654 				PROC_UNLOCK(p->p_leader);
655 			}
656 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
657 			    flp, flg);
658 			break;
659 		case F_UNLCK:
660 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
661 			    flp, flg);
662 			break;
663 		case F_UNLCKSYS:
664 			if (flg != F_REMOTE) {
665 				error = EINVAL;
666 				break;
667 			}
668 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
669 			    F_UNLCKSYS, flp, flg);
670 			break;
671 		default:
672 			error = EINVAL;
673 			break;
674 		}
675 		if (error != 0 || flp->l_type == F_UNLCK ||
676 		    flp->l_type == F_UNLCKSYS) {
677 			fdrop(fp, td);
678 			break;
679 		}
680 
681 		/*
682 		 * Check for a race with close.
683 		 *
684 		 * The vnode is now advisory locked (or unlocked, but this case
685 		 * is not really important) as the caller requested.
686 		 * We had to drop the filedesc lock, so we need to recheck if
687 		 * the descriptor is still valid, because if it was closed
688 		 * in the meantime we need to remove advisory lock from the
689 		 * vnode - close on any descriptor leading to an advisory
690 		 * locked vnode, removes that lock.
691 		 * We will return 0 on purpose in that case, as the result of
692 		 * successful advisory lock might have been externally visible
693 		 * already. This is fine - effectively we pretend to the caller
694 		 * that the closing thread was a bit slower and that the
695 		 * advisory lock succeeded before the close.
696 		 */
697 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
698 		if (error != 0) {
699 			fdrop(fp, td);
700 			break;
701 		}
702 		if (fp != fp2) {
703 			flp->l_whence = SEEK_SET;
704 			flp->l_start = 0;
705 			flp->l_len = 0;
706 			flp->l_type = F_UNLCK;
707 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
708 			    F_UNLCK, flp, F_POSIX);
709 		}
710 		fdrop(fp, td);
711 		fdrop(fp2, td);
712 		break;
713 
714 	case F_GETLK:
715 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
716 		if (error != 0)
717 			break;
718 		if (fp->f_type != DTYPE_VNODE) {
719 			error = EBADF;
720 			fdrop(fp, td);
721 			break;
722 		}
723 		flp = (struct flock *)arg;
724 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
725 		    flp->l_type != F_UNLCK) {
726 			error = EINVAL;
727 			fdrop(fp, td);
728 			break;
729 		}
730 		if (flp->l_whence == SEEK_CUR) {
731 			foffset = foffset_get(fp);
732 			if ((flp->l_start > 0 &&
733 			    foffset > OFF_MAX - flp->l_start) ||
734 			    (flp->l_start < 0 &&
735 			    foffset < OFF_MIN - flp->l_start)) {
736 				error = EOVERFLOW;
737 				fdrop(fp, td);
738 				break;
739 			}
740 			flp->l_start += foffset;
741 		}
742 		vp = fp->f_vnode;
743 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
744 		    F_POSIX);
745 		fdrop(fp, td);
746 		break;
747 
748 	case F_ADD_SEALS:
749 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
750 		if (error != 0)
751 			break;
752 		error = fo_add_seals(fp, arg);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_GET_SEALS:
757 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
758 		if (error != 0)
759 			break;
760 		if (fo_get_seals(fp, &seals) == 0)
761 			td->td_retval[0] = seals;
762 		else
763 			error = EINVAL;
764 		fdrop(fp, td);
765 		break;
766 
767 	case F_RDAHEAD:
768 		arg = arg ? 128 * 1024: 0;
769 		/* FALLTHROUGH */
770 	case F_READAHEAD:
771 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
772 		if (error != 0)
773 			break;
774 		if (fp->f_type != DTYPE_VNODE) {
775 			fdrop(fp, td);
776 			error = EBADF;
777 			break;
778 		}
779 		vp = fp->f_vnode;
780 		if (vp->v_type != VREG) {
781 			fdrop(fp, td);
782 			error = ENOTTY;
783 			break;
784 		}
785 
786 		/*
787 		 * Exclusive lock synchronizes against f_seqcount reads and
788 		 * writes in sequential_heuristic().
789 		 */
790 		error = vn_lock(vp, LK_EXCLUSIVE);
791 		if (error != 0) {
792 			fdrop(fp, td);
793 			break;
794 		}
795 		if (arg >= 0) {
796 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
797 			arg = MIN(arg, INT_MAX - bsize + 1);
798 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
799 			    (arg + bsize - 1) / bsize);
800 			atomic_set_int(&fp->f_flag, FRDAHEAD);
801 		} else {
802 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
803 		}
804 		VOP_UNLOCK(vp);
805 		fdrop(fp, td);
806 		break;
807 
808 	case F_ISUNIONSTACK:
809 		/*
810 		 * Check if the vnode is part of a union stack (either the
811 		 * "union" flag from mount(2) or unionfs).
812 		 *
813 		 * Prior to introduction of this op libc's readdir would call
814 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
815 		 * data just to check fs name and a mount flag.
816 		 *
817 		 * Fixing the code to handle everything in the kernel instead
818 		 * is a non-trivial endeavor and has low priority, thus this
819 		 * horrible kludge facilitates the current behavior in a much
820 		 * cheaper manner until someone(tm) sorts this out.
821 		 */
822 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
823 		if (error != 0)
824 			break;
825 		if (fp->f_type != DTYPE_VNODE) {
826 			fdrop(fp, td);
827 			error = EBADF;
828 			break;
829 		}
830 		vp = fp->f_vnode;
831 		/*
832 		 * Since we don't prevent dooming the vnode even non-null mp
833 		 * found can become immediately stale. This is tolerable since
834 		 * mount points are type-stable (providing safe memory access)
835 		 * and any vfs op on this vnode going forward will return an
836 		 * error (meaning return value in this case is meaningless).
837 		 */
838 		mp = atomic_load_ptr(&vp->v_mount);
839 		if (__predict_false(mp == NULL)) {
840 			fdrop(fp, td);
841 			error = EBADF;
842 			break;
843 		}
844 		td->td_retval[0] = 0;
845 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
846 		    mp->mnt_flag & MNT_UNION)
847 			td->td_retval[0] = 1;
848 		fdrop(fp, td);
849 		break;
850 
851 	default:
852 		error = EINVAL;
853 		break;
854 	}
855 	return (error);
856 }
857 
858 static int
859 getmaxfd(struct thread *td)
860 {
861 
862 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
863 }
864 
865 /*
866  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
867  */
868 int
869 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
870 {
871 	struct filedesc *fdp;
872 	struct filedescent *oldfde, *newfde;
873 	struct proc *p;
874 	struct file *delfp, *oldfp;
875 	u_long *oioctls, *nioctls;
876 	int error, maxfd;
877 
878 	p = td->td_proc;
879 	fdp = p->p_fd;
880 	oioctls = NULL;
881 
882 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
883 	MPASS(mode < FDDUP_LASTMODE);
884 
885 	AUDIT_ARG_FD(old);
886 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
887 
888 	/*
889 	 * Verify we have a valid descriptor to dup from and possibly to
890 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
891 	 * return EINVAL when the new descriptor is out of bounds.
892 	 */
893 	if (old < 0)
894 		return (EBADF);
895 	if (new < 0)
896 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
897 	maxfd = getmaxfd(td);
898 	if (new >= maxfd)
899 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
900 
901 	error = EBADF;
902 	FILEDESC_XLOCK(fdp);
903 	if (fget_locked(fdp, old) == NULL)
904 		goto unlock;
905 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
906 		td->td_retval[0] = new;
907 		if (flags & FDDUP_FLAG_CLOEXEC)
908 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
909 		error = 0;
910 		goto unlock;
911 	}
912 
913 	oldfde = &fdp->fd_ofiles[old];
914 	oldfp = oldfde->fde_file;
915 	if (!fhold(oldfp))
916 		goto unlock;
917 
918 	/*
919 	 * If the caller specified a file descriptor, make sure the file
920 	 * table is large enough to hold it, and grab it.  Otherwise, just
921 	 * allocate a new descriptor the usual way.
922 	 */
923 	switch (mode) {
924 	case FDDUP_NORMAL:
925 	case FDDUP_FCNTL:
926 		if ((error = fdalloc(td, new, &new)) != 0) {
927 			fdrop(oldfp, td);
928 			goto unlock;
929 		}
930 		break;
931 	case FDDUP_MUSTREPLACE:
932 		/* Target file descriptor must exist. */
933 		if (fget_locked(fdp, new) == NULL) {
934 			fdrop(oldfp, td);
935 			goto unlock;
936 		}
937 		break;
938 	case FDDUP_FIXED:
939 		if (new >= fdp->fd_nfiles) {
940 			/*
941 			 * The resource limits are here instead of e.g.
942 			 * fdalloc(), because the file descriptor table may be
943 			 * shared between processes, so we can't really use
944 			 * racct_add()/racct_sub().  Instead of counting the
945 			 * number of actually allocated descriptors, just put
946 			 * the limit on the size of the file descriptor table.
947 			 */
948 #ifdef RACCT
949 			if (RACCT_ENABLED()) {
950 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
951 				if (error != 0) {
952 					error = EMFILE;
953 					fdrop(oldfp, td);
954 					goto unlock;
955 				}
956 			}
957 #endif
958 			fdgrowtable_exp(fdp, new + 1);
959 		}
960 		if (!fdisused(fdp, new))
961 			fdused(fdp, new);
962 		break;
963 	default:
964 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
965 	}
966 
967 	KASSERT(old != new, ("new fd is same as old"));
968 
969 	/* Refetch oldfde because the table may have grown and old one freed. */
970 	oldfde = &fdp->fd_ofiles[old];
971 	KASSERT(oldfp == oldfde->fde_file,
972 	    ("fdt_ofiles shift from growth observed at fd %d",
973 	    old));
974 
975 	newfde = &fdp->fd_ofiles[new];
976 	delfp = newfde->fde_file;
977 
978 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
979 
980 	/*
981 	 * Duplicate the source descriptor.
982 	 */
983 #ifdef CAPABILITIES
984 	seqc_write_begin(&newfde->fde_seqc);
985 #endif
986 	oioctls = filecaps_free_prep(&newfde->fde_caps);
987 	memcpy(newfde, oldfde, fde_change_size);
988 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
989 	    nioctls);
990 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
991 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
992 	else
993 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
994 #ifdef CAPABILITIES
995 	seqc_write_end(&newfde->fde_seqc);
996 #endif
997 	td->td_retval[0] = new;
998 
999 	error = 0;
1000 
1001 	if (delfp != NULL) {
1002 		(void) closefp(fdp, new, delfp, td, true, false);
1003 		FILEDESC_UNLOCK_ASSERT(fdp);
1004 	} else {
1005 unlock:
1006 		FILEDESC_XUNLOCK(fdp);
1007 	}
1008 
1009 	filecaps_free_finish(oioctls);
1010 	return (error);
1011 }
1012 
1013 static void
1014 sigiofree(struct sigio *sigio)
1015 {
1016 	crfree(sigio->sio_ucred);
1017 	free(sigio, M_SIGIO);
1018 }
1019 
1020 static struct sigio *
1021 funsetown_locked(struct sigio *sigio)
1022 {
1023 	struct proc *p;
1024 	struct pgrp *pg;
1025 
1026 	SIGIO_ASSERT_LOCKED();
1027 
1028 	if (sigio == NULL)
1029 		return (NULL);
1030 	*(sigio->sio_myref) = NULL;
1031 	if (sigio->sio_pgid < 0) {
1032 		pg = sigio->sio_pgrp;
1033 		PGRP_LOCK(pg);
1034 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1035 		    sigio, sio_pgsigio);
1036 		PGRP_UNLOCK(pg);
1037 	} else {
1038 		p = sigio->sio_proc;
1039 		PROC_LOCK(p);
1040 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1041 		    sigio, sio_pgsigio);
1042 		PROC_UNLOCK(p);
1043 	}
1044 	return (sigio);
1045 }
1046 
1047 /*
1048  * If sigio is on the list associated with a process or process group,
1049  * disable signalling from the device, remove sigio from the list and
1050  * free sigio.
1051  */
1052 void
1053 funsetown(struct sigio **sigiop)
1054 {
1055 	struct sigio *sigio;
1056 
1057 	/* Racy check, consumers must provide synchronization. */
1058 	if (*sigiop == NULL)
1059 		return;
1060 
1061 	SIGIO_LOCK();
1062 	sigio = funsetown_locked(*sigiop);
1063 	SIGIO_UNLOCK();
1064 	if (sigio != NULL)
1065 		sigiofree(sigio);
1066 }
1067 
1068 /*
1069  * Free a list of sigio structures.  The caller must ensure that new sigio
1070  * structures cannot be added after this point.  For process groups this is
1071  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1072  * as an interlock.
1073  */
1074 void
1075 funsetownlst(struct sigiolst *sigiolst)
1076 {
1077 	struct proc *p;
1078 	struct pgrp *pg;
1079 	struct sigio *sigio, *tmp;
1080 
1081 	/* Racy check. */
1082 	sigio = SLIST_FIRST(sigiolst);
1083 	if (sigio == NULL)
1084 		return;
1085 
1086 	p = NULL;
1087 	pg = NULL;
1088 
1089 	SIGIO_LOCK();
1090 	sigio = SLIST_FIRST(sigiolst);
1091 	if (sigio == NULL) {
1092 		SIGIO_UNLOCK();
1093 		return;
1094 	}
1095 
1096 	/*
1097 	 * Every entry of the list should belong to a single proc or pgrp.
1098 	 */
1099 	if (sigio->sio_pgid < 0) {
1100 		pg = sigio->sio_pgrp;
1101 		sx_assert(&proctree_lock, SX_XLOCKED);
1102 		PGRP_LOCK(pg);
1103 	} else /* if (sigio->sio_pgid > 0) */ {
1104 		p = sigio->sio_proc;
1105 		PROC_LOCK(p);
1106 		KASSERT((p->p_flag & P_WEXIT) != 0,
1107 		    ("%s: process %p is not exiting", __func__, p));
1108 	}
1109 
1110 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1111 		*sigio->sio_myref = NULL;
1112 		if (pg != NULL) {
1113 			KASSERT(sigio->sio_pgid < 0,
1114 			    ("Proc sigio in pgrp sigio list"));
1115 			KASSERT(sigio->sio_pgrp == pg,
1116 			    ("Bogus pgrp in sigio list"));
1117 		} else /* if (p != NULL) */ {
1118 			KASSERT(sigio->sio_pgid > 0,
1119 			    ("Pgrp sigio in proc sigio list"));
1120 			KASSERT(sigio->sio_proc == p,
1121 			    ("Bogus proc in sigio list"));
1122 		}
1123 	}
1124 
1125 	if (pg != NULL)
1126 		PGRP_UNLOCK(pg);
1127 	else
1128 		PROC_UNLOCK(p);
1129 	SIGIO_UNLOCK();
1130 
1131 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1132 		sigiofree(sigio);
1133 }
1134 
1135 /*
1136  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1137  *
1138  * After permission checking, add a sigio structure to the sigio list for
1139  * the process or process group.
1140  */
1141 int
1142 fsetown(pid_t pgid, struct sigio **sigiop)
1143 {
1144 	struct proc *proc;
1145 	struct pgrp *pgrp;
1146 	struct sigio *osigio, *sigio;
1147 	int ret;
1148 
1149 	if (pgid == 0) {
1150 		funsetown(sigiop);
1151 		return (0);
1152 	}
1153 
1154 	ret = 0;
1155 
1156 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1157 	sigio->sio_pgid = pgid;
1158 	sigio->sio_ucred = crhold(curthread->td_ucred);
1159 	sigio->sio_myref = sigiop;
1160 
1161 	sx_slock(&proctree_lock);
1162 	SIGIO_LOCK();
1163 	osigio = funsetown_locked(*sigiop);
1164 	if (pgid > 0) {
1165 		proc = pfind(pgid);
1166 		if (proc == NULL) {
1167 			ret = ESRCH;
1168 			goto fail;
1169 		}
1170 
1171 		/*
1172 		 * Policy - Don't allow a process to FSETOWN a process
1173 		 * in another session.
1174 		 *
1175 		 * Remove this test to allow maximum flexibility or
1176 		 * restrict FSETOWN to the current process or process
1177 		 * group for maximum safety.
1178 		 */
1179 		if (proc->p_session != curthread->td_proc->p_session) {
1180 			PROC_UNLOCK(proc);
1181 			ret = EPERM;
1182 			goto fail;
1183 		}
1184 
1185 		sigio->sio_proc = proc;
1186 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1187 		PROC_UNLOCK(proc);
1188 	} else /* if (pgid < 0) */ {
1189 		pgrp = pgfind(-pgid);
1190 		if (pgrp == NULL) {
1191 			ret = ESRCH;
1192 			goto fail;
1193 		}
1194 
1195 		/*
1196 		 * Policy - Don't allow a process to FSETOWN a process
1197 		 * in another session.
1198 		 *
1199 		 * Remove this test to allow maximum flexibility or
1200 		 * restrict FSETOWN to the current process or process
1201 		 * group for maximum safety.
1202 		 */
1203 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1204 			PGRP_UNLOCK(pgrp);
1205 			ret = EPERM;
1206 			goto fail;
1207 		}
1208 
1209 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1210 		sigio->sio_pgrp = pgrp;
1211 		PGRP_UNLOCK(pgrp);
1212 	}
1213 	sx_sunlock(&proctree_lock);
1214 	*sigiop = sigio;
1215 	SIGIO_UNLOCK();
1216 	if (osigio != NULL)
1217 		sigiofree(osigio);
1218 	return (0);
1219 
1220 fail:
1221 	SIGIO_UNLOCK();
1222 	sx_sunlock(&proctree_lock);
1223 	sigiofree(sigio);
1224 	if (osigio != NULL)
1225 		sigiofree(osigio);
1226 	return (ret);
1227 }
1228 
1229 /*
1230  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1231  */
1232 pid_t
1233 fgetown(struct sigio **sigiop)
1234 {
1235 	pid_t pgid;
1236 
1237 	SIGIO_LOCK();
1238 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1239 	SIGIO_UNLOCK();
1240 	return (pgid);
1241 }
1242 
1243 static int
1244 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1245     bool audit)
1246 {
1247 	int error;
1248 
1249 	FILEDESC_XLOCK_ASSERT(fdp);
1250 
1251 	/*
1252 	 * We now hold the fp reference that used to be owned by the
1253 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1254 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1255 	 * added, and deleteing a knote for the new fd.
1256 	 */
1257 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1258 		knote_fdclose(td, fd);
1259 
1260 	/*
1261 	 * We need to notify mqueue if the object is of type mqueue.
1262 	 */
1263 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1264 		mq_fdclose(td, fd, fp);
1265 	FILEDESC_XUNLOCK(fdp);
1266 
1267 #ifdef AUDIT
1268 	if (AUDITING_TD(td) && audit)
1269 		audit_sysclose(td, fd, fp);
1270 #endif
1271 	error = closef(fp, td);
1272 
1273 	/*
1274 	 * All paths leading up to closefp() will have already removed or
1275 	 * replaced the fd in the filedesc table, so a restart would not
1276 	 * operate on the same file.
1277 	 */
1278 	if (error == ERESTART)
1279 		error = EINTR;
1280 
1281 	return (error);
1282 }
1283 
1284 static int
1285 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1286     bool holdleaders, bool audit)
1287 {
1288 	int error;
1289 
1290 	FILEDESC_XLOCK_ASSERT(fdp);
1291 
1292 	if (holdleaders) {
1293 		if (td->td_proc->p_fdtol != NULL) {
1294 			/*
1295 			 * Ask fdfree() to sleep to ensure that all relevant
1296 			 * process leaders can be traversed in closef().
1297 			 */
1298 			fdp->fd_holdleaderscount++;
1299 		} else {
1300 			holdleaders = false;
1301 		}
1302 	}
1303 
1304 	error = closefp_impl(fdp, fd, fp, td, audit);
1305 	if (holdleaders) {
1306 		FILEDESC_XLOCK(fdp);
1307 		fdp->fd_holdleaderscount--;
1308 		if (fdp->fd_holdleaderscount == 0 &&
1309 		    fdp->fd_holdleaderswakeup != 0) {
1310 			fdp->fd_holdleaderswakeup = 0;
1311 			wakeup(&fdp->fd_holdleaderscount);
1312 		}
1313 		FILEDESC_XUNLOCK(fdp);
1314 	}
1315 	return (error);
1316 }
1317 
1318 static int
1319 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1320     bool holdleaders, bool audit)
1321 {
1322 
1323 	FILEDESC_XLOCK_ASSERT(fdp);
1324 
1325 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1326 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1327 	} else {
1328 		return (closefp_impl(fdp, fd, fp, td, audit));
1329 	}
1330 }
1331 
1332 /*
1333  * Close a file descriptor.
1334  */
1335 #ifndef _SYS_SYSPROTO_H_
1336 struct close_args {
1337 	int     fd;
1338 };
1339 #endif
1340 /* ARGSUSED */
1341 int
1342 sys_close(struct thread *td, struct close_args *uap)
1343 {
1344 
1345 	return (kern_close(td, uap->fd));
1346 }
1347 
1348 int
1349 kern_close(struct thread *td, int fd)
1350 {
1351 	struct filedesc *fdp;
1352 	struct file *fp;
1353 
1354 	fdp = td->td_proc->p_fd;
1355 
1356 	FILEDESC_XLOCK(fdp);
1357 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1358 		FILEDESC_XUNLOCK(fdp);
1359 		return (EBADF);
1360 	}
1361 	fdfree(fdp, fd);
1362 
1363 	/* closefp() drops the FILEDESC lock for us. */
1364 	return (closefp(fdp, fd, fp, td, true, true));
1365 }
1366 
1367 int
1368 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1369 {
1370 	struct filedesc *fdp;
1371 	const struct fdescenttbl *fdt;
1372 	struct file *fp;
1373 	int fd;
1374 
1375 	/*
1376 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1377 	 * open should not be a usage error.  From a close_range() perspective,
1378 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1379 	 * be a usage error as all fd above 3 are in-fact already closed.
1380 	 */
1381 	if (highfd < lowfd) {
1382 		return (EINVAL);
1383 	}
1384 
1385 	fdp = td->td_proc->p_fd;
1386 	FILEDESC_XLOCK(fdp);
1387 	fdt = atomic_load_ptr(&fdp->fd_files);
1388 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1389 	fd = lowfd;
1390 	if (__predict_false(fd > highfd)) {
1391 		goto out_locked;
1392 	}
1393 	for (;;) {
1394 		fp = fdt->fdt_ofiles[fd].fde_file;
1395 		if (fp == NULL) {
1396 			if (fd == highfd)
1397 				goto out_locked;
1398 		} else {
1399 			fdfree(fdp, fd);
1400 			(void) closefp(fdp, fd, fp, td, true, true);
1401 			if (fd == highfd)
1402 				goto out_unlocked;
1403 			FILEDESC_XLOCK(fdp);
1404 			fdt = atomic_load_ptr(&fdp->fd_files);
1405 		}
1406 		fd++;
1407 	}
1408 out_locked:
1409 	FILEDESC_XUNLOCK(fdp);
1410 out_unlocked:
1411 	return (0);
1412 }
1413 
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct close_range_args {
1416 	u_int	lowfd;
1417 	u_int	highfd;
1418 	int	flags;
1419 };
1420 #endif
1421 int
1422 sys_close_range(struct thread *td, struct close_range_args *uap)
1423 {
1424 
1425 	/* No flags currently defined */
1426 	if (uap->flags != 0)
1427 		return (EINVAL);
1428 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1429 }
1430 
1431 #ifdef COMPAT_FREEBSD12
1432 /*
1433  * Close open file descriptors.
1434  */
1435 #ifndef _SYS_SYSPROTO_H_
1436 struct freebsd12_closefrom_args {
1437 	int	lowfd;
1438 };
1439 #endif
1440 /* ARGSUSED */
1441 int
1442 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1443 {
1444 	u_int lowfd;
1445 
1446 	AUDIT_ARG_FD(uap->lowfd);
1447 
1448 	/*
1449 	 * Treat negative starting file descriptor values identical to
1450 	 * closefrom(0) which closes all files.
1451 	 */
1452 	lowfd = MAX(0, uap->lowfd);
1453 	return (kern_close_range(td, lowfd, ~0U));
1454 }
1455 #endif	/* COMPAT_FREEBSD12 */
1456 
1457 #if defined(COMPAT_43)
1458 /*
1459  * Return status information about a file descriptor.
1460  */
1461 #ifndef _SYS_SYSPROTO_H_
1462 struct ofstat_args {
1463 	int	fd;
1464 	struct	ostat *sb;
1465 };
1466 #endif
1467 /* ARGSUSED */
1468 int
1469 ofstat(struct thread *td, struct ofstat_args *uap)
1470 {
1471 	struct ostat oub;
1472 	struct stat ub;
1473 	int error;
1474 
1475 	error = kern_fstat(td, uap->fd, &ub);
1476 	if (error == 0) {
1477 		cvtstat(&ub, &oub);
1478 		error = copyout(&oub, uap->sb, sizeof(oub));
1479 	}
1480 	return (error);
1481 }
1482 #endif /* COMPAT_43 */
1483 
1484 #if defined(COMPAT_FREEBSD11)
1485 int
1486 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1487 {
1488 	struct stat sb;
1489 	struct freebsd11_stat osb;
1490 	int error;
1491 
1492 	error = kern_fstat(td, uap->fd, &sb);
1493 	if (error != 0)
1494 		return (error);
1495 	error = freebsd11_cvtstat(&sb, &osb);
1496 	if (error == 0)
1497 		error = copyout(&osb, uap->sb, sizeof(osb));
1498 	return (error);
1499 }
1500 #endif	/* COMPAT_FREEBSD11 */
1501 
1502 /*
1503  * Return status information about a file descriptor.
1504  */
1505 #ifndef _SYS_SYSPROTO_H_
1506 struct fstat_args {
1507 	int	fd;
1508 	struct	stat *sb;
1509 };
1510 #endif
1511 /* ARGSUSED */
1512 int
1513 sys_fstat(struct thread *td, struct fstat_args *uap)
1514 {
1515 	struct stat ub;
1516 	int error;
1517 
1518 	error = kern_fstat(td, uap->fd, &ub);
1519 	if (error == 0)
1520 		error = copyout(&ub, uap->sb, sizeof(ub));
1521 	return (error);
1522 }
1523 
1524 int
1525 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1526 {
1527 	struct file *fp;
1528 	int error;
1529 
1530 	AUDIT_ARG_FD(fd);
1531 
1532 	error = fget(td, fd, &cap_fstat_rights, &fp);
1533 	if (__predict_false(error != 0))
1534 		return (error);
1535 
1536 	AUDIT_ARG_FILE(td->td_proc, fp);
1537 
1538 	error = fo_stat(fp, sbp, td->td_ucred, td);
1539 	fdrop(fp, td);
1540 #ifdef __STAT_TIME_T_EXT
1541 	sbp->st_atim_ext = 0;
1542 	sbp->st_mtim_ext = 0;
1543 	sbp->st_ctim_ext = 0;
1544 	sbp->st_btim_ext = 0;
1545 #endif
1546 #ifdef KTRACE
1547 	if (KTRPOINT(td, KTR_STRUCT))
1548 		ktrstat_error(sbp, error);
1549 #endif
1550 	return (error);
1551 }
1552 
1553 #if defined(COMPAT_FREEBSD11)
1554 /*
1555  * Return status information about a file descriptor.
1556  */
1557 #ifndef _SYS_SYSPROTO_H_
1558 struct freebsd11_nfstat_args {
1559 	int	fd;
1560 	struct	nstat *sb;
1561 };
1562 #endif
1563 /* ARGSUSED */
1564 int
1565 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1566 {
1567 	struct nstat nub;
1568 	struct stat ub;
1569 	int error;
1570 
1571 	error = kern_fstat(td, uap->fd, &ub);
1572 	if (error == 0) {
1573 		freebsd11_cvtnstat(&ub, &nub);
1574 		error = copyout(&nub, uap->sb, sizeof(nub));
1575 	}
1576 	return (error);
1577 }
1578 #endif /* COMPAT_FREEBSD11 */
1579 
1580 /*
1581  * Return pathconf information about a file descriptor.
1582  */
1583 #ifndef _SYS_SYSPROTO_H_
1584 struct fpathconf_args {
1585 	int	fd;
1586 	int	name;
1587 };
1588 #endif
1589 /* ARGSUSED */
1590 int
1591 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1592 {
1593 	long value;
1594 	int error;
1595 
1596 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1597 	if (error == 0)
1598 		td->td_retval[0] = value;
1599 	return (error);
1600 }
1601 
1602 int
1603 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1604 {
1605 	struct file *fp;
1606 	struct vnode *vp;
1607 	int error;
1608 
1609 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1610 	if (error != 0)
1611 		return (error);
1612 
1613 	if (name == _PC_ASYNC_IO) {
1614 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1615 		goto out;
1616 	}
1617 	vp = fp->f_vnode;
1618 	if (vp != NULL) {
1619 		vn_lock(vp, LK_SHARED | LK_RETRY);
1620 		error = VOP_PATHCONF(vp, name, valuep);
1621 		VOP_UNLOCK(vp);
1622 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1623 		if (name != _PC_PIPE_BUF) {
1624 			error = EINVAL;
1625 		} else {
1626 			*valuep = PIPE_BUF;
1627 			error = 0;
1628 		}
1629 	} else {
1630 		error = EOPNOTSUPP;
1631 	}
1632 out:
1633 	fdrop(fp, td);
1634 	return (error);
1635 }
1636 
1637 /*
1638  * Copy filecaps structure allocating memory for ioctls array if needed.
1639  *
1640  * The last parameter indicates whether the fdtable is locked. If it is not and
1641  * ioctls are encountered, copying fails and the caller must lock the table.
1642  *
1643  * Note that if the table was not locked, the caller has to check the relevant
1644  * sequence counter to determine whether the operation was successful.
1645  */
1646 bool
1647 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1648 {
1649 	size_t size;
1650 
1651 	if (src->fc_ioctls != NULL && !locked)
1652 		return (false);
1653 	memcpy(dst, src, sizeof(*src));
1654 	if (src->fc_ioctls == NULL)
1655 		return (true);
1656 
1657 	KASSERT(src->fc_nioctls > 0,
1658 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1659 
1660 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1661 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1662 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1663 	return (true);
1664 }
1665 
1666 static u_long *
1667 filecaps_copy_prep(const struct filecaps *src)
1668 {
1669 	u_long *ioctls;
1670 	size_t size;
1671 
1672 	if (__predict_true(src->fc_ioctls == NULL))
1673 		return (NULL);
1674 
1675 	KASSERT(src->fc_nioctls > 0,
1676 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1677 
1678 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1679 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1680 	return (ioctls);
1681 }
1682 
1683 static void
1684 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1685     u_long *ioctls)
1686 {
1687 	size_t size;
1688 
1689 	*dst = *src;
1690 	if (__predict_true(src->fc_ioctls == NULL)) {
1691 		MPASS(ioctls == NULL);
1692 		return;
1693 	}
1694 
1695 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1696 	dst->fc_ioctls = ioctls;
1697 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1698 }
1699 
1700 /*
1701  * Move filecaps structure to the new place and clear the old place.
1702  */
1703 void
1704 filecaps_move(struct filecaps *src, struct filecaps *dst)
1705 {
1706 
1707 	*dst = *src;
1708 	bzero(src, sizeof(*src));
1709 }
1710 
1711 /*
1712  * Fill the given filecaps structure with full rights.
1713  */
1714 static void
1715 filecaps_fill(struct filecaps *fcaps)
1716 {
1717 
1718 	CAP_ALL(&fcaps->fc_rights);
1719 	fcaps->fc_ioctls = NULL;
1720 	fcaps->fc_nioctls = -1;
1721 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1722 }
1723 
1724 /*
1725  * Free memory allocated within filecaps structure.
1726  */
1727 void
1728 filecaps_free(struct filecaps *fcaps)
1729 {
1730 
1731 	free(fcaps->fc_ioctls, M_FILECAPS);
1732 	bzero(fcaps, sizeof(*fcaps));
1733 }
1734 
1735 static u_long *
1736 filecaps_free_prep(struct filecaps *fcaps)
1737 {
1738 	u_long *ioctls;
1739 
1740 	ioctls = fcaps->fc_ioctls;
1741 	bzero(fcaps, sizeof(*fcaps));
1742 	return (ioctls);
1743 }
1744 
1745 static void
1746 filecaps_free_finish(u_long *ioctls)
1747 {
1748 
1749 	free(ioctls, M_FILECAPS);
1750 }
1751 
1752 /*
1753  * Validate the given filecaps structure.
1754  */
1755 static void
1756 filecaps_validate(const struct filecaps *fcaps, const char *func)
1757 {
1758 
1759 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1760 	    ("%s: invalid rights", func));
1761 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1762 	    ("%s: invalid fcntls", func));
1763 	KASSERT(fcaps->fc_fcntls == 0 ||
1764 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1765 	    ("%s: fcntls without CAP_FCNTL", func));
1766 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1767 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1768 	    ("%s: invalid ioctls", func));
1769 	KASSERT(fcaps->fc_nioctls == 0 ||
1770 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1771 	    ("%s: ioctls without CAP_IOCTL", func));
1772 }
1773 
1774 static void
1775 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1776 {
1777 	int nfd1;
1778 
1779 	FILEDESC_XLOCK_ASSERT(fdp);
1780 
1781 	nfd1 = fdp->fd_nfiles * 2;
1782 	if (nfd1 < nfd)
1783 		nfd1 = nfd;
1784 	fdgrowtable(fdp, nfd1);
1785 }
1786 
1787 /*
1788  * Grow the file table to accommodate (at least) nfd descriptors.
1789  */
1790 static void
1791 fdgrowtable(struct filedesc *fdp, int nfd)
1792 {
1793 	struct filedesc0 *fdp0;
1794 	struct freetable *ft;
1795 	struct fdescenttbl *ntable;
1796 	struct fdescenttbl *otable;
1797 	int nnfiles, onfiles;
1798 	NDSLOTTYPE *nmap, *omap;
1799 
1800 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1801 
1802 	/* save old values */
1803 	onfiles = fdp->fd_nfiles;
1804 	otable = fdp->fd_files;
1805 	omap = fdp->fd_map;
1806 
1807 	/* compute the size of the new table */
1808 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1809 	if (nnfiles <= onfiles)
1810 		/* the table is already large enough */
1811 		return;
1812 
1813 	/*
1814 	 * Allocate a new table.  We need enough space for the number of
1815 	 * entries, file entries themselves and the struct freetable we will use
1816 	 * when we decommission the table and place it on the freelist.
1817 	 * We place the struct freetable in the middle so we don't have
1818 	 * to worry about padding.
1819 	 */
1820 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1821 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1822 	    sizeof(struct freetable),
1823 	    M_FILEDESC, M_ZERO | M_WAITOK);
1824 	/* copy the old data */
1825 	ntable->fdt_nfiles = nnfiles;
1826 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1827 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1828 
1829 	/*
1830 	 * Allocate a new map only if the old is not large enough.  It will
1831 	 * grow at a slower rate than the table as it can map more
1832 	 * entries than the table can hold.
1833 	 */
1834 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1835 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1836 		    M_ZERO | M_WAITOK);
1837 		/* copy over the old data and update the pointer */
1838 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1839 		fdp->fd_map = nmap;
1840 	}
1841 
1842 	/*
1843 	 * Make sure that ntable is correctly initialized before we replace
1844 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1845 	 * data.
1846 	 */
1847 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1848 
1849 	/*
1850 	 * Free the old file table when not shared by other threads or processes.
1851 	 * The old file table is considered to be shared when either are true:
1852 	 * - The process has more than one thread.
1853 	 * - The file descriptor table has been shared via fdshare().
1854 	 *
1855 	 * When shared, the old file table will be placed on a freelist
1856 	 * which will be processed when the struct filedesc is released.
1857 	 *
1858 	 * Note that if onfiles == NDFILE, we're dealing with the original
1859 	 * static allocation contained within (struct filedesc0 *)fdp,
1860 	 * which must not be freed.
1861 	 */
1862 	if (onfiles > NDFILE) {
1863 		if (curproc->p_numthreads == 1 &&
1864 		    refcount_load(&fdp->fd_refcnt) == 1)
1865 			free(otable, M_FILEDESC);
1866 		else {
1867 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1868 			fdp0 = (struct filedesc0 *)fdp;
1869 			ft->ft_table = otable;
1870 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1871 		}
1872 	}
1873 	/*
1874 	 * The map does not have the same possibility of threads still
1875 	 * holding references to it.  So always free it as long as it
1876 	 * does not reference the original static allocation.
1877 	 */
1878 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1879 		free(omap, M_FILEDESC);
1880 }
1881 
1882 /*
1883  * Allocate a file descriptor for the process.
1884  */
1885 int
1886 fdalloc(struct thread *td, int minfd, int *result)
1887 {
1888 	struct proc *p = td->td_proc;
1889 	struct filedesc *fdp = p->p_fd;
1890 	int fd, maxfd, allocfd;
1891 #ifdef RACCT
1892 	int error;
1893 #endif
1894 
1895 	FILEDESC_XLOCK_ASSERT(fdp);
1896 
1897 	if (fdp->fd_freefile > minfd)
1898 		minfd = fdp->fd_freefile;
1899 
1900 	maxfd = getmaxfd(td);
1901 
1902 	/*
1903 	 * Search the bitmap for a free descriptor starting at minfd.
1904 	 * If none is found, grow the file table.
1905 	 */
1906 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1907 	if (__predict_false(fd >= maxfd))
1908 		return (EMFILE);
1909 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1910 		allocfd = min(fd * 2, maxfd);
1911 #ifdef RACCT
1912 		if (RACCT_ENABLED()) {
1913 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1914 			if (error != 0)
1915 				return (EMFILE);
1916 		}
1917 #endif
1918 		/*
1919 		 * fd is already equal to first free descriptor >= minfd, so
1920 		 * we only need to grow the table and we are done.
1921 		 */
1922 		fdgrowtable_exp(fdp, allocfd);
1923 	}
1924 
1925 	/*
1926 	 * Perform some sanity checks, then mark the file descriptor as
1927 	 * used and return it to the caller.
1928 	 */
1929 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1930 	    ("invalid descriptor %d", fd));
1931 	KASSERT(!fdisused(fdp, fd),
1932 	    ("fd_first_free() returned non-free descriptor"));
1933 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1934 	    ("file descriptor isn't free"));
1935 	fdused(fdp, fd);
1936 	*result = fd;
1937 	return (0);
1938 }
1939 
1940 /*
1941  * Allocate n file descriptors for the process.
1942  */
1943 int
1944 fdallocn(struct thread *td, int minfd, int *fds, int n)
1945 {
1946 	struct proc *p = td->td_proc;
1947 	struct filedesc *fdp = p->p_fd;
1948 	int i;
1949 
1950 	FILEDESC_XLOCK_ASSERT(fdp);
1951 
1952 	for (i = 0; i < n; i++)
1953 		if (fdalloc(td, 0, &fds[i]) != 0)
1954 			break;
1955 
1956 	if (i < n) {
1957 		for (i--; i >= 0; i--)
1958 			fdunused(fdp, fds[i]);
1959 		return (EMFILE);
1960 	}
1961 
1962 	return (0);
1963 }
1964 
1965 /*
1966  * Create a new open file structure and allocate a file descriptor for the
1967  * process that refers to it.  We add one reference to the file for the
1968  * descriptor table and one reference for resultfp. This is to prevent us
1969  * being preempted and the entry in the descriptor table closed after we
1970  * release the FILEDESC lock.
1971  */
1972 int
1973 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1974     struct filecaps *fcaps)
1975 {
1976 	struct file *fp;
1977 	int error, fd;
1978 
1979 	MPASS(resultfp != NULL);
1980 	MPASS(resultfd != NULL);
1981 
1982 	error = _falloc_noinstall(td, &fp, 2);
1983 	if (__predict_false(error != 0)) {
1984 		return (error);
1985 	}
1986 
1987 	error = finstall_refed(td, fp, &fd, flags, fcaps);
1988 	if (__predict_false(error != 0)) {
1989 		falloc_abort(td, fp);
1990 		return (error);
1991 	}
1992 
1993 	*resultfp = fp;
1994 	*resultfd = fd;
1995 
1996 	return (0);
1997 }
1998 
1999 /*
2000  * Create a new open file structure without allocating a file descriptor.
2001  */
2002 int
2003 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2004 {
2005 	struct file *fp;
2006 	int maxuserfiles = maxfiles - (maxfiles / 20);
2007 	int openfiles_new;
2008 	static struct timeval lastfail;
2009 	static int curfail;
2010 
2011 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2012 	MPASS(n > 0);
2013 
2014 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2015 	if ((openfiles_new >= maxuserfiles &&
2016 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2017 	    openfiles_new >= maxfiles) {
2018 		atomic_subtract_int(&openfiles, 1);
2019 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2020 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2021 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2022 		}
2023 		return (ENFILE);
2024 	}
2025 	fp = uma_zalloc(file_zone, M_WAITOK);
2026 	bzero(fp, sizeof(*fp));
2027 	refcount_init(&fp->f_count, n);
2028 	fp->f_cred = crhold(td->td_ucred);
2029 	fp->f_ops = &badfileops;
2030 	*resultfp = fp;
2031 	return (0);
2032 }
2033 
2034 void
2035 falloc_abort(struct thread *td, struct file *fp)
2036 {
2037 
2038 	/*
2039 	 * For assertion purposes.
2040 	 */
2041 	refcount_init(&fp->f_count, 0);
2042 	_fdrop(fp, td);
2043 }
2044 
2045 /*
2046  * Install a file in a file descriptor table.
2047  */
2048 void
2049 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2050     struct filecaps *fcaps)
2051 {
2052 	struct filedescent *fde;
2053 
2054 	MPASS(fp != NULL);
2055 	if (fcaps != NULL)
2056 		filecaps_validate(fcaps, __func__);
2057 	FILEDESC_XLOCK_ASSERT(fdp);
2058 
2059 	fde = &fdp->fd_ofiles[fd];
2060 #ifdef CAPABILITIES
2061 	seqc_write_begin(&fde->fde_seqc);
2062 #endif
2063 	fde->fde_file = fp;
2064 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2065 	if (fcaps != NULL)
2066 		filecaps_move(fcaps, &fde->fde_caps);
2067 	else
2068 		filecaps_fill(&fde->fde_caps);
2069 #ifdef CAPABILITIES
2070 	seqc_write_end(&fde->fde_seqc);
2071 #endif
2072 }
2073 
2074 int
2075 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2076     struct filecaps *fcaps)
2077 {
2078 	struct filedesc *fdp = td->td_proc->p_fd;
2079 	int error;
2080 
2081 	MPASS(fd != NULL);
2082 
2083 	FILEDESC_XLOCK(fdp);
2084 	error = fdalloc(td, 0, fd);
2085 	if (__predict_true(error == 0)) {
2086 		_finstall(fdp, fp, *fd, flags, fcaps);
2087 	}
2088 	FILEDESC_XUNLOCK(fdp);
2089 	return (error);
2090 }
2091 
2092 int
2093 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2094     struct filecaps *fcaps)
2095 {
2096 	int error;
2097 
2098 	MPASS(fd != NULL);
2099 
2100 	if (!fhold(fp))
2101 		return (EBADF);
2102 	error = finstall_refed(td, fp, fd, flags, fcaps);
2103 	if (__predict_false(error != 0)) {
2104 		fdrop(fp, td);
2105 	}
2106 	return (error);
2107 }
2108 
2109 /*
2110  * Build a new filedesc structure from another.
2111  *
2112  * If fdp is not NULL, return with it shared locked.
2113  */
2114 struct filedesc *
2115 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2116 {
2117 	struct filedesc0 *newfdp0;
2118 	struct filedesc *newfdp;
2119 
2120 	if (prepfiles)
2121 		MPASS(lastfile != NULL);
2122 	else
2123 		MPASS(lastfile == NULL);
2124 
2125 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2126 	newfdp = &newfdp0->fd_fd;
2127 
2128 	/* Create the file descriptor table. */
2129 	FILEDESC_LOCK_INIT(newfdp);
2130 	refcount_init(&newfdp->fd_refcnt, 1);
2131 	refcount_init(&newfdp->fd_holdcnt, 1);
2132 	newfdp->fd_map = newfdp0->fd_dmap;
2133 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2134 	newfdp->fd_files->fdt_nfiles = NDFILE;
2135 
2136 	if (fdp == NULL)
2137 		return (newfdp);
2138 
2139 	FILEDESC_SLOCK(fdp);
2140 	if (!prepfiles) {
2141 		FILEDESC_SUNLOCK(fdp);
2142 		return (newfdp);
2143 	}
2144 
2145 	for (;;) {
2146 		*lastfile = fdlastfile(fdp);
2147 		if (*lastfile < newfdp->fd_nfiles)
2148 			break;
2149 		FILEDESC_SUNLOCK(fdp);
2150 		fdgrowtable(newfdp, *lastfile + 1);
2151 		FILEDESC_SLOCK(fdp);
2152 	}
2153 
2154 	return (newfdp);
2155 }
2156 
2157 /*
2158  * Build a pwddesc structure from another.
2159  * Copy the current, root, and jail root vnode references.
2160  *
2161  * If pdp is not NULL, return with it shared locked.
2162  */
2163 struct pwddesc *
2164 pdinit(struct pwddesc *pdp, bool keeplock)
2165 {
2166 	struct pwddesc *newpdp;
2167 	struct pwd *newpwd;
2168 
2169 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2170 
2171 	PWDDESC_LOCK_INIT(newpdp);
2172 	refcount_init(&newpdp->pd_refcount, 1);
2173 	newpdp->pd_cmask = CMASK;
2174 
2175 	if (pdp == NULL) {
2176 		newpwd = pwd_alloc();
2177 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2178 		return (newpdp);
2179 	}
2180 
2181 	PWDDESC_XLOCK(pdp);
2182 	newpwd = pwd_hold_pwddesc(pdp);
2183 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2184 	if (!keeplock)
2185 		PWDDESC_XUNLOCK(pdp);
2186 	return (newpdp);
2187 }
2188 
2189 static struct filedesc *
2190 fdhold(struct proc *p)
2191 {
2192 	struct filedesc *fdp;
2193 
2194 	PROC_LOCK_ASSERT(p, MA_OWNED);
2195 	fdp = p->p_fd;
2196 	if (fdp != NULL)
2197 		refcount_acquire(&fdp->fd_holdcnt);
2198 	return (fdp);
2199 }
2200 
2201 static struct pwddesc *
2202 pdhold(struct proc *p)
2203 {
2204 	struct pwddesc *pdp;
2205 
2206 	PROC_LOCK_ASSERT(p, MA_OWNED);
2207 	pdp = p->p_pd;
2208 	if (pdp != NULL)
2209 		refcount_acquire(&pdp->pd_refcount);
2210 	return (pdp);
2211 }
2212 
2213 static void
2214 fddrop(struct filedesc *fdp)
2215 {
2216 
2217 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2218 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2219 			return;
2220 	}
2221 
2222 	FILEDESC_LOCK_DESTROY(fdp);
2223 	uma_zfree(filedesc0_zone, fdp);
2224 }
2225 
2226 static void
2227 pddrop(struct pwddesc *pdp)
2228 {
2229 	struct pwd *pwd;
2230 
2231 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2232 		return;
2233 
2234 	PWDDESC_XLOCK(pdp);
2235 	if (refcount_release(&pdp->pd_refcount) == 0) {
2236 		PWDDESC_XUNLOCK(pdp);
2237 		return;
2238 	}
2239 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2240 	pwd_set(pdp, NULL);
2241 	PWDDESC_XUNLOCK(pdp);
2242 	pwd_drop(pwd);
2243 
2244 	PWDDESC_LOCK_DESTROY(pdp);
2245 	free(pdp, M_PWDDESC);
2246 }
2247 
2248 /*
2249  * Share a filedesc structure.
2250  */
2251 struct filedesc *
2252 fdshare(struct filedesc *fdp)
2253 {
2254 
2255 	refcount_acquire(&fdp->fd_refcnt);
2256 	return (fdp);
2257 }
2258 
2259 /*
2260  * Share a pwddesc structure.
2261  */
2262 struct pwddesc *
2263 pdshare(struct pwddesc *pdp)
2264 {
2265 	refcount_acquire(&pdp->pd_refcount);
2266 	return (pdp);
2267 }
2268 
2269 /*
2270  * Unshare a filedesc structure, if necessary by making a copy
2271  */
2272 void
2273 fdunshare(struct thread *td)
2274 {
2275 	struct filedesc *tmp;
2276 	struct proc *p = td->td_proc;
2277 
2278 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2279 		return;
2280 
2281 	tmp = fdcopy(p->p_fd);
2282 	fdescfree(td);
2283 	p->p_fd = tmp;
2284 }
2285 
2286 /*
2287  * Unshare a pwddesc structure.
2288  */
2289 void
2290 pdunshare(struct thread *td)
2291 {
2292 	struct pwddesc *pdp;
2293 	struct proc *p;
2294 
2295 	p = td->td_proc;
2296 	/* Not shared. */
2297 	if (p->p_pd->pd_refcount == 1)
2298 		return;
2299 
2300 	pdp = pdcopy(p->p_pd);
2301 	pdescfree(td);
2302 	p->p_pd = pdp;
2303 }
2304 
2305 void
2306 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2307 {
2308 
2309 	fdescfree(td);
2310 	td->td_proc->p_fd = fdp;
2311 }
2312 
2313 /*
2314  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2315  * this is to ease callers, not catch errors.
2316  */
2317 struct filedesc *
2318 fdcopy(struct filedesc *fdp)
2319 {
2320 	struct filedesc *newfdp;
2321 	struct filedescent *nfde, *ofde;
2322 	int i, lastfile;
2323 
2324 	MPASS(fdp != NULL);
2325 
2326 	newfdp = fdinit(fdp, true, &lastfile);
2327 	/* copy all passable descriptors (i.e. not kqueue) */
2328 	newfdp->fd_freefile = -1;
2329 	for (i = 0; i <= lastfile; ++i) {
2330 		ofde = &fdp->fd_ofiles[i];
2331 		if (ofde->fde_file == NULL ||
2332 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2333 		    !fhold(ofde->fde_file)) {
2334 			if (newfdp->fd_freefile == -1)
2335 				newfdp->fd_freefile = i;
2336 			continue;
2337 		}
2338 		nfde = &newfdp->fd_ofiles[i];
2339 		*nfde = *ofde;
2340 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2341 		fdused_init(newfdp, i);
2342 	}
2343 	if (newfdp->fd_freefile == -1)
2344 		newfdp->fd_freefile = i;
2345 	FILEDESC_SUNLOCK(fdp);
2346 	return (newfdp);
2347 }
2348 
2349 /*
2350  * Copy a pwddesc structure.
2351  */
2352 struct pwddesc *
2353 pdcopy(struct pwddesc *pdp)
2354 {
2355 	struct pwddesc *newpdp;
2356 
2357 	MPASS(pdp != NULL);
2358 
2359 	newpdp = pdinit(pdp, true);
2360 	newpdp->pd_cmask = pdp->pd_cmask;
2361 	PWDDESC_XUNLOCK(pdp);
2362 	return (newpdp);
2363 }
2364 
2365 /*
2366  * Copies a filedesc structure, while remapping all file descriptors
2367  * stored inside using a translation table.
2368  *
2369  * File descriptors are copied over to the new file descriptor table,
2370  * regardless of whether the close-on-exec flag is set.
2371  */
2372 int
2373 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2374     struct filedesc **ret)
2375 {
2376 	struct filedesc *newfdp;
2377 	struct filedescent *nfde, *ofde;
2378 	int error, i, lastfile;
2379 
2380 	MPASS(fdp != NULL);
2381 
2382 	newfdp = fdinit(fdp, true, &lastfile);
2383 	if (nfds > lastfile + 1) {
2384 		/* New table cannot be larger than the old one. */
2385 		error = E2BIG;
2386 		goto bad;
2387 	}
2388 	/* Copy all passable descriptors (i.e. not kqueue). */
2389 	newfdp->fd_freefile = nfds;
2390 	for (i = 0; i < nfds; ++i) {
2391 		if (fds[i] < 0 || fds[i] > lastfile) {
2392 			/* File descriptor out of bounds. */
2393 			error = EBADF;
2394 			goto bad;
2395 		}
2396 		ofde = &fdp->fd_ofiles[fds[i]];
2397 		if (ofde->fde_file == NULL) {
2398 			/* Unused file descriptor. */
2399 			error = EBADF;
2400 			goto bad;
2401 		}
2402 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2403 			/* File descriptor cannot be passed. */
2404 			error = EINVAL;
2405 			goto bad;
2406 		}
2407 		if (!fhold(ofde->fde_file)) {
2408 			error = EBADF;
2409 			goto bad;
2410 		}
2411 		nfde = &newfdp->fd_ofiles[i];
2412 		*nfde = *ofde;
2413 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2414 		fdused_init(newfdp, i);
2415 	}
2416 	FILEDESC_SUNLOCK(fdp);
2417 	*ret = newfdp;
2418 	return (0);
2419 bad:
2420 	FILEDESC_SUNLOCK(fdp);
2421 	fdescfree_remapped(newfdp);
2422 	return (error);
2423 }
2424 
2425 /*
2426  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2427  * one of processes using it exits) and the table used to be shared.
2428  */
2429 static void
2430 fdclearlocks(struct thread *td)
2431 {
2432 	struct filedesc *fdp;
2433 	struct filedesc_to_leader *fdtol;
2434 	struct flock lf;
2435 	struct file *fp;
2436 	struct proc *p;
2437 	struct vnode *vp;
2438 	int i, lastfile;
2439 
2440 	p = td->td_proc;
2441 	fdp = p->p_fd;
2442 	fdtol = p->p_fdtol;
2443 	MPASS(fdtol != NULL);
2444 
2445 	FILEDESC_XLOCK(fdp);
2446 	KASSERT(fdtol->fdl_refcount > 0,
2447 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2448 	    fdtol->fdl_refcount));
2449 	if (fdtol->fdl_refcount == 1 &&
2450 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2451 		lastfile = fdlastfile(fdp);
2452 		for (i = 0; i <= lastfile; i++) {
2453 			fp = fdp->fd_ofiles[i].fde_file;
2454 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2455 			    !fhold(fp))
2456 				continue;
2457 			FILEDESC_XUNLOCK(fdp);
2458 			lf.l_whence = SEEK_SET;
2459 			lf.l_start = 0;
2460 			lf.l_len = 0;
2461 			lf.l_type = F_UNLCK;
2462 			vp = fp->f_vnode;
2463 			(void) VOP_ADVLOCK(vp,
2464 			    (caddr_t)p->p_leader, F_UNLCK,
2465 			    &lf, F_POSIX);
2466 			FILEDESC_XLOCK(fdp);
2467 			fdrop(fp, td);
2468 		}
2469 	}
2470 retry:
2471 	if (fdtol->fdl_refcount == 1) {
2472 		if (fdp->fd_holdleaderscount > 0 &&
2473 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2474 			/*
2475 			 * close() or kern_dup() has cleared a reference
2476 			 * in a shared file descriptor table.
2477 			 */
2478 			fdp->fd_holdleaderswakeup = 1;
2479 			sx_sleep(&fdp->fd_holdleaderscount,
2480 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2481 			goto retry;
2482 		}
2483 		if (fdtol->fdl_holdcount > 0) {
2484 			/*
2485 			 * Ensure that fdtol->fdl_leader remains
2486 			 * valid in closef().
2487 			 */
2488 			fdtol->fdl_wakeup = 1;
2489 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2490 			    "fdlhold", 0);
2491 			goto retry;
2492 		}
2493 	}
2494 	fdtol->fdl_refcount--;
2495 	if (fdtol->fdl_refcount == 0 &&
2496 	    fdtol->fdl_holdcount == 0) {
2497 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2498 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2499 	} else
2500 		fdtol = NULL;
2501 	p->p_fdtol = NULL;
2502 	FILEDESC_XUNLOCK(fdp);
2503 	if (fdtol != NULL)
2504 		free(fdtol, M_FILEDESC_TO_LEADER);
2505 }
2506 
2507 /*
2508  * Release a filedesc structure.
2509  */
2510 static void
2511 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2512 {
2513 	struct filedesc0 *fdp0;
2514 	struct freetable *ft, *tft;
2515 	struct filedescent *fde;
2516 	struct file *fp;
2517 	int i, lastfile;
2518 
2519 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2520 	    ("%s: fd table %p carries references", __func__, fdp));
2521 
2522 	/*
2523 	 * Serialize with threads iterating over the table, if any.
2524 	 */
2525 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2526 		FILEDESC_XLOCK(fdp);
2527 		FILEDESC_XUNLOCK(fdp);
2528 	}
2529 
2530 	lastfile = fdlastfile_single(fdp);
2531 	for (i = 0; i <= lastfile; i++) {
2532 		fde = &fdp->fd_ofiles[i];
2533 		fp = fde->fde_file;
2534 		if (fp != NULL) {
2535 			fdefree_last(fde);
2536 			if (needclose)
2537 				(void) closef(fp, td);
2538 			else
2539 				fdrop(fp, td);
2540 		}
2541 	}
2542 
2543 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2544 		free(fdp->fd_map, M_FILEDESC);
2545 	if (fdp->fd_nfiles > NDFILE)
2546 		free(fdp->fd_files, M_FILEDESC);
2547 
2548 	fdp0 = (struct filedesc0 *)fdp;
2549 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2550 		free(ft->ft_table, M_FILEDESC);
2551 
2552 	fddrop(fdp);
2553 }
2554 
2555 void
2556 fdescfree(struct thread *td)
2557 {
2558 	struct proc *p;
2559 	struct filedesc *fdp;
2560 
2561 	p = td->td_proc;
2562 	fdp = p->p_fd;
2563 	MPASS(fdp != NULL);
2564 
2565 #ifdef RACCT
2566 	if (RACCT_ENABLED())
2567 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2568 #endif
2569 
2570 	if (p->p_fdtol != NULL)
2571 		fdclearlocks(td);
2572 
2573 	PROC_LOCK(p);
2574 	p->p_fd = NULL;
2575 	PROC_UNLOCK(p);
2576 
2577 	if (refcount_release(&fdp->fd_refcnt) == 0)
2578 		return;
2579 
2580 	fdescfree_fds(td, fdp, 1);
2581 }
2582 
2583 void
2584 pdescfree(struct thread *td)
2585 {
2586 	struct proc *p;
2587 	struct pwddesc *pdp;
2588 
2589 	p = td->td_proc;
2590 	pdp = p->p_pd;
2591 	MPASS(pdp != NULL);
2592 
2593 	PROC_LOCK(p);
2594 	p->p_pd = NULL;
2595 	PROC_UNLOCK(p);
2596 
2597 	pddrop(pdp);
2598 }
2599 
2600 void
2601 fdescfree_remapped(struct filedesc *fdp)
2602 {
2603 #ifdef INVARIANTS
2604 	/* fdescfree_fds() asserts that fd_refcnt == 0. */
2605 	if (!refcount_release(&fdp->fd_refcnt))
2606 		panic("%s: fd table %p has extra references", __func__, fdp);
2607 #endif
2608 	fdescfree_fds(curthread, fdp, 0);
2609 }
2610 
2611 /*
2612  * For setugid programs, we don't want to people to use that setugidness
2613  * to generate error messages which write to a file which otherwise would
2614  * otherwise be off-limits to the process.  We check for filesystems where
2615  * the vnode can change out from under us after execve (like [lin]procfs).
2616  *
2617  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2618  * sufficient.  We also don't check for setugidness since we know we are.
2619  */
2620 static bool
2621 is_unsafe(struct file *fp)
2622 {
2623 	struct vnode *vp;
2624 
2625 	if (fp->f_type != DTYPE_VNODE)
2626 		return (false);
2627 
2628 	vp = fp->f_vnode;
2629 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2630 }
2631 
2632 /*
2633  * Make this setguid thing safe, if at all possible.
2634  */
2635 void
2636 fdsetugidsafety(struct thread *td)
2637 {
2638 	struct filedesc *fdp;
2639 	struct file *fp;
2640 	int i;
2641 
2642 	fdp = td->td_proc->p_fd;
2643 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2644 	    ("the fdtable should not be shared"));
2645 	MPASS(fdp->fd_nfiles >= 3);
2646 	for (i = 0; i <= 2; i++) {
2647 		fp = fdp->fd_ofiles[i].fde_file;
2648 		if (fp != NULL && is_unsafe(fp)) {
2649 			FILEDESC_XLOCK(fdp);
2650 			knote_fdclose(td, i);
2651 			/*
2652 			 * NULL-out descriptor prior to close to avoid
2653 			 * a race while close blocks.
2654 			 */
2655 			fdfree(fdp, i);
2656 			FILEDESC_XUNLOCK(fdp);
2657 			(void) closef(fp, td);
2658 		}
2659 	}
2660 }
2661 
2662 /*
2663  * If a specific file object occupies a specific file descriptor, close the
2664  * file descriptor entry and drop a reference on the file object.  This is a
2665  * convenience function to handle a subsequent error in a function that calls
2666  * falloc() that handles the race that another thread might have closed the
2667  * file descriptor out from under the thread creating the file object.
2668  */
2669 void
2670 fdclose(struct thread *td, struct file *fp, int idx)
2671 {
2672 	struct filedesc *fdp = td->td_proc->p_fd;
2673 
2674 	FILEDESC_XLOCK(fdp);
2675 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2676 		fdfree(fdp, idx);
2677 		FILEDESC_XUNLOCK(fdp);
2678 		fdrop(fp, td);
2679 	} else
2680 		FILEDESC_XUNLOCK(fdp);
2681 }
2682 
2683 /*
2684  * Close any files on exec?
2685  */
2686 void
2687 fdcloseexec(struct thread *td)
2688 {
2689 	struct filedesc *fdp;
2690 	struct filedescent *fde;
2691 	struct file *fp;
2692 	int i, lastfile;
2693 
2694 	fdp = td->td_proc->p_fd;
2695 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2696 	    ("the fdtable should not be shared"));
2697 	lastfile = fdlastfile_single(fdp);
2698 	for (i = 0; i <= lastfile; i++) {
2699 		fde = &fdp->fd_ofiles[i];
2700 		fp = fde->fde_file;
2701 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2702 		    (fde->fde_flags & UF_EXCLOSE))) {
2703 			FILEDESC_XLOCK(fdp);
2704 			fdfree(fdp, i);
2705 			(void) closefp(fdp, i, fp, td, false, false);
2706 			FILEDESC_UNLOCK_ASSERT(fdp);
2707 		}
2708 	}
2709 }
2710 
2711 /*
2712  * It is unsafe for set[ug]id processes to be started with file
2713  * descriptors 0..2 closed, as these descriptors are given implicit
2714  * significance in the Standard C library.  fdcheckstd() will create a
2715  * descriptor referencing /dev/null for each of stdin, stdout, and
2716  * stderr that is not already open.
2717  */
2718 int
2719 fdcheckstd(struct thread *td)
2720 {
2721 	struct filedesc *fdp;
2722 	register_t save;
2723 	int i, error, devnull;
2724 
2725 	fdp = td->td_proc->p_fd;
2726 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2727 	    ("the fdtable should not be shared"));
2728 	MPASS(fdp->fd_nfiles >= 3);
2729 	devnull = -1;
2730 	for (i = 0; i <= 2; i++) {
2731 		if (fdp->fd_ofiles[i].fde_file != NULL)
2732 			continue;
2733 
2734 		save = td->td_retval[0];
2735 		if (devnull != -1) {
2736 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2737 		} else {
2738 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2739 			    UIO_SYSSPACE, O_RDWR, 0);
2740 			if (error == 0) {
2741 				devnull = td->td_retval[0];
2742 				KASSERT(devnull == i, ("we didn't get our fd"));
2743 			}
2744 		}
2745 		td->td_retval[0] = save;
2746 		if (error != 0)
2747 			return (error);
2748 	}
2749 	return (0);
2750 }
2751 
2752 /*
2753  * Internal form of close.  Decrement reference count on file structure.
2754  * Note: td may be NULL when closing a file that was being passed in a
2755  * message.
2756  */
2757 int
2758 closef(struct file *fp, struct thread *td)
2759 {
2760 	struct vnode *vp;
2761 	struct flock lf;
2762 	struct filedesc_to_leader *fdtol;
2763 	struct filedesc *fdp;
2764 
2765 	MPASS(td != NULL);
2766 
2767 	/*
2768 	 * POSIX record locking dictates that any close releases ALL
2769 	 * locks owned by this process.  This is handled by setting
2770 	 * a flag in the unlock to free ONLY locks obeying POSIX
2771 	 * semantics, and not to free BSD-style file locks.
2772 	 * If the descriptor was in a message, POSIX-style locks
2773 	 * aren't passed with the descriptor, and the thread pointer
2774 	 * will be NULL.  Callers should be careful only to pass a
2775 	 * NULL thread pointer when there really is no owning
2776 	 * context that might have locks, or the locks will be
2777 	 * leaked.
2778 	 */
2779 	if (fp->f_type == DTYPE_VNODE) {
2780 		vp = fp->f_vnode;
2781 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2782 			lf.l_whence = SEEK_SET;
2783 			lf.l_start = 0;
2784 			lf.l_len = 0;
2785 			lf.l_type = F_UNLCK;
2786 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2787 			    F_UNLCK, &lf, F_POSIX);
2788 		}
2789 		fdtol = td->td_proc->p_fdtol;
2790 		if (fdtol != NULL) {
2791 			/*
2792 			 * Handle special case where file descriptor table is
2793 			 * shared between multiple process leaders.
2794 			 */
2795 			fdp = td->td_proc->p_fd;
2796 			FILEDESC_XLOCK(fdp);
2797 			for (fdtol = fdtol->fdl_next;
2798 			    fdtol != td->td_proc->p_fdtol;
2799 			    fdtol = fdtol->fdl_next) {
2800 				if ((fdtol->fdl_leader->p_flag &
2801 				    P_ADVLOCK) == 0)
2802 					continue;
2803 				fdtol->fdl_holdcount++;
2804 				FILEDESC_XUNLOCK(fdp);
2805 				lf.l_whence = SEEK_SET;
2806 				lf.l_start = 0;
2807 				lf.l_len = 0;
2808 				lf.l_type = F_UNLCK;
2809 				vp = fp->f_vnode;
2810 				(void) VOP_ADVLOCK(vp,
2811 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2812 				    F_POSIX);
2813 				FILEDESC_XLOCK(fdp);
2814 				fdtol->fdl_holdcount--;
2815 				if (fdtol->fdl_holdcount == 0 &&
2816 				    fdtol->fdl_wakeup != 0) {
2817 					fdtol->fdl_wakeup = 0;
2818 					wakeup(fdtol);
2819 				}
2820 			}
2821 			FILEDESC_XUNLOCK(fdp);
2822 		}
2823 	}
2824 	return (fdrop_close(fp, td));
2825 }
2826 
2827 /*
2828  * Hack for file descriptor passing code.
2829  */
2830 void
2831 closef_nothread(struct file *fp)
2832 {
2833 
2834 	fdrop(fp, NULL);
2835 }
2836 
2837 /*
2838  * Initialize the file pointer with the specified properties.
2839  *
2840  * The ops are set with release semantics to be certain that the flags, type,
2841  * and data are visible when ops is.  This is to prevent ops methods from being
2842  * called with bad data.
2843  */
2844 void
2845 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2846 {
2847 	fp->f_data = data;
2848 	fp->f_flag = flag;
2849 	fp->f_type = type;
2850 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2851 }
2852 
2853 void
2854 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2855 {
2856 	fp->f_seqcount[UIO_READ] = 1;
2857 	fp->f_seqcount[UIO_WRITE] = 1;
2858 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2859 	    data, ops);
2860 }
2861 
2862 int
2863 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2864     struct file **fpp, struct filecaps *havecapsp)
2865 {
2866 	struct filedescent *fde;
2867 	int error;
2868 
2869 	FILEDESC_LOCK_ASSERT(fdp);
2870 
2871 	fde = fdeget_locked(fdp, fd);
2872 	if (fde == NULL) {
2873 		error = EBADF;
2874 		goto out;
2875 	}
2876 
2877 #ifdef CAPABILITIES
2878 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2879 	if (error != 0)
2880 		goto out;
2881 #endif
2882 
2883 	if (havecapsp != NULL)
2884 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2885 
2886 	*fpp = fde->fde_file;
2887 
2888 	error = 0;
2889 out:
2890 	return (error);
2891 }
2892 
2893 int
2894 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2895     struct file **fpp, struct filecaps *havecapsp)
2896 {
2897 	struct filedesc *fdp = td->td_proc->p_fd;
2898 	int error;
2899 #ifndef CAPABILITIES
2900 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2901 	if (havecapsp != NULL && error == 0)
2902 		filecaps_fill(havecapsp);
2903 #else
2904 	struct file *fp;
2905 	seqc_t seq;
2906 
2907 	*fpp = NULL;
2908 	for (;;) {
2909 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2910 		if (error != 0)
2911 			return (error);
2912 
2913 		if (havecapsp != NULL) {
2914 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2915 			    havecapsp, false)) {
2916 				fdrop(fp, td);
2917 				goto get_locked;
2918 			}
2919 		}
2920 
2921 		if (!fd_modified(fdp, fd, seq))
2922 			break;
2923 		fdrop(fp, td);
2924 	}
2925 
2926 	*fpp = fp;
2927 	return (0);
2928 
2929 get_locked:
2930 	FILEDESC_SLOCK(fdp);
2931 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2932 	if (error == 0 && !fhold(*fpp))
2933 		error = EBADF;
2934 	FILEDESC_SUNLOCK(fdp);
2935 #endif
2936 	return (error);
2937 }
2938 
2939 #ifdef CAPABILITIES
2940 int
2941 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2942 {
2943 	const struct filedescent *fde;
2944 	const struct fdescenttbl *fdt;
2945 	struct filedesc *fdp;
2946 	struct file *fp;
2947 	struct vnode *vp;
2948 	const cap_rights_t *haverights;
2949 	cap_rights_t rights;
2950 	seqc_t seq;
2951 
2952 	VFS_SMR_ASSERT_ENTERED();
2953 
2954 	rights = *ndp->ni_rightsneeded;
2955 	cap_rights_set_one(&rights, CAP_LOOKUP);
2956 
2957 	fdp = curproc->p_fd;
2958 	fdt = fdp->fd_files;
2959 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2960 		return (EBADF);
2961 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2962 	fde = &fdt->fdt_ofiles[fd];
2963 	haverights = cap_rights_fde_inline(fde);
2964 	fp = fde->fde_file;
2965 	if (__predict_false(fp == NULL))
2966 		return (EAGAIN);
2967 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2968 		return (EAGAIN);
2969 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2970 	vp = fp->f_vnode;
2971 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
2972 		return (EAGAIN);
2973 	}
2974 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2975 		return (EAGAIN);
2976 	}
2977 	/*
2978 	 * Use an acquire barrier to force re-reading of fdt so it is
2979 	 * refreshed for verification.
2980 	 */
2981 	atomic_thread_fence_acq();
2982 	fdt = fdp->fd_files;
2983 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2984 		return (EAGAIN);
2985 	/*
2986 	 * If file descriptor doesn't have all rights,
2987 	 * all lookups relative to it must also be
2988 	 * strictly relative.
2989 	 *
2990 	 * Not yet supported by fast path.
2991 	 */
2992 	CAP_ALL(&rights);
2993 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
2994 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
2995 	    ndp->ni_filecaps.fc_nioctls != -1) {
2996 #ifdef notyet
2997 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
2998 #else
2999 		return (EAGAIN);
3000 #endif
3001 	}
3002 	*vpp = vp;
3003 	return (0);
3004 }
3005 #else
3006 int
3007 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3008 {
3009 	const struct fdescenttbl *fdt;
3010 	struct filedesc *fdp;
3011 	struct file *fp;
3012 	struct vnode *vp;
3013 
3014 	VFS_SMR_ASSERT_ENTERED();
3015 
3016 	fdp = curproc->p_fd;
3017 	fdt = fdp->fd_files;
3018 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3019 		return (EBADF);
3020 	fp = fdt->fdt_ofiles[fd].fde_file;
3021 	if (__predict_false(fp == NULL))
3022 		return (EAGAIN);
3023 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
3024 	vp = fp->f_vnode;
3025 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3026 		return (EAGAIN);
3027 	}
3028 	/*
3029 	 * Use an acquire barrier to force re-reading of fdt so it is
3030 	 * refreshed for verification.
3031 	 */
3032 	atomic_thread_fence_acq();
3033 	fdt = fdp->fd_files;
3034 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3035 		return (EAGAIN);
3036 	filecaps_fill(&ndp->ni_filecaps);
3037 	*vpp = vp;
3038 	return (0);
3039 }
3040 #endif
3041 
3042 int
3043 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3044     struct file **fpp, seqc_t *seqp)
3045 {
3046 #ifdef CAPABILITIES
3047 	const struct filedescent *fde;
3048 #endif
3049 	const struct fdescenttbl *fdt;
3050 	struct file *fp;
3051 #ifdef CAPABILITIES
3052 	seqc_t seq;
3053 	cap_rights_t haverights;
3054 	int error;
3055 #endif
3056 
3057 	fdt = fdp->fd_files;
3058 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3059 		return (EBADF);
3060 	/*
3061 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
3062 	 * never raising a refcount above 0.  To accomplish this we have
3063 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
3064 	 * must be re-verified once we acquire a reference to be certain
3065 	 * that the identity is still correct and we did not lose a race
3066 	 * due to preemption.
3067 	 */
3068 	for (;;) {
3069 #ifdef CAPABILITIES
3070 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3071 		fde = &fdt->fdt_ofiles[fd];
3072 		haverights = *cap_rights_fde_inline(fde);
3073 		fp = fde->fde_file;
3074 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
3075 			continue;
3076 #else
3077 		fp = fdt->fdt_ofiles[fd].fde_file;
3078 #endif
3079 		if (fp == NULL)
3080 			return (EBADF);
3081 #ifdef CAPABILITIES
3082 		error = cap_check_inline(&haverights, needrightsp);
3083 		if (error != 0)
3084 			return (error);
3085 #endif
3086 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3087 			/*
3088 			 * Force a reload. Other thread could reallocate the
3089 			 * table before this fd was closed, so it is possible
3090 			 * that there is a stale fp pointer in cached version.
3091 			 */
3092 			fdt = atomic_load_ptr(&fdp->fd_files);
3093 			continue;
3094 		}
3095 		/*
3096 		 * Use an acquire barrier to force re-reading of fdt so it is
3097 		 * refreshed for verification.
3098 		 */
3099 		atomic_thread_fence_acq();
3100 		fdt = fdp->fd_files;
3101 #ifdef	CAPABILITIES
3102 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
3103 #else
3104 		if (fp == fdt->fdt_ofiles[fd].fde_file)
3105 #endif
3106 			break;
3107 		fdrop(fp, curthread);
3108 	}
3109 	*fpp = fp;
3110 	if (seqp != NULL) {
3111 #ifdef CAPABILITIES
3112 		*seqp = seq;
3113 #endif
3114 	}
3115 	return (0);
3116 }
3117 
3118 /*
3119  * See the comments in fget_unlocked_seq for an explanation of how this works.
3120  *
3121  * This is a simplified variant which bails out to the aforementioned routine
3122  * if anything goes wrong. In practice this only happens when userspace is
3123  * racing with itself.
3124  */
3125 int
3126 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3127     struct file **fpp)
3128 {
3129 #ifdef CAPABILITIES
3130 	const struct filedescent *fde;
3131 #endif
3132 	const struct fdescenttbl *fdt;
3133 	struct file *fp;
3134 #ifdef CAPABILITIES
3135 	seqc_t seq;
3136 	const cap_rights_t *haverights;
3137 #endif
3138 
3139 	fdt = fdp->fd_files;
3140 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3141 		return (EBADF);
3142 #ifdef CAPABILITIES
3143 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3144 	fde = &fdt->fdt_ofiles[fd];
3145 	haverights = cap_rights_fde_inline(fde);
3146 	fp = fde->fde_file;
3147 #else
3148 	fp = fdt->fdt_ofiles[fd].fde_file;
3149 #endif
3150 	if (__predict_false(fp == NULL))
3151 		goto out_fallback;
3152 #ifdef CAPABILITIES
3153 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3154 		goto out_fallback;
3155 #endif
3156 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3157 		goto out_fallback;
3158 
3159 	/*
3160 	 * Use an acquire barrier to force re-reading of fdt so it is
3161 	 * refreshed for verification.
3162 	 */
3163 	atomic_thread_fence_acq();
3164 	fdt = fdp->fd_files;
3165 #ifdef	CAPABILITIES
3166 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
3167 #else
3168 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3169 #endif
3170 		goto out_fdrop;
3171 	*fpp = fp;
3172 	return (0);
3173 out_fdrop:
3174 	fdrop(fp, curthread);
3175 out_fallback:
3176 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
3177 }
3178 
3179 /*
3180  * Extract the file pointer associated with the specified descriptor for the
3181  * current user process.
3182  *
3183  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3184  * returned.
3185  *
3186  * File's rights will be checked against the capability rights mask.
3187  *
3188  * If an error occurred the non-zero error is returned and *fpp is set to
3189  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3190  * responsible for fdrop().
3191  */
3192 static __inline int
3193 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3194     cap_rights_t *needrightsp)
3195 {
3196 	struct filedesc *fdp;
3197 	struct file *fp;
3198 	int error;
3199 
3200 	*fpp = NULL;
3201 	fdp = td->td_proc->p_fd;
3202 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
3203 	if (__predict_false(error != 0))
3204 		return (error);
3205 	if (__predict_false(fp->f_ops == &badfileops)) {
3206 		fdrop(fp, td);
3207 		return (EBADF);
3208 	}
3209 
3210 	/*
3211 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3212 	 */
3213 	error = 0;
3214 	switch (flags) {
3215 	case FREAD:
3216 	case FWRITE:
3217 		if ((fp->f_flag & flags) == 0)
3218 			error = EBADF;
3219 		break;
3220 	case FEXEC:
3221 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3222 		    ((fp->f_flag & FWRITE) != 0))
3223 			error = EBADF;
3224 		break;
3225 	case 0:
3226 		break;
3227 	default:
3228 		KASSERT(0, ("wrong flags"));
3229 	}
3230 
3231 	if (error != 0) {
3232 		fdrop(fp, td);
3233 		return (error);
3234 	}
3235 
3236 	*fpp = fp;
3237 	return (0);
3238 }
3239 
3240 int
3241 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3242 {
3243 
3244 	return (_fget(td, fd, fpp, 0, rightsp));
3245 }
3246 
3247 int
3248 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3249     struct file **fpp)
3250 {
3251 	int error;
3252 #ifndef CAPABILITIES
3253 	error = _fget(td, fd, fpp, 0, rightsp);
3254 	if (maxprotp != NULL)
3255 		*maxprotp = VM_PROT_ALL;
3256 	return (error);
3257 #else
3258 	cap_rights_t fdrights;
3259 	struct filedesc *fdp;
3260 	struct file *fp;
3261 	seqc_t seq;
3262 
3263 	*fpp = NULL;
3264 	fdp = td->td_proc->p_fd;
3265 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3266 	for (;;) {
3267 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3268 		if (__predict_false(error != 0))
3269 			return (error);
3270 		if (__predict_false(fp->f_ops == &badfileops)) {
3271 			fdrop(fp, td);
3272 			return (EBADF);
3273 		}
3274 		if (maxprotp != NULL)
3275 			fdrights = *cap_rights(fdp, fd);
3276 		if (!fd_modified(fdp, fd, seq))
3277 			break;
3278 		fdrop(fp, td);
3279 	}
3280 
3281 	/*
3282 	 * If requested, convert capability rights to access flags.
3283 	 */
3284 	if (maxprotp != NULL)
3285 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3286 	*fpp = fp;
3287 	return (0);
3288 #endif
3289 }
3290 
3291 int
3292 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3293 {
3294 
3295 	return (_fget(td, fd, fpp, FREAD, rightsp));
3296 }
3297 
3298 int
3299 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3300 {
3301 
3302 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3303 }
3304 
3305 int
3306 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3307     struct file **fpp)
3308 {
3309 	struct filedesc *fdp = td->td_proc->p_fd;
3310 #ifndef CAPABILITIES
3311 	return (fget_unlocked(fdp, fd, rightsp, fpp));
3312 #else
3313 	struct file *fp;
3314 	int error;
3315 	seqc_t seq;
3316 
3317 	*fpp = NULL;
3318 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3319 	for (;;) {
3320 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
3321 		if (error != 0)
3322 			return (error);
3323 		error = cap_fcntl_check(fdp, fd, needfcntl);
3324 		if (!fd_modified(fdp, fd, seq))
3325 			break;
3326 		fdrop(fp, td);
3327 	}
3328 	if (error != 0) {
3329 		fdrop(fp, td);
3330 		return (error);
3331 	}
3332 	*fpp = fp;
3333 	return (0);
3334 #endif
3335 }
3336 
3337 /*
3338  * Like fget() but loads the underlying vnode, or returns an error if the
3339  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3340  * never have VM objects.  The returned vnode will be vref()'d.
3341  *
3342  * XXX: what about the unused flags ?
3343  */
3344 static __inline int
3345 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3346     struct vnode **vpp)
3347 {
3348 	struct file *fp;
3349 	int error;
3350 
3351 	*vpp = NULL;
3352 	error = _fget(td, fd, &fp, flags, needrightsp);
3353 	if (error != 0)
3354 		return (error);
3355 	if (fp->f_vnode == NULL) {
3356 		error = EINVAL;
3357 	} else {
3358 		*vpp = fp->f_vnode;
3359 		vrefact(*vpp);
3360 	}
3361 	fdrop(fp, td);
3362 
3363 	return (error);
3364 }
3365 
3366 int
3367 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3368 {
3369 
3370 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3371 }
3372 
3373 int
3374 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3375     struct filecaps *havecaps, struct vnode **vpp)
3376 {
3377 	struct filecaps caps;
3378 	struct file *fp;
3379 	int error;
3380 
3381 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3382 	if (error != 0)
3383 		return (error);
3384 	if (fp->f_ops == &badfileops) {
3385 		error = EBADF;
3386 		goto out;
3387 	}
3388 	if (fp->f_vnode == NULL) {
3389 		error = EINVAL;
3390 		goto out;
3391 	}
3392 
3393 	*havecaps = caps;
3394 	*vpp = fp->f_vnode;
3395 	vrefact(*vpp);
3396 	fdrop(fp, td);
3397 
3398 	return (0);
3399 out:
3400 	filecaps_free(&caps);
3401 	fdrop(fp, td);
3402 	return (error);
3403 }
3404 
3405 int
3406 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3407 {
3408 
3409 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3410 }
3411 
3412 int
3413 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3414 {
3415 
3416 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3417 }
3418 
3419 #ifdef notyet
3420 int
3421 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3422     struct vnode **vpp)
3423 {
3424 
3425 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3426 }
3427 #endif
3428 
3429 /*
3430  * Handle the last reference to a file being closed.
3431  *
3432  * Without the noinline attribute clang keeps inlining the func thorough this
3433  * file when fdrop is used.
3434  */
3435 int __noinline
3436 _fdrop(struct file *fp, struct thread *td)
3437 {
3438 	int error;
3439 #ifdef INVARIANTS
3440 	int count;
3441 
3442 	count = refcount_load(&fp->f_count);
3443 	if (count != 0)
3444 		panic("fdrop: fp %p count %d", fp, count);
3445 #endif
3446 	error = fo_close(fp, td);
3447 	atomic_subtract_int(&openfiles, 1);
3448 	crfree(fp->f_cred);
3449 	free(fp->f_advice, M_FADVISE);
3450 	uma_zfree(file_zone, fp);
3451 
3452 	return (error);
3453 }
3454 
3455 /*
3456  * Apply an advisory lock on a file descriptor.
3457  *
3458  * Just attempt to get a record lock of the requested type on the entire file
3459  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3460  */
3461 #ifndef _SYS_SYSPROTO_H_
3462 struct flock_args {
3463 	int	fd;
3464 	int	how;
3465 };
3466 #endif
3467 /* ARGSUSED */
3468 int
3469 sys_flock(struct thread *td, struct flock_args *uap)
3470 {
3471 	struct file *fp;
3472 	struct vnode *vp;
3473 	struct flock lf;
3474 	int error;
3475 
3476 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3477 	if (error != 0)
3478 		return (error);
3479 	if (fp->f_type != DTYPE_VNODE) {
3480 		fdrop(fp, td);
3481 		return (EOPNOTSUPP);
3482 	}
3483 
3484 	vp = fp->f_vnode;
3485 	lf.l_whence = SEEK_SET;
3486 	lf.l_start = 0;
3487 	lf.l_len = 0;
3488 	if (uap->how & LOCK_UN) {
3489 		lf.l_type = F_UNLCK;
3490 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3491 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3492 		goto done2;
3493 	}
3494 	if (uap->how & LOCK_EX)
3495 		lf.l_type = F_WRLCK;
3496 	else if (uap->how & LOCK_SH)
3497 		lf.l_type = F_RDLCK;
3498 	else {
3499 		error = EBADF;
3500 		goto done2;
3501 	}
3502 	atomic_set_int(&fp->f_flag, FHASLOCK);
3503 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3504 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3505 done2:
3506 	fdrop(fp, td);
3507 	return (error);
3508 }
3509 /*
3510  * Duplicate the specified descriptor to a free descriptor.
3511  */
3512 int
3513 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3514     int openerror, int *indxp)
3515 {
3516 	struct filedescent *newfde, *oldfde;
3517 	struct file *fp;
3518 	u_long *ioctls;
3519 	int error, indx;
3520 
3521 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3522 	    ("unexpected error %d in %s", openerror, __func__));
3523 
3524 	/*
3525 	 * If the to-be-dup'd fd number is greater than the allowed number
3526 	 * of file descriptors, or the fd to be dup'd has already been
3527 	 * closed, then reject.
3528 	 */
3529 	FILEDESC_XLOCK(fdp);
3530 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3531 		FILEDESC_XUNLOCK(fdp);
3532 		return (EBADF);
3533 	}
3534 
3535 	error = fdalloc(td, 0, &indx);
3536 	if (error != 0) {
3537 		FILEDESC_XUNLOCK(fdp);
3538 		return (error);
3539 	}
3540 
3541 	/*
3542 	 * There are two cases of interest here.
3543 	 *
3544 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3545 	 *
3546 	 * For ENXIO steal away the file structure from (dfd) and store it in
3547 	 * (indx).  (dfd) is effectively closed by this operation.
3548 	 */
3549 	switch (openerror) {
3550 	case ENODEV:
3551 		/*
3552 		 * Check that the mode the file is being opened for is a
3553 		 * subset of the mode of the existing descriptor.
3554 		 */
3555 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3556 			fdunused(fdp, indx);
3557 			FILEDESC_XUNLOCK(fdp);
3558 			return (EACCES);
3559 		}
3560 		if (!fhold(fp)) {
3561 			fdunused(fdp, indx);
3562 			FILEDESC_XUNLOCK(fdp);
3563 			return (EBADF);
3564 		}
3565 		newfde = &fdp->fd_ofiles[indx];
3566 		oldfde = &fdp->fd_ofiles[dfd];
3567 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3568 #ifdef CAPABILITIES
3569 		seqc_write_begin(&newfde->fde_seqc);
3570 #endif
3571 		memcpy(newfde, oldfde, fde_change_size);
3572 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3573 		    ioctls);
3574 #ifdef CAPABILITIES
3575 		seqc_write_end(&newfde->fde_seqc);
3576 #endif
3577 		break;
3578 	case ENXIO:
3579 		/*
3580 		 * Steal away the file pointer from dfd and stuff it into indx.
3581 		 */
3582 		newfde = &fdp->fd_ofiles[indx];
3583 		oldfde = &fdp->fd_ofiles[dfd];
3584 #ifdef CAPABILITIES
3585 		seqc_write_begin(&newfde->fde_seqc);
3586 #endif
3587 		memcpy(newfde, oldfde, fde_change_size);
3588 		oldfde->fde_file = NULL;
3589 		fdunused(fdp, dfd);
3590 #ifdef CAPABILITIES
3591 		seqc_write_end(&newfde->fde_seqc);
3592 #endif
3593 		break;
3594 	}
3595 	FILEDESC_XUNLOCK(fdp);
3596 	*indxp = indx;
3597 	return (0);
3598 }
3599 
3600 /*
3601  * This sysctl determines if we will allow a process to chroot(2) if it
3602  * has a directory open:
3603  *	0: disallowed for all processes.
3604  *	1: allowed for processes that were not already chroot(2)'ed.
3605  *	2: allowed for all processes.
3606  */
3607 
3608 static int chroot_allow_open_directories = 1;
3609 
3610 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3611     &chroot_allow_open_directories, 0,
3612     "Allow a process to chroot(2) if it has a directory open");
3613 
3614 /*
3615  * Helper function for raised chroot(2) security function:  Refuse if
3616  * any filedescriptors are open directories.
3617  */
3618 static int
3619 chroot_refuse_vdir_fds(struct filedesc *fdp)
3620 {
3621 	struct vnode *vp;
3622 	struct file *fp;
3623 	int fd, lastfile;
3624 
3625 	FILEDESC_LOCK_ASSERT(fdp);
3626 
3627 	lastfile = fdlastfile(fdp);
3628 	for (fd = 0; fd <= lastfile; fd++) {
3629 		fp = fget_locked(fdp, fd);
3630 		if (fp == NULL)
3631 			continue;
3632 		if (fp->f_type == DTYPE_VNODE) {
3633 			vp = fp->f_vnode;
3634 			if (vp->v_type == VDIR)
3635 				return (EPERM);
3636 		}
3637 	}
3638 	return (0);
3639 }
3640 
3641 static void
3642 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3643 {
3644 
3645 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3646 		vrefact(oldpwd->pwd_cdir);
3647 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3648 	}
3649 
3650 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3651 		vrefact(oldpwd->pwd_rdir);
3652 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3653 	}
3654 
3655 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3656 		vrefact(oldpwd->pwd_jdir);
3657 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3658 	}
3659 }
3660 
3661 struct pwd *
3662 pwd_hold_pwddesc(struct pwddesc *pdp)
3663 {
3664 	struct pwd *pwd;
3665 
3666 	PWDDESC_ASSERT_XLOCKED(pdp);
3667 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3668 	if (pwd != NULL)
3669 		refcount_acquire(&pwd->pwd_refcount);
3670 	return (pwd);
3671 }
3672 
3673 bool
3674 pwd_hold_smr(struct pwd *pwd)
3675 {
3676 
3677 	MPASS(pwd != NULL);
3678 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3679 		return (true);
3680 	}
3681 	return (false);
3682 }
3683 
3684 struct pwd *
3685 pwd_hold(struct thread *td)
3686 {
3687 	struct pwddesc *pdp;
3688 	struct pwd *pwd;
3689 
3690 	pdp = td->td_proc->p_pd;
3691 
3692 	vfs_smr_enter();
3693 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3694 	if (pwd_hold_smr(pwd)) {
3695 		vfs_smr_exit();
3696 		return (pwd);
3697 	}
3698 	vfs_smr_exit();
3699 	PWDDESC_XLOCK(pdp);
3700 	pwd = pwd_hold_pwddesc(pdp);
3701 	MPASS(pwd != NULL);
3702 	PWDDESC_XUNLOCK(pdp);
3703 	return (pwd);
3704 }
3705 
3706 static struct pwd *
3707 pwd_alloc(void)
3708 {
3709 	struct pwd *pwd;
3710 
3711 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3712 	bzero(pwd, sizeof(*pwd));
3713 	refcount_init(&pwd->pwd_refcount, 1);
3714 	return (pwd);
3715 }
3716 
3717 void
3718 pwd_drop(struct pwd *pwd)
3719 {
3720 
3721 	if (!refcount_release(&pwd->pwd_refcount))
3722 		return;
3723 
3724 	if (pwd->pwd_cdir != NULL)
3725 		vrele(pwd->pwd_cdir);
3726 	if (pwd->pwd_rdir != NULL)
3727 		vrele(pwd->pwd_rdir);
3728 	if (pwd->pwd_jdir != NULL)
3729 		vrele(pwd->pwd_jdir);
3730 	uma_zfree_smr(pwd_zone, pwd);
3731 }
3732 
3733 /*
3734 * Common routine for kern_chroot() and jail_attach().  The caller is
3735 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3736 * authorize this operation.
3737 */
3738 int
3739 pwd_chroot(struct thread *td, struct vnode *vp)
3740 {
3741 	struct pwddesc *pdp;
3742 	struct filedesc *fdp;
3743 	struct pwd *newpwd, *oldpwd;
3744 	int error;
3745 
3746 	fdp = td->td_proc->p_fd;
3747 	pdp = td->td_proc->p_pd;
3748 	newpwd = pwd_alloc();
3749 	FILEDESC_SLOCK(fdp);
3750 	PWDDESC_XLOCK(pdp);
3751 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3752 	if (chroot_allow_open_directories == 0 ||
3753 	    (chroot_allow_open_directories == 1 &&
3754 	    oldpwd->pwd_rdir != rootvnode)) {
3755 		error = chroot_refuse_vdir_fds(fdp);
3756 		FILEDESC_SUNLOCK(fdp);
3757 		if (error != 0) {
3758 			PWDDESC_XUNLOCK(pdp);
3759 			pwd_drop(newpwd);
3760 			return (error);
3761 		}
3762 	} else {
3763 		FILEDESC_SUNLOCK(fdp);
3764 	}
3765 
3766 	vrefact(vp);
3767 	newpwd->pwd_rdir = vp;
3768 	if (oldpwd->pwd_jdir == NULL) {
3769 		vrefact(vp);
3770 		newpwd->pwd_jdir = vp;
3771 	}
3772 	pwd_fill(oldpwd, newpwd);
3773 	pwd_set(pdp, newpwd);
3774 	PWDDESC_XUNLOCK(pdp);
3775 	pwd_drop(oldpwd);
3776 	return (0);
3777 }
3778 
3779 void
3780 pwd_chdir(struct thread *td, struct vnode *vp)
3781 {
3782 	struct pwddesc *pdp;
3783 	struct pwd *newpwd, *oldpwd;
3784 
3785 	VNPASS(vp->v_usecount > 0, vp);
3786 
3787 	newpwd = pwd_alloc();
3788 	pdp = td->td_proc->p_pd;
3789 	PWDDESC_XLOCK(pdp);
3790 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3791 	newpwd->pwd_cdir = vp;
3792 	pwd_fill(oldpwd, newpwd);
3793 	pwd_set(pdp, newpwd);
3794 	PWDDESC_XUNLOCK(pdp);
3795 	pwd_drop(oldpwd);
3796 }
3797 
3798 void
3799 pwd_ensure_dirs(void)
3800 {
3801 	struct pwddesc *pdp;
3802 	struct pwd *oldpwd, *newpwd;
3803 
3804 	pdp = curproc->p_pd;
3805 	PWDDESC_XLOCK(pdp);
3806 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3807 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3808 		PWDDESC_XUNLOCK(pdp);
3809 		return;
3810 	}
3811 	PWDDESC_XUNLOCK(pdp);
3812 
3813 	newpwd = pwd_alloc();
3814 	PWDDESC_XLOCK(pdp);
3815 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3816 	pwd_fill(oldpwd, newpwd);
3817 	if (newpwd->pwd_cdir == NULL) {
3818 		vrefact(rootvnode);
3819 		newpwd->pwd_cdir = rootvnode;
3820 	}
3821 	if (newpwd->pwd_rdir == NULL) {
3822 		vrefact(rootvnode);
3823 		newpwd->pwd_rdir = rootvnode;
3824 	}
3825 	pwd_set(pdp, newpwd);
3826 	PWDDESC_XUNLOCK(pdp);
3827 	pwd_drop(oldpwd);
3828 }
3829 
3830 void
3831 pwd_set_rootvnode(void)
3832 {
3833 	struct pwddesc *pdp;
3834 	struct pwd *oldpwd, *newpwd;
3835 
3836 	pdp = curproc->p_pd;
3837 
3838 	newpwd = pwd_alloc();
3839 	PWDDESC_XLOCK(pdp);
3840 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3841 	vrefact(rootvnode);
3842 	newpwd->pwd_cdir = rootvnode;
3843 	vrefact(rootvnode);
3844 	newpwd->pwd_rdir = rootvnode;
3845 	pwd_fill(oldpwd, newpwd);
3846 	pwd_set(pdp, newpwd);
3847 	PWDDESC_XUNLOCK(pdp);
3848 	pwd_drop(oldpwd);
3849 }
3850 
3851 /*
3852  * Scan all active processes and prisons to see if any of them have a current
3853  * or root directory of `olddp'. If so, replace them with the new mount point.
3854  */
3855 void
3856 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3857 {
3858 	struct pwddesc *pdp;
3859 	struct pwd *newpwd, *oldpwd;
3860 	struct prison *pr;
3861 	struct proc *p;
3862 	int nrele;
3863 
3864 	if (vrefcnt(olddp) == 1)
3865 		return;
3866 	nrele = 0;
3867 	newpwd = pwd_alloc();
3868 	sx_slock(&allproc_lock);
3869 	FOREACH_PROC_IN_SYSTEM(p) {
3870 		PROC_LOCK(p);
3871 		pdp = pdhold(p);
3872 		PROC_UNLOCK(p);
3873 		if (pdp == NULL)
3874 			continue;
3875 		PWDDESC_XLOCK(pdp);
3876 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3877 		if (oldpwd == NULL ||
3878 		    (oldpwd->pwd_cdir != olddp &&
3879 		    oldpwd->pwd_rdir != olddp &&
3880 		    oldpwd->pwd_jdir != olddp)) {
3881 			PWDDESC_XUNLOCK(pdp);
3882 			pddrop(pdp);
3883 			continue;
3884 		}
3885 		if (oldpwd->pwd_cdir == olddp) {
3886 			vrefact(newdp);
3887 			newpwd->pwd_cdir = newdp;
3888 		}
3889 		if (oldpwd->pwd_rdir == olddp) {
3890 			vrefact(newdp);
3891 			newpwd->pwd_rdir = newdp;
3892 		}
3893 		if (oldpwd->pwd_jdir == olddp) {
3894 			vrefact(newdp);
3895 			newpwd->pwd_jdir = newdp;
3896 		}
3897 		pwd_fill(oldpwd, newpwd);
3898 		pwd_set(pdp, newpwd);
3899 		PWDDESC_XUNLOCK(pdp);
3900 		pwd_drop(oldpwd);
3901 		pddrop(pdp);
3902 		newpwd = pwd_alloc();
3903 	}
3904 	sx_sunlock(&allproc_lock);
3905 	pwd_drop(newpwd);
3906 	if (rootvnode == olddp) {
3907 		vrefact(newdp);
3908 		rootvnode = newdp;
3909 		nrele++;
3910 	}
3911 	mtx_lock(&prison0.pr_mtx);
3912 	if (prison0.pr_root == olddp) {
3913 		vrefact(newdp);
3914 		prison0.pr_root = newdp;
3915 		nrele++;
3916 	}
3917 	mtx_unlock(&prison0.pr_mtx);
3918 	sx_slock(&allprison_lock);
3919 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3920 		mtx_lock(&pr->pr_mtx);
3921 		if (pr->pr_root == olddp) {
3922 			vrefact(newdp);
3923 			pr->pr_root = newdp;
3924 			nrele++;
3925 		}
3926 		mtx_unlock(&pr->pr_mtx);
3927 	}
3928 	sx_sunlock(&allprison_lock);
3929 	while (nrele--)
3930 		vrele(olddp);
3931 }
3932 
3933 struct filedesc_to_leader *
3934 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3935 {
3936 	struct filedesc_to_leader *fdtol;
3937 
3938 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3939 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3940 	fdtol->fdl_refcount = 1;
3941 	fdtol->fdl_holdcount = 0;
3942 	fdtol->fdl_wakeup = 0;
3943 	fdtol->fdl_leader = leader;
3944 	if (old != NULL) {
3945 		FILEDESC_XLOCK(fdp);
3946 		fdtol->fdl_next = old->fdl_next;
3947 		fdtol->fdl_prev = old;
3948 		old->fdl_next = fdtol;
3949 		fdtol->fdl_next->fdl_prev = fdtol;
3950 		FILEDESC_XUNLOCK(fdp);
3951 	} else {
3952 		fdtol->fdl_next = fdtol;
3953 		fdtol->fdl_prev = fdtol;
3954 	}
3955 	return (fdtol);
3956 }
3957 
3958 static int
3959 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3960 {
3961 	NDSLOTTYPE *map;
3962 	struct filedesc *fdp;
3963 	int count, off, minoff;
3964 
3965 	if (*(int *)arg1 != 0)
3966 		return (EINVAL);
3967 
3968 	fdp = curproc->p_fd;
3969 	count = 0;
3970 	FILEDESC_SLOCK(fdp);
3971 	map = fdp->fd_map;
3972 	off = NDSLOT(fdp->fd_nfiles - 1);
3973 	for (minoff = NDSLOT(0); off >= minoff; --off)
3974 		count += bitcountl(map[off]);
3975 	FILEDESC_SUNLOCK(fdp);
3976 
3977 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3978 }
3979 
3980 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3981     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3982     "Number of open file descriptors");
3983 
3984 /*
3985  * Get file structures globally.
3986  */
3987 static int
3988 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3989 {
3990 	struct xfile xf;
3991 	struct filedesc *fdp;
3992 	struct file *fp;
3993 	struct proc *p;
3994 	int error, n, lastfile;
3995 
3996 	error = sysctl_wire_old_buffer(req, 0);
3997 	if (error != 0)
3998 		return (error);
3999 	if (req->oldptr == NULL) {
4000 		n = 0;
4001 		sx_slock(&allproc_lock);
4002 		FOREACH_PROC_IN_SYSTEM(p) {
4003 			PROC_LOCK(p);
4004 			if (p->p_state == PRS_NEW) {
4005 				PROC_UNLOCK(p);
4006 				continue;
4007 			}
4008 			fdp = fdhold(p);
4009 			PROC_UNLOCK(p);
4010 			if (fdp == NULL)
4011 				continue;
4012 			/* overestimates sparse tables. */
4013 			n += fdp->fd_nfiles;
4014 			fddrop(fdp);
4015 		}
4016 		sx_sunlock(&allproc_lock);
4017 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4018 	}
4019 	error = 0;
4020 	bzero(&xf, sizeof(xf));
4021 	xf.xf_size = sizeof(xf);
4022 	sx_slock(&allproc_lock);
4023 	FOREACH_PROC_IN_SYSTEM(p) {
4024 		PROC_LOCK(p);
4025 		if (p->p_state == PRS_NEW) {
4026 			PROC_UNLOCK(p);
4027 			continue;
4028 		}
4029 		if (p_cansee(req->td, p) != 0) {
4030 			PROC_UNLOCK(p);
4031 			continue;
4032 		}
4033 		xf.xf_pid = p->p_pid;
4034 		xf.xf_uid = p->p_ucred->cr_uid;
4035 		fdp = fdhold(p);
4036 		PROC_UNLOCK(p);
4037 		if (fdp == NULL)
4038 			continue;
4039 		FILEDESC_SLOCK(fdp);
4040 		lastfile = fdlastfile(fdp);
4041 		for (n = 0; refcount_load(&fdp->fd_refcnt) > 0 && n <= lastfile;
4042 		    n++) {
4043 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4044 				continue;
4045 			xf.xf_fd = n;
4046 			xf.xf_file = (uintptr_t)fp;
4047 			xf.xf_data = (uintptr_t)fp->f_data;
4048 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4049 			xf.xf_type = (uintptr_t)fp->f_type;
4050 			xf.xf_count = refcount_load(&fp->f_count);
4051 			xf.xf_msgcount = 0;
4052 			xf.xf_offset = foffset_get(fp);
4053 			xf.xf_flag = fp->f_flag;
4054 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4055 			if (error)
4056 				break;
4057 		}
4058 		FILEDESC_SUNLOCK(fdp);
4059 		fddrop(fdp);
4060 		if (error)
4061 			break;
4062 	}
4063 	sx_sunlock(&allproc_lock);
4064 	return (error);
4065 }
4066 
4067 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4068     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4069 
4070 #ifdef KINFO_FILE_SIZE
4071 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4072 #endif
4073 
4074 static int
4075 xlate_fflags(int fflags)
4076 {
4077 	static const struct {
4078 		int	fflag;
4079 		int	kf_fflag;
4080 	} fflags_table[] = {
4081 		{ FAPPEND, KF_FLAG_APPEND },
4082 		{ FASYNC, KF_FLAG_ASYNC },
4083 		{ FFSYNC, KF_FLAG_FSYNC },
4084 		{ FHASLOCK, KF_FLAG_HASLOCK },
4085 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4086 		{ FREAD, KF_FLAG_READ },
4087 		{ FWRITE, KF_FLAG_WRITE },
4088 		{ O_CREAT, KF_FLAG_CREAT },
4089 		{ O_DIRECT, KF_FLAG_DIRECT },
4090 		{ O_EXCL, KF_FLAG_EXCL },
4091 		{ O_EXEC, KF_FLAG_EXEC },
4092 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4093 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4094 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4095 		{ O_TRUNC, KF_FLAG_TRUNC }
4096 	};
4097 	unsigned int i;
4098 	int kflags;
4099 
4100 	kflags = 0;
4101 	for (i = 0; i < nitems(fflags_table); i++)
4102 		if (fflags & fflags_table[i].fflag)
4103 			kflags |=  fflags_table[i].kf_fflag;
4104 	return (kflags);
4105 }
4106 
4107 /* Trim unused data from kf_path by truncating the structure size. */
4108 void
4109 pack_kinfo(struct kinfo_file *kif)
4110 {
4111 
4112 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4113 	    strlen(kif->kf_path) + 1;
4114 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4115 }
4116 
4117 static void
4118 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4119     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4120 {
4121 	int error;
4122 
4123 	bzero(kif, sizeof(*kif));
4124 
4125 	/* Set a default type to allow for empty fill_kinfo() methods. */
4126 	kif->kf_type = KF_TYPE_UNKNOWN;
4127 	kif->kf_flags = xlate_fflags(fp->f_flag);
4128 	if (rightsp != NULL)
4129 		kif->kf_cap_rights = *rightsp;
4130 	else
4131 		cap_rights_init_zero(&kif->kf_cap_rights);
4132 	kif->kf_fd = fd;
4133 	kif->kf_ref_count = refcount_load(&fp->f_count);
4134 	kif->kf_offset = foffset_get(fp);
4135 
4136 	/*
4137 	 * This may drop the filedesc lock, so the 'fp' cannot be
4138 	 * accessed after this call.
4139 	 */
4140 	error = fo_fill_kinfo(fp, kif, fdp);
4141 	if (error == 0)
4142 		kif->kf_status |= KF_ATTR_VALID;
4143 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4144 		pack_kinfo(kif);
4145 	else
4146 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4147 }
4148 
4149 static void
4150 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4151     struct kinfo_file *kif, int flags)
4152 {
4153 	int error;
4154 
4155 	bzero(kif, sizeof(*kif));
4156 
4157 	kif->kf_type = KF_TYPE_VNODE;
4158 	error = vn_fill_kinfo_vnode(vp, kif);
4159 	if (error == 0)
4160 		kif->kf_status |= KF_ATTR_VALID;
4161 	kif->kf_flags = xlate_fflags(fflags);
4162 	cap_rights_init_zero(&kif->kf_cap_rights);
4163 	kif->kf_fd = fd;
4164 	kif->kf_ref_count = -1;
4165 	kif->kf_offset = -1;
4166 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4167 		pack_kinfo(kif);
4168 	else
4169 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4170 	vrele(vp);
4171 }
4172 
4173 struct export_fd_buf {
4174 	struct filedesc		*fdp;
4175 	struct pwddesc	*pdp;
4176 	struct sbuf 		*sb;
4177 	ssize_t			remainder;
4178 	struct kinfo_file	kif;
4179 	int			flags;
4180 };
4181 
4182 static int
4183 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4184 {
4185 	struct kinfo_file *kif;
4186 
4187 	kif = &efbuf->kif;
4188 	if (efbuf->remainder != -1) {
4189 		if (efbuf->remainder < kif->kf_structsize) {
4190 			/* Terminate export. */
4191 			efbuf->remainder = 0;
4192 			return (0);
4193 		}
4194 		efbuf->remainder -= kif->kf_structsize;
4195 	}
4196 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
4197 }
4198 
4199 static int
4200 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4201     struct export_fd_buf *efbuf)
4202 {
4203 	int error;
4204 
4205 	if (efbuf->remainder == 0)
4206 		return (0);
4207 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4208 	    efbuf->flags);
4209 	FILEDESC_SUNLOCK(efbuf->fdp);
4210 	error = export_kinfo_to_sb(efbuf);
4211 	FILEDESC_SLOCK(efbuf->fdp);
4212 	return (error);
4213 }
4214 
4215 static int
4216 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4217     struct export_fd_buf *efbuf)
4218 {
4219 	int error;
4220 
4221 	if (efbuf->remainder == 0)
4222 		return (0);
4223 	if (efbuf->pdp != NULL)
4224 		PWDDESC_XUNLOCK(efbuf->pdp);
4225 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4226 	error = export_kinfo_to_sb(efbuf);
4227 	if (efbuf->pdp != NULL)
4228 		PWDDESC_XLOCK(efbuf->pdp);
4229 	return (error);
4230 }
4231 
4232 /*
4233  * Store a process file descriptor information to sbuf.
4234  *
4235  * Takes a locked proc as argument, and returns with the proc unlocked.
4236  */
4237 int
4238 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4239     int flags)
4240 {
4241 	struct file *fp;
4242 	struct filedesc *fdp;
4243 	struct pwddesc *pdp;
4244 	struct export_fd_buf *efbuf;
4245 	struct vnode *cttyvp, *textvp, *tracevp;
4246 	struct pwd *pwd;
4247 	int error, i, lastfile;
4248 	cap_rights_t rights;
4249 
4250 	PROC_LOCK_ASSERT(p, MA_OWNED);
4251 
4252 	/* ktrace vnode */
4253 	tracevp = p->p_tracevp;
4254 	if (tracevp != NULL)
4255 		vrefact(tracevp);
4256 	/* text vnode */
4257 	textvp = p->p_textvp;
4258 	if (textvp != NULL)
4259 		vrefact(textvp);
4260 	/* Controlling tty. */
4261 	cttyvp = NULL;
4262 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4263 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4264 		if (cttyvp != NULL)
4265 			vrefact(cttyvp);
4266 	}
4267 	fdp = fdhold(p);
4268 	pdp = pdhold(p);
4269 	PROC_UNLOCK(p);
4270 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4271 	efbuf->fdp = NULL;
4272 	efbuf->pdp = NULL;
4273 	efbuf->sb = sb;
4274 	efbuf->remainder = maxlen;
4275 	efbuf->flags = flags;
4276 	if (tracevp != NULL)
4277 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
4278 		    efbuf);
4279 	if (textvp != NULL)
4280 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
4281 	if (cttyvp != NULL)
4282 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
4283 		    efbuf);
4284 	error = 0;
4285 	if (pdp == NULL || fdp == NULL)
4286 		goto fail;
4287 	efbuf->fdp = fdp;
4288 	efbuf->pdp = pdp;
4289 	PWDDESC_XLOCK(pdp);
4290 	pwd = pwd_hold_pwddesc(pdp);
4291 	if (pwd != NULL) {
4292 		/* working directory */
4293 		if (pwd->pwd_cdir != NULL) {
4294 			vrefact(pwd->pwd_cdir);
4295 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4296 		}
4297 		/* root directory */
4298 		if (pwd->pwd_rdir != NULL) {
4299 			vrefact(pwd->pwd_rdir);
4300 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
4301 		}
4302 		/* jail directory */
4303 		if (pwd->pwd_jdir != NULL) {
4304 			vrefact(pwd->pwd_jdir);
4305 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
4306 		}
4307 	}
4308 	PWDDESC_XUNLOCK(pdp);
4309 	if (pwd != NULL)
4310 		pwd_drop(pwd);
4311 	FILEDESC_SLOCK(fdp);
4312 	lastfile = fdlastfile(fdp);
4313 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4314 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4315 			continue;
4316 #ifdef CAPABILITIES
4317 		rights = *cap_rights(fdp, i);
4318 #else /* !CAPABILITIES */
4319 		rights = cap_no_rights;
4320 #endif
4321 		/*
4322 		 * Create sysctl entry.  It is OK to drop the filedesc
4323 		 * lock inside of export_file_to_sb() as we will
4324 		 * re-validate and re-evaluate its properties when the
4325 		 * loop continues.
4326 		 */
4327 		error = export_file_to_sb(fp, i, &rights, efbuf);
4328 		if (error != 0 || efbuf->remainder == 0)
4329 			break;
4330 	}
4331 	FILEDESC_SUNLOCK(fdp);
4332 fail:
4333 	if (fdp != NULL)
4334 		fddrop(fdp);
4335 	if (pdp != NULL)
4336 		pddrop(pdp);
4337 	free(efbuf, M_TEMP);
4338 	return (error);
4339 }
4340 
4341 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4342 
4343 /*
4344  * Get per-process file descriptors for use by procstat(1), et al.
4345  */
4346 static int
4347 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4348 {
4349 	struct sbuf sb;
4350 	struct proc *p;
4351 	ssize_t maxlen;
4352 	int error, error2, *name;
4353 
4354 	name = (int *)arg1;
4355 
4356 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4357 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4358 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4359 	if (error != 0) {
4360 		sbuf_delete(&sb);
4361 		return (error);
4362 	}
4363 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4364 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4365 	    KERN_FILEDESC_PACK_KINFO);
4366 	error2 = sbuf_finish(&sb);
4367 	sbuf_delete(&sb);
4368 	return (error != 0 ? error : error2);
4369 }
4370 
4371 #ifdef COMPAT_FREEBSD7
4372 #ifdef KINFO_OFILE_SIZE
4373 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4374 #endif
4375 
4376 static void
4377 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4378 {
4379 
4380 	okif->kf_structsize = sizeof(*okif);
4381 	okif->kf_type = kif->kf_type;
4382 	okif->kf_fd = kif->kf_fd;
4383 	okif->kf_ref_count = kif->kf_ref_count;
4384 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4385 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4386 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4387 	okif->kf_offset = kif->kf_offset;
4388 	if (kif->kf_type == KF_TYPE_VNODE)
4389 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4390 	else
4391 		okif->kf_vnode_type = KF_VTYPE_VNON;
4392 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4393 	if (kif->kf_type == KF_TYPE_SOCKET) {
4394 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4395 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4396 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4397 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4398 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4399 	} else {
4400 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4401 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4402 	}
4403 }
4404 
4405 static int
4406 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4407     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4408 {
4409 	int error;
4410 
4411 	vrefact(vp);
4412 	PWDDESC_XUNLOCK(pdp);
4413 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4414 	kinfo_to_okinfo(kif, okif);
4415 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4416 	PWDDESC_XLOCK(pdp);
4417 	return (error);
4418 }
4419 
4420 /*
4421  * Get per-process file descriptors for use by procstat(1), et al.
4422  */
4423 static int
4424 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4425 {
4426 	struct kinfo_ofile *okif;
4427 	struct kinfo_file *kif;
4428 	struct filedesc *fdp;
4429 	struct pwddesc *pdp;
4430 	struct pwd *pwd;
4431 	int error, i, lastfile, *name;
4432 	struct file *fp;
4433 	struct proc *p;
4434 
4435 	name = (int *)arg1;
4436 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4437 	if (error != 0)
4438 		return (error);
4439 	fdp = fdhold(p);
4440 	if (fdp != NULL)
4441 		pdp = pdhold(p);
4442 	PROC_UNLOCK(p);
4443 	if (fdp == NULL || pdp == NULL) {
4444 		if (fdp != NULL)
4445 			fddrop(fdp);
4446 		return (ENOENT);
4447 	}
4448 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4449 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4450 	PWDDESC_XLOCK(pdp);
4451 	pwd = pwd_hold_pwddesc(pdp);
4452 	if (pwd != NULL) {
4453 		if (pwd->pwd_cdir != NULL)
4454 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4455 			    okif, pdp, req);
4456 		if (pwd->pwd_rdir != NULL)
4457 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4458 			    okif, pdp, req);
4459 		if (pwd->pwd_jdir != NULL)
4460 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4461 			    okif, pdp, req);
4462 	}
4463 	PWDDESC_XUNLOCK(pdp);
4464 	if (pwd != NULL)
4465 		pwd_drop(pwd);
4466 	FILEDESC_SLOCK(fdp);
4467 	lastfile = fdlastfile(fdp);
4468 	for (i = 0; refcount_load(&fdp->fd_refcnt) > 0 && i <= lastfile; i++) {
4469 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4470 			continue;
4471 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4472 		    KERN_FILEDESC_PACK_KINFO);
4473 		FILEDESC_SUNLOCK(fdp);
4474 		kinfo_to_okinfo(kif, okif);
4475 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4476 		FILEDESC_SLOCK(fdp);
4477 		if (error)
4478 			break;
4479 	}
4480 	FILEDESC_SUNLOCK(fdp);
4481 	fddrop(fdp);
4482 	pddrop(pdp);
4483 	free(kif, M_TEMP);
4484 	free(okif, M_TEMP);
4485 	return (0);
4486 }
4487 
4488 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4489     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4490     "Process ofiledesc entries");
4491 #endif	/* COMPAT_FREEBSD7 */
4492 
4493 int
4494 vntype_to_kinfo(int vtype)
4495 {
4496 	struct {
4497 		int	vtype;
4498 		int	kf_vtype;
4499 	} vtypes_table[] = {
4500 		{ VBAD, KF_VTYPE_VBAD },
4501 		{ VBLK, KF_VTYPE_VBLK },
4502 		{ VCHR, KF_VTYPE_VCHR },
4503 		{ VDIR, KF_VTYPE_VDIR },
4504 		{ VFIFO, KF_VTYPE_VFIFO },
4505 		{ VLNK, KF_VTYPE_VLNK },
4506 		{ VNON, KF_VTYPE_VNON },
4507 		{ VREG, KF_VTYPE_VREG },
4508 		{ VSOCK, KF_VTYPE_VSOCK }
4509 	};
4510 	unsigned int i;
4511 
4512 	/*
4513 	 * Perform vtype translation.
4514 	 */
4515 	for (i = 0; i < nitems(vtypes_table); i++)
4516 		if (vtypes_table[i].vtype == vtype)
4517 			return (vtypes_table[i].kf_vtype);
4518 
4519 	return (KF_VTYPE_UNKNOWN);
4520 }
4521 
4522 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4523     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4524     "Process filedesc entries");
4525 
4526 /*
4527  * Store a process current working directory information to sbuf.
4528  *
4529  * Takes a locked proc as argument, and returns with the proc unlocked.
4530  */
4531 int
4532 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4533 {
4534 	struct pwddesc *pdp;
4535 	struct pwd *pwd;
4536 	struct export_fd_buf *efbuf;
4537 	struct vnode *cdir;
4538 	int error;
4539 
4540 	PROC_LOCK_ASSERT(p, MA_OWNED);
4541 
4542 	pdp = pdhold(p);
4543 	PROC_UNLOCK(p);
4544 	if (pdp == NULL)
4545 		return (EINVAL);
4546 
4547 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4548 	efbuf->pdp = pdp;
4549 	efbuf->sb = sb;
4550 	efbuf->remainder = maxlen;
4551 
4552 	PWDDESC_XLOCK(pdp);
4553 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4554 	cdir = pwd->pwd_cdir;
4555 	if (cdir == NULL) {
4556 		error = EINVAL;
4557 	} else {
4558 		vrefact(cdir);
4559 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4560 	}
4561 	PWDDESC_XUNLOCK(pdp);
4562 	pddrop(pdp);
4563 	free(efbuf, M_TEMP);
4564 	return (error);
4565 }
4566 
4567 /*
4568  * Get per-process current working directory.
4569  */
4570 static int
4571 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4572 {
4573 	struct sbuf sb;
4574 	struct proc *p;
4575 	ssize_t maxlen;
4576 	int error, error2, *name;
4577 
4578 	name = (int *)arg1;
4579 
4580 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4581 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4582 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4583 	if (error != 0) {
4584 		sbuf_delete(&sb);
4585 		return (error);
4586 	}
4587 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4588 	error = kern_proc_cwd_out(p, &sb, maxlen);
4589 	error2 = sbuf_finish(&sb);
4590 	sbuf_delete(&sb);
4591 	return (error != 0 ? error : error2);
4592 }
4593 
4594 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4595     sysctl_kern_proc_cwd, "Process current working directory");
4596 
4597 #ifdef DDB
4598 /*
4599  * For the purposes of debugging, generate a human-readable string for the
4600  * file type.
4601  */
4602 static const char *
4603 file_type_to_name(short type)
4604 {
4605 
4606 	switch (type) {
4607 	case 0:
4608 		return ("zero");
4609 	case DTYPE_VNODE:
4610 		return ("vnode");
4611 	case DTYPE_SOCKET:
4612 		return ("socket");
4613 	case DTYPE_PIPE:
4614 		return ("pipe");
4615 	case DTYPE_FIFO:
4616 		return ("fifo");
4617 	case DTYPE_KQUEUE:
4618 		return ("kqueue");
4619 	case DTYPE_CRYPTO:
4620 		return ("crypto");
4621 	case DTYPE_MQUEUE:
4622 		return ("mqueue");
4623 	case DTYPE_SHM:
4624 		return ("shm");
4625 	case DTYPE_SEM:
4626 		return ("ksem");
4627 	case DTYPE_PTS:
4628 		return ("pts");
4629 	case DTYPE_DEV:
4630 		return ("dev");
4631 	case DTYPE_PROCDESC:
4632 		return ("proc");
4633 	case DTYPE_EVENTFD:
4634 		return ("eventfd");
4635 	case DTYPE_LINUXTFD:
4636 		return ("ltimer");
4637 	default:
4638 		return ("unkn");
4639 	}
4640 }
4641 
4642 /*
4643  * For the purposes of debugging, identify a process (if any, perhaps one of
4644  * many) that references the passed file in its file descriptor array. Return
4645  * NULL if none.
4646  */
4647 static struct proc *
4648 file_to_first_proc(struct file *fp)
4649 {
4650 	struct filedesc *fdp;
4651 	struct proc *p;
4652 	int n;
4653 
4654 	FOREACH_PROC_IN_SYSTEM(p) {
4655 		if (p->p_state == PRS_NEW)
4656 			continue;
4657 		fdp = p->p_fd;
4658 		if (fdp == NULL)
4659 			continue;
4660 		for (n = 0; n < fdp->fd_nfiles; n++) {
4661 			if (fp == fdp->fd_ofiles[n].fde_file)
4662 				return (p);
4663 		}
4664 	}
4665 	return (NULL);
4666 }
4667 
4668 static void
4669 db_print_file(struct file *fp, int header)
4670 {
4671 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4672 	struct proc *p;
4673 
4674 	if (header)
4675 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4676 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4677 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4678 		    "FCmd");
4679 	p = file_to_first_proc(fp);
4680 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4681 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4682 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4683 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4684 
4685 #undef XPTRWIDTH
4686 }
4687 
4688 DB_SHOW_COMMAND(file, db_show_file)
4689 {
4690 	struct file *fp;
4691 
4692 	if (!have_addr) {
4693 		db_printf("usage: show file <addr>\n");
4694 		return;
4695 	}
4696 	fp = (struct file *)addr;
4697 	db_print_file(fp, 1);
4698 }
4699 
4700 DB_SHOW_COMMAND(files, db_show_files)
4701 {
4702 	struct filedesc *fdp;
4703 	struct file *fp;
4704 	struct proc *p;
4705 	int header;
4706 	int n;
4707 
4708 	header = 1;
4709 	FOREACH_PROC_IN_SYSTEM(p) {
4710 		if (p->p_state == PRS_NEW)
4711 			continue;
4712 		if ((fdp = p->p_fd) == NULL)
4713 			continue;
4714 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4715 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4716 				continue;
4717 			db_print_file(fp, header);
4718 			header = 0;
4719 		}
4720 	}
4721 }
4722 #endif
4723 
4724 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4725     &maxfilesperproc, 0, "Maximum files allowed open per process");
4726 
4727 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4728     &maxfiles, 0, "Maximum number of files");
4729 
4730 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4731     &openfiles, 0, "System-wide number of open files");
4732 
4733 /* ARGSUSED*/
4734 static void
4735 filelistinit(void *dummy)
4736 {
4737 
4738 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4739 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4740 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4741 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4742 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4743 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4744 	/*
4745 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4746 	 * the vnode zone.
4747 	 */
4748 	vfs_smr = uma_zone_get_smr(pwd_zone);
4749 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4750 }
4751 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4752 
4753 /*-------------------------------------------------------------------*/
4754 
4755 static int
4756 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4757     int flags, struct thread *td)
4758 {
4759 
4760 	return (EBADF);
4761 }
4762 
4763 static int
4764 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4765     struct thread *td)
4766 {
4767 
4768 	return (EINVAL);
4769 }
4770 
4771 static int
4772 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4773     struct thread *td)
4774 {
4775 
4776 	return (EBADF);
4777 }
4778 
4779 static int
4780 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4781     struct thread *td)
4782 {
4783 
4784 	return (0);
4785 }
4786 
4787 static int
4788 badfo_kqfilter(struct file *fp, struct knote *kn)
4789 {
4790 
4791 	return (EBADF);
4792 }
4793 
4794 static int
4795 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4796     struct thread *td)
4797 {
4798 
4799 	return (EBADF);
4800 }
4801 
4802 static int
4803 badfo_close(struct file *fp, struct thread *td)
4804 {
4805 
4806 	return (0);
4807 }
4808 
4809 static int
4810 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4811     struct thread *td)
4812 {
4813 
4814 	return (EBADF);
4815 }
4816 
4817 static int
4818 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4819     struct thread *td)
4820 {
4821 
4822 	return (EBADF);
4823 }
4824 
4825 static int
4826 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4827     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4828     struct thread *td)
4829 {
4830 
4831 	return (EBADF);
4832 }
4833 
4834 static int
4835 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4836 {
4837 
4838 	return (0);
4839 }
4840 
4841 struct fileops badfileops = {
4842 	.fo_read = badfo_readwrite,
4843 	.fo_write = badfo_readwrite,
4844 	.fo_truncate = badfo_truncate,
4845 	.fo_ioctl = badfo_ioctl,
4846 	.fo_poll = badfo_poll,
4847 	.fo_kqfilter = badfo_kqfilter,
4848 	.fo_stat = badfo_stat,
4849 	.fo_close = badfo_close,
4850 	.fo_chmod = badfo_chmod,
4851 	.fo_chown = badfo_chown,
4852 	.fo_sendfile = badfo_sendfile,
4853 	.fo_fill_kinfo = badfo_fill_kinfo,
4854 };
4855 
4856 int
4857 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4858     int flags, struct thread *td)
4859 {
4860 
4861 	return (EOPNOTSUPP);
4862 }
4863 
4864 int
4865 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4866     struct thread *td)
4867 {
4868 
4869 	return (EINVAL);
4870 }
4871 
4872 int
4873 invfo_ioctl(struct file *fp, u_long com, void *data,
4874     struct ucred *active_cred, struct thread *td)
4875 {
4876 
4877 	return (ENOTTY);
4878 }
4879 
4880 int
4881 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4882     struct thread *td)
4883 {
4884 
4885 	return (poll_no_poll(events));
4886 }
4887 
4888 int
4889 invfo_kqfilter(struct file *fp, struct knote *kn)
4890 {
4891 
4892 	return (EINVAL);
4893 }
4894 
4895 int
4896 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4897     struct thread *td)
4898 {
4899 
4900 	return (EINVAL);
4901 }
4902 
4903 int
4904 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4905     struct thread *td)
4906 {
4907 
4908 	return (EINVAL);
4909 }
4910 
4911 int
4912 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4913     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4914     struct thread *td)
4915 {
4916 
4917 	return (EINVAL);
4918 }
4919 
4920 /*-------------------------------------------------------------------*/
4921 
4922 /*
4923  * File Descriptor pseudo-device driver (/dev/fd/).
4924  *
4925  * Opening minor device N dup()s the file (if any) connected to file
4926  * descriptor N belonging to the calling process.  Note that this driver
4927  * consists of only the ``open()'' routine, because all subsequent
4928  * references to this file will be direct to the other driver.
4929  *
4930  * XXX: we could give this one a cloning event handler if necessary.
4931  */
4932 
4933 /* ARGSUSED */
4934 static int
4935 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4936 {
4937 
4938 	/*
4939 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4940 	 * the file descriptor being sought for duplication. The error
4941 	 * return ensures that the vnode for this device will be released
4942 	 * by vn_open. Open will detect this special error and take the
4943 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4944 	 * will simply report the error.
4945 	 */
4946 	td->td_dupfd = dev2unit(dev);
4947 	return (ENODEV);
4948 }
4949 
4950 static struct cdevsw fildesc_cdevsw = {
4951 	.d_version =	D_VERSION,
4952 	.d_open =	fdopen,
4953 	.d_name =	"FD",
4954 };
4955 
4956 static void
4957 fildesc_drvinit(void *unused)
4958 {
4959 	struct cdev *dev;
4960 
4961 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4962 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4963 	make_dev_alias(dev, "stdin");
4964 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4965 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4966 	make_dev_alias(dev, "stdout");
4967 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4968 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4969 	make_dev_alias(dev, "stderr");
4970 }
4971 
4972 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4973