xref: /freebsd/sys/kern/kern_descrip.c (revision 47dd1d1b619cc035b82b49a91a25544309ff95ae)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/stat.h>
73 #include <sys/sx.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/unistd.h>
78 #include <sys/user.h>
79 #include <sys/vnode.h>
80 #ifdef KTRACE
81 #include <sys/ktrace.h>
82 #endif
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
95     "file desc to leader structures");
96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
98 
99 MALLOC_DECLARE(M_FADVISE);
100 
101 static __read_mostly uma_zone_t file_zone;
102 static __read_mostly uma_zone_t filedesc0_zone;
103 
104 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
105 		    struct thread *td, int holdleaders);
106 static int	fd_first_free(struct filedesc *fdp, int low, int size);
107 static int	fd_last_used(struct filedesc *fdp, int size);
108 static void	fdgrowtable(struct filedesc *fdp, int nfd);
109 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void	fdunused(struct filedesc *fdp, int fd);
111 static void	fdused(struct filedesc *fdp, int fd);
112 static int	getmaxfd(struct thread *td);
113 static u_long	*filecaps_copy_prep(const struct filecaps *src);
114 static void	filecaps_copy_finish(const struct filecaps *src,
115 		    struct filecaps *dst, u_long *ioctls);
116 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
117 static void	filecaps_free_finish(u_long *ioctls);
118 
119 /*
120  * Each process has:
121  *
122  * - An array of open file descriptors (fd_ofiles)
123  * - An array of file flags (fd_ofileflags)
124  * - A bitmap recording which descriptors are in use (fd_map)
125  *
126  * A process starts out with NDFILE descriptors.  The value of NDFILE has
127  * been selected based the historical limit of 20 open files, and an
128  * assumption that the majority of processes, especially short-lived
129  * processes like shells, will never need more.
130  *
131  * If this initial allocation is exhausted, a larger descriptor table and
132  * map are allocated dynamically, and the pointers in the process's struct
133  * filedesc are updated to point to those.  This is repeated every time
134  * the process runs out of file descriptors (provided it hasn't hit its
135  * resource limit).
136  *
137  * Since threads may hold references to individual descriptor table
138  * entries, the tables are never freed.  Instead, they are placed on a
139  * linked list and freed only when the struct filedesc is released.
140  */
141 #define NDFILE		20
142 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
143 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x)	((x) / NDENTRIES)
145 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
147 
148 /*
149  * SLIST entry used to keep track of ofiles which must be reclaimed when
150  * the process exits.
151  */
152 struct freetable {
153 	struct fdescenttbl *ft_table;
154 	SLIST_ENTRY(freetable) ft_next;
155 };
156 
157 /*
158  * Initial allocation: a filedesc structure + the head of SLIST used to
159  * keep track of old ofiles + enough space for NDFILE descriptors.
160  */
161 
162 struct fdescenttbl0 {
163 	int	fdt_nfiles;
164 	struct	filedescent fdt_ofiles[NDFILE];
165 };
166 
167 struct filedesc0 {
168 	struct filedesc fd_fd;
169 	SLIST_HEAD(, freetable) fd_free;
170 	struct	fdescenttbl0 fd_dfiles;
171 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
172 };
173 
174 /*
175  * Descriptor management.
176  */
177 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
178 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
180 
181 /*
182  * If low >= size, just return low. Otherwise find the first zero bit in the
183  * given bitmap, starting at low and not exceeding size - 1. Return size if
184  * not found.
185  */
186 static int
187 fd_first_free(struct filedesc *fdp, int low, int size)
188 {
189 	NDSLOTTYPE *map = fdp->fd_map;
190 	NDSLOTTYPE mask;
191 	int off, maxoff;
192 
193 	if (low >= size)
194 		return (low);
195 
196 	off = NDSLOT(low);
197 	if (low % NDENTRIES) {
198 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
199 		if ((mask &= ~map[off]) != 0UL)
200 			return (off * NDENTRIES + ffsl(mask) - 1);
201 		++off;
202 	}
203 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
204 		if (map[off] != ~0UL)
205 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
206 	return (size);
207 }
208 
209 /*
210  * Find the highest non-zero bit in the given bitmap, starting at 0 and
211  * not exceeding size - 1. Return -1 if not found.
212  */
213 static int
214 fd_last_used(struct filedesc *fdp, int size)
215 {
216 	NDSLOTTYPE *map = fdp->fd_map;
217 	NDSLOTTYPE mask;
218 	int off, minoff;
219 
220 	off = NDSLOT(size);
221 	if (size % NDENTRIES) {
222 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
223 		if ((mask &= map[off]) != 0)
224 			return (off * NDENTRIES + flsl(mask) - 1);
225 		--off;
226 	}
227 	for (minoff = NDSLOT(0); off >= minoff; --off)
228 		if (map[off] != 0)
229 			return (off * NDENTRIES + flsl(map[off]) - 1);
230 	return (-1);
231 }
232 
233 static int
234 fdisused(struct filedesc *fdp, int fd)
235 {
236 
237 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
238 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
239 
240 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
241 }
242 
243 /*
244  * Mark a file descriptor as used.
245  */
246 static void
247 fdused_init(struct filedesc *fdp, int fd)
248 {
249 
250 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
251 
252 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
253 }
254 
255 static void
256 fdused(struct filedesc *fdp, int fd)
257 {
258 
259 	FILEDESC_XLOCK_ASSERT(fdp);
260 
261 	fdused_init(fdp, fd);
262 	if (fd > fdp->fd_lastfile)
263 		fdp->fd_lastfile = fd;
264 	if (fd == fdp->fd_freefile)
265 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
266 }
267 
268 /*
269  * Mark a file descriptor as unused.
270  */
271 static void
272 fdunused(struct filedesc *fdp, int fd)
273 {
274 
275 	FILEDESC_XLOCK_ASSERT(fdp);
276 
277 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
278 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
279 	    ("fd=%d is still in use", fd));
280 
281 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
282 	if (fd < fdp->fd_freefile)
283 		fdp->fd_freefile = fd;
284 	if (fd == fdp->fd_lastfile)
285 		fdp->fd_lastfile = fd_last_used(fdp, fd);
286 }
287 
288 /*
289  * Free a file descriptor.
290  *
291  * Avoid some work if fdp is about to be destroyed.
292  */
293 static inline void
294 fdefree_last(struct filedescent *fde)
295 {
296 
297 	filecaps_free(&fde->fde_caps);
298 }
299 
300 static inline void
301 fdfree(struct filedesc *fdp, int fd)
302 {
303 	struct filedescent *fde;
304 
305 	fde = &fdp->fd_ofiles[fd];
306 #ifdef CAPABILITIES
307 	seq_write_begin(&fde->fde_seq);
308 #endif
309 	fde->fde_file = NULL;
310 #ifdef CAPABILITIES
311 	seq_write_end(&fde->fde_seq);
312 #endif
313 	fdefree_last(fde);
314 	fdunused(fdp, fd);
315 }
316 
317 void
318 pwd_ensure_dirs(void)
319 {
320 	struct filedesc *fdp;
321 
322 	fdp = curproc->p_fd;
323 	FILEDESC_XLOCK(fdp);
324 	if (fdp->fd_cdir == NULL) {
325 		fdp->fd_cdir = rootvnode;
326 		vrefact(rootvnode);
327 	}
328 	if (fdp->fd_rdir == NULL) {
329 		fdp->fd_rdir = rootvnode;
330 		vrefact(rootvnode);
331 	}
332 	FILEDESC_XUNLOCK(fdp);
333 }
334 
335 /*
336  * System calls on descriptors.
337  */
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
340 	int	dummy;
341 };
342 #endif
343 /* ARGSUSED */
344 int
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
346 {
347 #ifdef	RACCT
348 	uint64_t lim;
349 #endif
350 
351 	td->td_retval[0] =
352 	    min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
353 #ifdef	RACCT
354 	PROC_LOCK(td->td_proc);
355 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
356 	PROC_UNLOCK(td->td_proc);
357 	if (lim < td->td_retval[0])
358 		td->td_retval[0] = lim;
359 #endif
360 	return (0);
361 }
362 
363 /*
364  * Duplicate a file descriptor to a particular value.
365  *
366  * Note: keep in mind that a potential race condition exists when closing
367  * descriptors from a shared descriptor table (via rfork).
368  */
369 #ifndef _SYS_SYSPROTO_H_
370 struct dup2_args {
371 	u_int	from;
372 	u_int	to;
373 };
374 #endif
375 /* ARGSUSED */
376 int
377 sys_dup2(struct thread *td, struct dup2_args *uap)
378 {
379 
380 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
381 }
382 
383 /*
384  * Duplicate a file descriptor.
385  */
386 #ifndef _SYS_SYSPROTO_H_
387 struct dup_args {
388 	u_int	fd;
389 };
390 #endif
391 /* ARGSUSED */
392 int
393 sys_dup(struct thread *td, struct dup_args *uap)
394 {
395 
396 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
397 }
398 
399 /*
400  * The file control system call.
401  */
402 #ifndef _SYS_SYSPROTO_H_
403 struct fcntl_args {
404 	int	fd;
405 	int	cmd;
406 	long	arg;
407 };
408 #endif
409 /* ARGSUSED */
410 int
411 sys_fcntl(struct thread *td, struct fcntl_args *uap)
412 {
413 
414 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
415 }
416 
417 int
418 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
419 {
420 	struct flock fl;
421 	struct __oflock ofl;
422 	intptr_t arg1;
423 	int error, newcmd;
424 
425 	error = 0;
426 	newcmd = cmd;
427 	switch (cmd) {
428 	case F_OGETLK:
429 	case F_OSETLK:
430 	case F_OSETLKW:
431 		/*
432 		 * Convert old flock structure to new.
433 		 */
434 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
435 		fl.l_start = ofl.l_start;
436 		fl.l_len = ofl.l_len;
437 		fl.l_pid = ofl.l_pid;
438 		fl.l_type = ofl.l_type;
439 		fl.l_whence = ofl.l_whence;
440 		fl.l_sysid = 0;
441 
442 		switch (cmd) {
443 		case F_OGETLK:
444 			newcmd = F_GETLK;
445 			break;
446 		case F_OSETLK:
447 			newcmd = F_SETLK;
448 			break;
449 		case F_OSETLKW:
450 			newcmd = F_SETLKW;
451 			break;
452 		}
453 		arg1 = (intptr_t)&fl;
454 		break;
455 	case F_GETLK:
456 	case F_SETLK:
457 	case F_SETLKW:
458 	case F_SETLK_REMOTE:
459 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
460 		arg1 = (intptr_t)&fl;
461 		break;
462 	default:
463 		arg1 = arg;
464 		break;
465 	}
466 	if (error)
467 		return (error);
468 	error = kern_fcntl(td, fd, newcmd, arg1);
469 	if (error)
470 		return (error);
471 	if (cmd == F_OGETLK) {
472 		ofl.l_start = fl.l_start;
473 		ofl.l_len = fl.l_len;
474 		ofl.l_pid = fl.l_pid;
475 		ofl.l_type = fl.l_type;
476 		ofl.l_whence = fl.l_whence;
477 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
478 	} else if (cmd == F_GETLK) {
479 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
480 	}
481 	return (error);
482 }
483 
484 int
485 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
486 {
487 	struct filedesc *fdp;
488 	struct flock *flp;
489 	struct file *fp, *fp2;
490 	struct filedescent *fde;
491 	struct proc *p;
492 	struct vnode *vp;
493 	int error, flg, tmp;
494 	uint64_t bsize;
495 	off_t foffset;
496 
497 	error = 0;
498 	flg = F_POSIX;
499 	p = td->td_proc;
500 	fdp = p->p_fd;
501 
502 	AUDIT_ARG_FD(cmd);
503 	AUDIT_ARG_CMD(cmd);
504 	switch (cmd) {
505 	case F_DUPFD:
506 		tmp = arg;
507 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
508 		break;
509 
510 	case F_DUPFD_CLOEXEC:
511 		tmp = arg;
512 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
513 		break;
514 
515 	case F_DUP2FD:
516 		tmp = arg;
517 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
518 		break;
519 
520 	case F_DUP2FD_CLOEXEC:
521 		tmp = arg;
522 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
523 		break;
524 
525 	case F_GETFD:
526 		error = EBADF;
527 		FILEDESC_SLOCK(fdp);
528 		fde = fdeget_locked(fdp, fd);
529 		if (fde != NULL) {
530 			td->td_retval[0] =
531 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
532 			error = 0;
533 		}
534 		FILEDESC_SUNLOCK(fdp);
535 		break;
536 
537 	case F_SETFD:
538 		error = EBADF;
539 		FILEDESC_XLOCK(fdp);
540 		fde = fdeget_locked(fdp, fd);
541 		if (fde != NULL) {
542 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
543 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
544 			error = 0;
545 		}
546 		FILEDESC_XUNLOCK(fdp);
547 		break;
548 
549 	case F_GETFL:
550 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
551 		if (error != 0)
552 			break;
553 		td->td_retval[0] = OFLAGS(fp->f_flag);
554 		fdrop(fp, td);
555 		break;
556 
557 	case F_SETFL:
558 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
559 		if (error != 0)
560 			break;
561 		do {
562 			tmp = flg = fp->f_flag;
563 			tmp &= ~FCNTLFLAGS;
564 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
565 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
566 		tmp = fp->f_flag & FNONBLOCK;
567 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
568 		if (error != 0) {
569 			fdrop(fp, td);
570 			break;
571 		}
572 		tmp = fp->f_flag & FASYNC;
573 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
574 		if (error == 0) {
575 			fdrop(fp, td);
576 			break;
577 		}
578 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
579 		tmp = 0;
580 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
581 		fdrop(fp, td);
582 		break;
583 
584 	case F_GETOWN:
585 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
586 		if (error != 0)
587 			break;
588 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
589 		if (error == 0)
590 			td->td_retval[0] = tmp;
591 		fdrop(fp, td);
592 		break;
593 
594 	case F_SETOWN:
595 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
596 		if (error != 0)
597 			break;
598 		tmp = arg;
599 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
600 		fdrop(fp, td);
601 		break;
602 
603 	case F_SETLK_REMOTE:
604 		error = priv_check(td, PRIV_NFS_LOCKD);
605 		if (error)
606 			return (error);
607 		flg = F_REMOTE;
608 		goto do_setlk;
609 
610 	case F_SETLKW:
611 		flg |= F_WAIT;
612 		/* FALLTHROUGH F_SETLK */
613 
614 	case F_SETLK:
615 	do_setlk:
616 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
617 		if (error != 0)
618 			break;
619 		if (fp->f_type != DTYPE_VNODE) {
620 			error = EBADF;
621 			fdrop(fp, td);
622 			break;
623 		}
624 
625 		flp = (struct flock *)arg;
626 		if (flp->l_whence == SEEK_CUR) {
627 			foffset = foffset_get(fp);
628 			if (foffset < 0 ||
629 			    (flp->l_start > 0 &&
630 			     foffset > OFF_MAX - flp->l_start)) {
631 				error = EOVERFLOW;
632 				fdrop(fp, td);
633 				break;
634 			}
635 			flp->l_start += foffset;
636 		}
637 
638 		vp = fp->f_vnode;
639 		switch (flp->l_type) {
640 		case F_RDLCK:
641 			if ((fp->f_flag & FREAD) == 0) {
642 				error = EBADF;
643 				break;
644 			}
645 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
646 				PROC_LOCK(p->p_leader);
647 				p->p_leader->p_flag |= P_ADVLOCK;
648 				PROC_UNLOCK(p->p_leader);
649 			}
650 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
651 			    flp, flg);
652 			break;
653 		case F_WRLCK:
654 			if ((fp->f_flag & FWRITE) == 0) {
655 				error = EBADF;
656 				break;
657 			}
658 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
659 				PROC_LOCK(p->p_leader);
660 				p->p_leader->p_flag |= P_ADVLOCK;
661 				PROC_UNLOCK(p->p_leader);
662 			}
663 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
664 			    flp, flg);
665 			break;
666 		case F_UNLCK:
667 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
668 			    flp, flg);
669 			break;
670 		case F_UNLCKSYS:
671 			/*
672 			 * Temporary api for testing remote lock
673 			 * infrastructure.
674 			 */
675 			if (flg != F_REMOTE) {
676 				error = EINVAL;
677 				break;
678 			}
679 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
680 			    F_UNLCKSYS, flp, flg);
681 			break;
682 		default:
683 			error = EINVAL;
684 			break;
685 		}
686 		if (error != 0 || flp->l_type == F_UNLCK ||
687 		    flp->l_type == F_UNLCKSYS) {
688 			fdrop(fp, td);
689 			break;
690 		}
691 
692 		/*
693 		 * Check for a race with close.
694 		 *
695 		 * The vnode is now advisory locked (or unlocked, but this case
696 		 * is not really important) as the caller requested.
697 		 * We had to drop the filedesc lock, so we need to recheck if
698 		 * the descriptor is still valid, because if it was closed
699 		 * in the meantime we need to remove advisory lock from the
700 		 * vnode - close on any descriptor leading to an advisory
701 		 * locked vnode, removes that lock.
702 		 * We will return 0 on purpose in that case, as the result of
703 		 * successful advisory lock might have been externally visible
704 		 * already. This is fine - effectively we pretend to the caller
705 		 * that the closing thread was a bit slower and that the
706 		 * advisory lock succeeded before the close.
707 		 */
708 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
709 		if (error != 0) {
710 			fdrop(fp, td);
711 			break;
712 		}
713 		if (fp != fp2) {
714 			flp->l_whence = SEEK_SET;
715 			flp->l_start = 0;
716 			flp->l_len = 0;
717 			flp->l_type = F_UNLCK;
718 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
719 			    F_UNLCK, flp, F_POSIX);
720 		}
721 		fdrop(fp, td);
722 		fdrop(fp2, td);
723 		break;
724 
725 	case F_GETLK:
726 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
727 		if (error != 0)
728 			break;
729 		if (fp->f_type != DTYPE_VNODE) {
730 			error = EBADF;
731 			fdrop(fp, td);
732 			break;
733 		}
734 		flp = (struct flock *)arg;
735 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
736 		    flp->l_type != F_UNLCK) {
737 			error = EINVAL;
738 			fdrop(fp, td);
739 			break;
740 		}
741 		if (flp->l_whence == SEEK_CUR) {
742 			foffset = foffset_get(fp);
743 			if ((flp->l_start > 0 &&
744 			    foffset > OFF_MAX - flp->l_start) ||
745 			    (flp->l_start < 0 &&
746 			    foffset < OFF_MIN - flp->l_start)) {
747 				error = EOVERFLOW;
748 				fdrop(fp, td);
749 				break;
750 			}
751 			flp->l_start += foffset;
752 		}
753 		vp = fp->f_vnode;
754 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
755 		    F_POSIX);
756 		fdrop(fp, td);
757 		break;
758 
759 	case F_RDAHEAD:
760 		arg = arg ? 128 * 1024: 0;
761 		/* FALLTHROUGH */
762 	case F_READAHEAD:
763 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
764 		if (error != 0)
765 			break;
766 		if (fp->f_type != DTYPE_VNODE) {
767 			fdrop(fp, td);
768 			error = EBADF;
769 			break;
770 		}
771 		vp = fp->f_vnode;
772 		/*
773 		 * Exclusive lock synchronizes against f_seqcount reads and
774 		 * writes in sequential_heuristic().
775 		 */
776 		error = vn_lock(vp, LK_EXCLUSIVE);
777 		if (error != 0) {
778 			fdrop(fp, td);
779 			break;
780 		}
781 		if (arg >= 0) {
782 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
783 			fp->f_seqcount = (arg + bsize - 1) / bsize;
784 			atomic_set_int(&fp->f_flag, FRDAHEAD);
785 		} else {
786 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
787 		}
788 		VOP_UNLOCK(vp, 0);
789 		fdrop(fp, td);
790 		break;
791 
792 	default:
793 		error = EINVAL;
794 		break;
795 	}
796 	return (error);
797 }
798 
799 static int
800 getmaxfd(struct thread *td)
801 {
802 
803 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
804 }
805 
806 /*
807  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
808  */
809 int
810 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
811 {
812 	struct filedesc *fdp;
813 	struct filedescent *oldfde, *newfde;
814 	struct proc *p;
815 	struct file *delfp;
816 	u_long *oioctls, *nioctls;
817 	int error, maxfd;
818 
819 	p = td->td_proc;
820 	fdp = p->p_fd;
821 
822 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
823 	MPASS(mode < FDDUP_LASTMODE);
824 
825 	AUDIT_ARG_FD(old);
826 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
827 
828 	/*
829 	 * Verify we have a valid descriptor to dup from and possibly to
830 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
831 	 * return EINVAL when the new descriptor is out of bounds.
832 	 */
833 	if (old < 0)
834 		return (EBADF);
835 	if (new < 0)
836 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
837 	maxfd = getmaxfd(td);
838 	if (new >= maxfd)
839 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
840 
841 	error = EBADF;
842 	FILEDESC_XLOCK(fdp);
843 	if (fget_locked(fdp, old) == NULL)
844 		goto unlock;
845 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
846 		td->td_retval[0] = new;
847 		if (flags & FDDUP_FLAG_CLOEXEC)
848 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
849 		error = 0;
850 		goto unlock;
851 	}
852 
853 	/*
854 	 * If the caller specified a file descriptor, make sure the file
855 	 * table is large enough to hold it, and grab it.  Otherwise, just
856 	 * allocate a new descriptor the usual way.
857 	 */
858 	switch (mode) {
859 	case FDDUP_NORMAL:
860 	case FDDUP_FCNTL:
861 		if ((error = fdalloc(td, new, &new)) != 0)
862 			goto unlock;
863 		break;
864 	case FDDUP_MUSTREPLACE:
865 		/* Target file descriptor must exist. */
866 		if (fget_locked(fdp, new) == NULL)
867 			goto unlock;
868 		break;
869 	case FDDUP_FIXED:
870 		if (new >= fdp->fd_nfiles) {
871 			/*
872 			 * The resource limits are here instead of e.g.
873 			 * fdalloc(), because the file descriptor table may be
874 			 * shared between processes, so we can't really use
875 			 * racct_add()/racct_sub().  Instead of counting the
876 			 * number of actually allocated descriptors, just put
877 			 * the limit on the size of the file descriptor table.
878 			 */
879 #ifdef RACCT
880 			if (racct_enable) {
881 				PROC_LOCK(p);
882 				error = racct_set(p, RACCT_NOFILE, new + 1);
883 				PROC_UNLOCK(p);
884 				if (error != 0) {
885 					error = EMFILE;
886 					goto unlock;
887 				}
888 			}
889 #endif
890 			fdgrowtable_exp(fdp, new + 1);
891 		}
892 		if (!fdisused(fdp, new))
893 			fdused(fdp, new);
894 		break;
895 	default:
896 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
897 	}
898 
899 	KASSERT(old != new, ("new fd is same as old"));
900 
901 	oldfde = &fdp->fd_ofiles[old];
902 	fhold(oldfde->fde_file);
903 	newfde = &fdp->fd_ofiles[new];
904 	delfp = newfde->fde_file;
905 
906 	oioctls = filecaps_free_prep(&newfde->fde_caps);
907 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
908 
909 	/*
910 	 * Duplicate the source descriptor.
911 	 */
912 #ifdef CAPABILITIES
913 	seq_write_begin(&newfde->fde_seq);
914 #endif
915 	memcpy(newfde, oldfde, fde_change_size);
916 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
917 	    nioctls);
918 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
919 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
920 	else
921 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
922 #ifdef CAPABILITIES
923 	seq_write_end(&newfde->fde_seq);
924 #endif
925 	filecaps_free_finish(oioctls);
926 	td->td_retval[0] = new;
927 
928 	error = 0;
929 
930 	if (delfp != NULL) {
931 		(void) closefp(fdp, new, delfp, td, 1);
932 		FILEDESC_UNLOCK_ASSERT(fdp);
933 	} else {
934 unlock:
935 		FILEDESC_XUNLOCK(fdp);
936 	}
937 
938 	return (error);
939 }
940 
941 /*
942  * If sigio is on the list associated with a process or process group,
943  * disable signalling from the device, remove sigio from the list and
944  * free sigio.
945  */
946 void
947 funsetown(struct sigio **sigiop)
948 {
949 	struct sigio *sigio;
950 
951 	if (*sigiop == NULL)
952 		return;
953 	SIGIO_LOCK();
954 	sigio = *sigiop;
955 	if (sigio == NULL) {
956 		SIGIO_UNLOCK();
957 		return;
958 	}
959 	*(sigio->sio_myref) = NULL;
960 	if ((sigio)->sio_pgid < 0) {
961 		struct pgrp *pg = (sigio)->sio_pgrp;
962 		PGRP_LOCK(pg);
963 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
964 			    sigio, sio_pgsigio);
965 		PGRP_UNLOCK(pg);
966 	} else {
967 		struct proc *p = (sigio)->sio_proc;
968 		PROC_LOCK(p);
969 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
970 			    sigio, sio_pgsigio);
971 		PROC_UNLOCK(p);
972 	}
973 	SIGIO_UNLOCK();
974 	crfree(sigio->sio_ucred);
975 	free(sigio, M_SIGIO);
976 }
977 
978 /*
979  * Free a list of sigio structures.
980  * We only need to lock the SIGIO_LOCK because we have made ourselves
981  * inaccessible to callers of fsetown and therefore do not need to lock
982  * the proc or pgrp struct for the list manipulation.
983  */
984 void
985 funsetownlst(struct sigiolst *sigiolst)
986 {
987 	struct proc *p;
988 	struct pgrp *pg;
989 	struct sigio *sigio;
990 
991 	sigio = SLIST_FIRST(sigiolst);
992 	if (sigio == NULL)
993 		return;
994 	p = NULL;
995 	pg = NULL;
996 
997 	/*
998 	 * Every entry of the list should belong
999 	 * to a single proc or pgrp.
1000 	 */
1001 	if (sigio->sio_pgid < 0) {
1002 		pg = sigio->sio_pgrp;
1003 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1004 	} else /* if (sigio->sio_pgid > 0) */ {
1005 		p = sigio->sio_proc;
1006 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1007 	}
1008 
1009 	SIGIO_LOCK();
1010 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1011 		*(sigio->sio_myref) = NULL;
1012 		if (pg != NULL) {
1013 			KASSERT(sigio->sio_pgid < 0,
1014 			    ("Proc sigio in pgrp sigio list"));
1015 			KASSERT(sigio->sio_pgrp == pg,
1016 			    ("Bogus pgrp in sigio list"));
1017 			PGRP_LOCK(pg);
1018 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1019 			    sio_pgsigio);
1020 			PGRP_UNLOCK(pg);
1021 		} else /* if (p != NULL) */ {
1022 			KASSERT(sigio->sio_pgid > 0,
1023 			    ("Pgrp sigio in proc sigio list"));
1024 			KASSERT(sigio->sio_proc == p,
1025 			    ("Bogus proc in sigio list"));
1026 			PROC_LOCK(p);
1027 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1028 			    sio_pgsigio);
1029 			PROC_UNLOCK(p);
1030 		}
1031 		SIGIO_UNLOCK();
1032 		crfree(sigio->sio_ucred);
1033 		free(sigio, M_SIGIO);
1034 		SIGIO_LOCK();
1035 	}
1036 	SIGIO_UNLOCK();
1037 }
1038 
1039 /*
1040  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1041  *
1042  * After permission checking, add a sigio structure to the sigio list for
1043  * the process or process group.
1044  */
1045 int
1046 fsetown(pid_t pgid, struct sigio **sigiop)
1047 {
1048 	struct proc *proc;
1049 	struct pgrp *pgrp;
1050 	struct sigio *sigio;
1051 	int ret;
1052 
1053 	if (pgid == 0) {
1054 		funsetown(sigiop);
1055 		return (0);
1056 	}
1057 
1058 	ret = 0;
1059 
1060 	/* Allocate and fill in the new sigio out of locks. */
1061 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1062 	sigio->sio_pgid = pgid;
1063 	sigio->sio_ucred = crhold(curthread->td_ucred);
1064 	sigio->sio_myref = sigiop;
1065 
1066 	sx_slock(&proctree_lock);
1067 	if (pgid > 0) {
1068 		proc = pfind(pgid);
1069 		if (proc == NULL) {
1070 			ret = ESRCH;
1071 			goto fail;
1072 		}
1073 
1074 		/*
1075 		 * Policy - Don't allow a process to FSETOWN a process
1076 		 * in another session.
1077 		 *
1078 		 * Remove this test to allow maximum flexibility or
1079 		 * restrict FSETOWN to the current process or process
1080 		 * group for maximum safety.
1081 		 */
1082 		PROC_UNLOCK(proc);
1083 		if (proc->p_session != curthread->td_proc->p_session) {
1084 			ret = EPERM;
1085 			goto fail;
1086 		}
1087 
1088 		pgrp = NULL;
1089 	} else /* if (pgid < 0) */ {
1090 		pgrp = pgfind(-pgid);
1091 		if (pgrp == NULL) {
1092 			ret = ESRCH;
1093 			goto fail;
1094 		}
1095 		PGRP_UNLOCK(pgrp);
1096 
1097 		/*
1098 		 * Policy - Don't allow a process to FSETOWN a process
1099 		 * in another session.
1100 		 *
1101 		 * Remove this test to allow maximum flexibility or
1102 		 * restrict FSETOWN to the current process or process
1103 		 * group for maximum safety.
1104 		 */
1105 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1106 			ret = EPERM;
1107 			goto fail;
1108 		}
1109 
1110 		proc = NULL;
1111 	}
1112 	funsetown(sigiop);
1113 	if (pgid > 0) {
1114 		PROC_LOCK(proc);
1115 		/*
1116 		 * Since funsetownlst() is called without the proctree
1117 		 * locked, we need to check for P_WEXIT.
1118 		 * XXX: is ESRCH correct?
1119 		 */
1120 		if ((proc->p_flag & P_WEXIT) != 0) {
1121 			PROC_UNLOCK(proc);
1122 			ret = ESRCH;
1123 			goto fail;
1124 		}
1125 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1126 		sigio->sio_proc = proc;
1127 		PROC_UNLOCK(proc);
1128 	} else {
1129 		PGRP_LOCK(pgrp);
1130 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1131 		sigio->sio_pgrp = pgrp;
1132 		PGRP_UNLOCK(pgrp);
1133 	}
1134 	sx_sunlock(&proctree_lock);
1135 	SIGIO_LOCK();
1136 	*sigiop = sigio;
1137 	SIGIO_UNLOCK();
1138 	return (0);
1139 
1140 fail:
1141 	sx_sunlock(&proctree_lock);
1142 	crfree(sigio->sio_ucred);
1143 	free(sigio, M_SIGIO);
1144 	return (ret);
1145 }
1146 
1147 /*
1148  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1149  */
1150 pid_t
1151 fgetown(sigiop)
1152 	struct sigio **sigiop;
1153 {
1154 	pid_t pgid;
1155 
1156 	SIGIO_LOCK();
1157 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1158 	SIGIO_UNLOCK();
1159 	return (pgid);
1160 }
1161 
1162 /*
1163  * Function drops the filedesc lock on return.
1164  */
1165 static int
1166 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1167     int holdleaders)
1168 {
1169 	int error;
1170 
1171 	FILEDESC_XLOCK_ASSERT(fdp);
1172 
1173 	if (holdleaders) {
1174 		if (td->td_proc->p_fdtol != NULL) {
1175 			/*
1176 			 * Ask fdfree() to sleep to ensure that all relevant
1177 			 * process leaders can be traversed in closef().
1178 			 */
1179 			fdp->fd_holdleaderscount++;
1180 		} else {
1181 			holdleaders = 0;
1182 		}
1183 	}
1184 
1185 	/*
1186 	 * We now hold the fp reference that used to be owned by the
1187 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1188 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1189 	 * added, and deleteing a knote for the new fd.
1190 	 */
1191 	knote_fdclose(td, fd);
1192 
1193 	/*
1194 	 * We need to notify mqueue if the object is of type mqueue.
1195 	 */
1196 	if (fp->f_type == DTYPE_MQUEUE)
1197 		mq_fdclose(td, fd, fp);
1198 	FILEDESC_XUNLOCK(fdp);
1199 
1200 	error = closef(fp, td);
1201 	if (holdleaders) {
1202 		FILEDESC_XLOCK(fdp);
1203 		fdp->fd_holdleaderscount--;
1204 		if (fdp->fd_holdleaderscount == 0 &&
1205 		    fdp->fd_holdleaderswakeup != 0) {
1206 			fdp->fd_holdleaderswakeup = 0;
1207 			wakeup(&fdp->fd_holdleaderscount);
1208 		}
1209 		FILEDESC_XUNLOCK(fdp);
1210 	}
1211 	return (error);
1212 }
1213 
1214 /*
1215  * Close a file descriptor.
1216  */
1217 #ifndef _SYS_SYSPROTO_H_
1218 struct close_args {
1219 	int     fd;
1220 };
1221 #endif
1222 /* ARGSUSED */
1223 int
1224 sys_close(struct thread *td, struct close_args *uap)
1225 {
1226 
1227 	return (kern_close(td, uap->fd));
1228 }
1229 
1230 int
1231 kern_close(struct thread *td, int fd)
1232 {
1233 	struct filedesc *fdp;
1234 	struct file *fp;
1235 
1236 	fdp = td->td_proc->p_fd;
1237 
1238 	AUDIT_SYSCLOSE(td, fd);
1239 
1240 	FILEDESC_XLOCK(fdp);
1241 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1242 		FILEDESC_XUNLOCK(fdp);
1243 		return (EBADF);
1244 	}
1245 	fdfree(fdp, fd);
1246 
1247 	/* closefp() drops the FILEDESC lock for us. */
1248 	return (closefp(fdp, fd, fp, td, 1));
1249 }
1250 
1251 /*
1252  * Close open file descriptors.
1253  */
1254 #ifndef _SYS_SYSPROTO_H_
1255 struct closefrom_args {
1256 	int	lowfd;
1257 };
1258 #endif
1259 /* ARGSUSED */
1260 int
1261 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1262 {
1263 	struct filedesc *fdp;
1264 	int fd;
1265 
1266 	fdp = td->td_proc->p_fd;
1267 	AUDIT_ARG_FD(uap->lowfd);
1268 
1269 	/*
1270 	 * Treat negative starting file descriptor values identical to
1271 	 * closefrom(0) which closes all files.
1272 	 */
1273 	if (uap->lowfd < 0)
1274 		uap->lowfd = 0;
1275 	FILEDESC_SLOCK(fdp);
1276 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1277 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1278 			FILEDESC_SUNLOCK(fdp);
1279 			(void)kern_close(td, fd);
1280 			FILEDESC_SLOCK(fdp);
1281 		}
1282 	}
1283 	FILEDESC_SUNLOCK(fdp);
1284 	return (0);
1285 }
1286 
1287 #if defined(COMPAT_43)
1288 /*
1289  * Return status information about a file descriptor.
1290  */
1291 #ifndef _SYS_SYSPROTO_H_
1292 struct ofstat_args {
1293 	int	fd;
1294 	struct	ostat *sb;
1295 };
1296 #endif
1297 /* ARGSUSED */
1298 int
1299 ofstat(struct thread *td, struct ofstat_args *uap)
1300 {
1301 	struct ostat oub;
1302 	struct stat ub;
1303 	int error;
1304 
1305 	error = kern_fstat(td, uap->fd, &ub);
1306 	if (error == 0) {
1307 		cvtstat(&ub, &oub);
1308 		error = copyout(&oub, uap->sb, sizeof(oub));
1309 	}
1310 	return (error);
1311 }
1312 #endif /* COMPAT_43 */
1313 
1314 #if defined(COMPAT_FREEBSD11)
1315 int
1316 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1317 {
1318 	struct stat sb;
1319 	struct freebsd11_stat osb;
1320 	int error;
1321 
1322 	error = kern_fstat(td, uap->fd, &sb);
1323 	if (error != 0)
1324 		return (error);
1325 	error = freebsd11_cvtstat(&sb, &osb);
1326 	if (error == 0)
1327 		error = copyout(&osb, uap->sb, sizeof(osb));
1328 	return (error);
1329 }
1330 #endif	/* COMPAT_FREEBSD11 */
1331 
1332 /*
1333  * Return status information about a file descriptor.
1334  */
1335 #ifndef _SYS_SYSPROTO_H_
1336 struct fstat_args {
1337 	int	fd;
1338 	struct	stat *sb;
1339 };
1340 #endif
1341 /* ARGSUSED */
1342 int
1343 sys_fstat(struct thread *td, struct fstat_args *uap)
1344 {
1345 	struct stat ub;
1346 	int error;
1347 
1348 	error = kern_fstat(td, uap->fd, &ub);
1349 	if (error == 0)
1350 		error = copyout(&ub, uap->sb, sizeof(ub));
1351 	return (error);
1352 }
1353 
1354 int
1355 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1356 {
1357 	struct file *fp;
1358 	int error;
1359 
1360 	AUDIT_ARG_FD(fd);
1361 
1362 	error = fget(td, fd, &cap_fstat_rights, &fp);
1363 	if (error != 0)
1364 		return (error);
1365 
1366 	AUDIT_ARG_FILE(td->td_proc, fp);
1367 
1368 	error = fo_stat(fp, sbp, td->td_ucred, td);
1369 	fdrop(fp, td);
1370 #ifdef __STAT_TIME_T_EXT
1371 	if (error == 0) {
1372 		sbp->st_atim_ext = 0;
1373 		sbp->st_mtim_ext = 0;
1374 		sbp->st_ctim_ext = 0;
1375 		sbp->st_btim_ext = 0;
1376 	}
1377 #endif
1378 #ifdef KTRACE
1379 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1380 		ktrstat(sbp);
1381 #endif
1382 	return (error);
1383 }
1384 
1385 #if defined(COMPAT_FREEBSD11)
1386 /*
1387  * Return status information about a file descriptor.
1388  */
1389 #ifndef _SYS_SYSPROTO_H_
1390 struct freebsd11_nfstat_args {
1391 	int	fd;
1392 	struct	nstat *sb;
1393 };
1394 #endif
1395 /* ARGSUSED */
1396 int
1397 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1398 {
1399 	struct nstat nub;
1400 	struct stat ub;
1401 	int error;
1402 
1403 	error = kern_fstat(td, uap->fd, &ub);
1404 	if (error == 0) {
1405 		freebsd11_cvtnstat(&ub, &nub);
1406 		error = copyout(&nub, uap->sb, sizeof(nub));
1407 	}
1408 	return (error);
1409 }
1410 #endif /* COMPAT_FREEBSD11 */
1411 
1412 /*
1413  * Return pathconf information about a file descriptor.
1414  */
1415 #ifndef _SYS_SYSPROTO_H_
1416 struct fpathconf_args {
1417 	int	fd;
1418 	int	name;
1419 };
1420 #endif
1421 /* ARGSUSED */
1422 int
1423 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1424 {
1425 	long value;
1426 	int error;
1427 
1428 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1429 	if (error == 0)
1430 		td->td_retval[0] = value;
1431 	return (error);
1432 }
1433 
1434 int
1435 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1436 {
1437 	struct file *fp;
1438 	struct vnode *vp;
1439 	int error;
1440 
1441 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1442 	if (error != 0)
1443 		return (error);
1444 
1445 	if (name == _PC_ASYNC_IO) {
1446 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1447 		goto out;
1448 	}
1449 	vp = fp->f_vnode;
1450 	if (vp != NULL) {
1451 		vn_lock(vp, LK_SHARED | LK_RETRY);
1452 		error = VOP_PATHCONF(vp, name, valuep);
1453 		VOP_UNLOCK(vp, 0);
1454 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1455 		if (name != _PC_PIPE_BUF) {
1456 			error = EINVAL;
1457 		} else {
1458 			*valuep = PIPE_BUF;
1459 			error = 0;
1460 		}
1461 	} else {
1462 		error = EOPNOTSUPP;
1463 	}
1464 out:
1465 	fdrop(fp, td);
1466 	return (error);
1467 }
1468 
1469 /*
1470  * Initialize filecaps structure.
1471  */
1472 void
1473 filecaps_init(struct filecaps *fcaps)
1474 {
1475 
1476 	bzero(fcaps, sizeof(*fcaps));
1477 	fcaps->fc_nioctls = -1;
1478 }
1479 
1480 /*
1481  * Copy filecaps structure allocating memory for ioctls array if needed.
1482  *
1483  * The last parameter indicates whether the fdtable is locked. If it is not and
1484  * ioctls are encountered, copying fails and the caller must lock the table.
1485  *
1486  * Note that if the table was not locked, the caller has to check the relevant
1487  * sequence counter to determine whether the operation was successful.
1488  */
1489 bool
1490 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1491 {
1492 	size_t size;
1493 
1494 	if (src->fc_ioctls != NULL && !locked)
1495 		return (false);
1496 	memcpy(dst, src, sizeof(*src));
1497 	if (src->fc_ioctls == NULL)
1498 		return (true);
1499 
1500 	KASSERT(src->fc_nioctls > 0,
1501 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1502 
1503 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1504 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1505 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1506 	return (true);
1507 }
1508 
1509 static u_long *
1510 filecaps_copy_prep(const struct filecaps *src)
1511 {
1512 	u_long *ioctls;
1513 	size_t size;
1514 
1515 	if (src->fc_ioctls == NULL)
1516 		return (NULL);
1517 
1518 	KASSERT(src->fc_nioctls > 0,
1519 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1520 
1521 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1522 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1523 	return (ioctls);
1524 }
1525 
1526 static void
1527 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1528     u_long *ioctls)
1529 {
1530 	size_t size;
1531 
1532 	*dst = *src;
1533 	if (src->fc_ioctls == NULL) {
1534 		MPASS(ioctls == NULL);
1535 		return;
1536 	}
1537 
1538 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1539 	dst->fc_ioctls = ioctls;
1540 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1541 }
1542 
1543 /*
1544  * Move filecaps structure to the new place and clear the old place.
1545  */
1546 void
1547 filecaps_move(struct filecaps *src, struct filecaps *dst)
1548 {
1549 
1550 	*dst = *src;
1551 	bzero(src, sizeof(*src));
1552 }
1553 
1554 /*
1555  * Fill the given filecaps structure with full rights.
1556  */
1557 static void
1558 filecaps_fill(struct filecaps *fcaps)
1559 {
1560 
1561 	CAP_ALL(&fcaps->fc_rights);
1562 	fcaps->fc_ioctls = NULL;
1563 	fcaps->fc_nioctls = -1;
1564 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1565 }
1566 
1567 /*
1568  * Free memory allocated within filecaps structure.
1569  */
1570 void
1571 filecaps_free(struct filecaps *fcaps)
1572 {
1573 
1574 	free(fcaps->fc_ioctls, M_FILECAPS);
1575 	bzero(fcaps, sizeof(*fcaps));
1576 }
1577 
1578 static u_long *
1579 filecaps_free_prep(struct filecaps *fcaps)
1580 {
1581 	u_long *ioctls;
1582 
1583 	ioctls = fcaps->fc_ioctls;
1584 	bzero(fcaps, sizeof(*fcaps));
1585 	return (ioctls);
1586 }
1587 
1588 static void
1589 filecaps_free_finish(u_long *ioctls)
1590 {
1591 
1592 	free(ioctls, M_FILECAPS);
1593 }
1594 
1595 /*
1596  * Validate the given filecaps structure.
1597  */
1598 static void
1599 filecaps_validate(const struct filecaps *fcaps, const char *func)
1600 {
1601 
1602 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1603 	    ("%s: invalid rights", func));
1604 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1605 	    ("%s: invalid fcntls", func));
1606 	KASSERT(fcaps->fc_fcntls == 0 ||
1607 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1608 	    ("%s: fcntls without CAP_FCNTL", func));
1609 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1610 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1611 	    ("%s: invalid ioctls", func));
1612 	KASSERT(fcaps->fc_nioctls == 0 ||
1613 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1614 	    ("%s: ioctls without CAP_IOCTL", func));
1615 }
1616 
1617 static void
1618 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1619 {
1620 	int nfd1;
1621 
1622 	FILEDESC_XLOCK_ASSERT(fdp);
1623 
1624 	nfd1 = fdp->fd_nfiles * 2;
1625 	if (nfd1 < nfd)
1626 		nfd1 = nfd;
1627 	fdgrowtable(fdp, nfd1);
1628 }
1629 
1630 /*
1631  * Grow the file table to accommodate (at least) nfd descriptors.
1632  */
1633 static void
1634 fdgrowtable(struct filedesc *fdp, int nfd)
1635 {
1636 	struct filedesc0 *fdp0;
1637 	struct freetable *ft;
1638 	struct fdescenttbl *ntable;
1639 	struct fdescenttbl *otable;
1640 	int nnfiles, onfiles;
1641 	NDSLOTTYPE *nmap, *omap;
1642 
1643 	/*
1644 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1645 	 * growing it to accommodate for the one we are going to copy from. There
1646 	 * is no need to have a lock on this one as it's not visible to anyone.
1647 	 */
1648 	if (fdp->fd_lastfile != -1)
1649 		FILEDESC_XLOCK_ASSERT(fdp);
1650 
1651 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1652 
1653 	/* save old values */
1654 	onfiles = fdp->fd_nfiles;
1655 	otable = fdp->fd_files;
1656 	omap = fdp->fd_map;
1657 
1658 	/* compute the size of the new table */
1659 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1660 	if (nnfiles <= onfiles)
1661 		/* the table is already large enough */
1662 		return;
1663 
1664 	/*
1665 	 * Allocate a new table.  We need enough space for the number of
1666 	 * entries, file entries themselves and the struct freetable we will use
1667 	 * when we decommission the table and place it on the freelist.
1668 	 * We place the struct freetable in the middle so we don't have
1669 	 * to worry about padding.
1670 	 */
1671 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1672 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1673 	    sizeof(struct freetable),
1674 	    M_FILEDESC, M_ZERO | M_WAITOK);
1675 	/* copy the old data */
1676 	ntable->fdt_nfiles = nnfiles;
1677 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1678 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1679 
1680 	/*
1681 	 * Allocate a new map only if the old is not large enough.  It will
1682 	 * grow at a slower rate than the table as it can map more
1683 	 * entries than the table can hold.
1684 	 */
1685 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1686 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1687 		    M_ZERO | M_WAITOK);
1688 		/* copy over the old data and update the pointer */
1689 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1690 		fdp->fd_map = nmap;
1691 	}
1692 
1693 	/*
1694 	 * Make sure that ntable is correctly initialized before we replace
1695 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1696 	 * data.
1697 	 */
1698 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1699 
1700 	/*
1701 	 * Do not free the old file table, as some threads may still
1702 	 * reference entries within it.  Instead, place it on a freelist
1703 	 * which will be processed when the struct filedesc is released.
1704 	 *
1705 	 * Note that if onfiles == NDFILE, we're dealing with the original
1706 	 * static allocation contained within (struct filedesc0 *)fdp,
1707 	 * which must not be freed.
1708 	 */
1709 	if (onfiles > NDFILE) {
1710 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1711 		fdp0 = (struct filedesc0 *)fdp;
1712 		ft->ft_table = otable;
1713 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1714 	}
1715 	/*
1716 	 * The map does not have the same possibility of threads still
1717 	 * holding references to it.  So always free it as long as it
1718 	 * does not reference the original static allocation.
1719 	 */
1720 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1721 		free(omap, M_FILEDESC);
1722 }
1723 
1724 /*
1725  * Allocate a file descriptor for the process.
1726  */
1727 int
1728 fdalloc(struct thread *td, int minfd, int *result)
1729 {
1730 	struct proc *p = td->td_proc;
1731 	struct filedesc *fdp = p->p_fd;
1732 	int fd, maxfd, allocfd;
1733 #ifdef RACCT
1734 	int error;
1735 #endif
1736 
1737 	FILEDESC_XLOCK_ASSERT(fdp);
1738 
1739 	if (fdp->fd_freefile > minfd)
1740 		minfd = fdp->fd_freefile;
1741 
1742 	maxfd = getmaxfd(td);
1743 
1744 	/*
1745 	 * Search the bitmap for a free descriptor starting at minfd.
1746 	 * If none is found, grow the file table.
1747 	 */
1748 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1749 	if (fd >= maxfd)
1750 		return (EMFILE);
1751 	if (fd >= fdp->fd_nfiles) {
1752 		allocfd = min(fd * 2, maxfd);
1753 #ifdef RACCT
1754 		if (racct_enable) {
1755 			PROC_LOCK(p);
1756 			error = racct_set(p, RACCT_NOFILE, allocfd);
1757 			PROC_UNLOCK(p);
1758 			if (error != 0)
1759 				return (EMFILE);
1760 		}
1761 #endif
1762 		/*
1763 		 * fd is already equal to first free descriptor >= minfd, so
1764 		 * we only need to grow the table and we are done.
1765 		 */
1766 		fdgrowtable_exp(fdp, allocfd);
1767 	}
1768 
1769 	/*
1770 	 * Perform some sanity checks, then mark the file descriptor as
1771 	 * used and return it to the caller.
1772 	 */
1773 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1774 	    ("invalid descriptor %d", fd));
1775 	KASSERT(!fdisused(fdp, fd),
1776 	    ("fd_first_free() returned non-free descriptor"));
1777 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1778 	    ("file descriptor isn't free"));
1779 	fdused(fdp, fd);
1780 	*result = fd;
1781 	return (0);
1782 }
1783 
1784 /*
1785  * Allocate n file descriptors for the process.
1786  */
1787 int
1788 fdallocn(struct thread *td, int minfd, int *fds, int n)
1789 {
1790 	struct proc *p = td->td_proc;
1791 	struct filedesc *fdp = p->p_fd;
1792 	int i;
1793 
1794 	FILEDESC_XLOCK_ASSERT(fdp);
1795 
1796 	for (i = 0; i < n; i++)
1797 		if (fdalloc(td, 0, &fds[i]) != 0)
1798 			break;
1799 
1800 	if (i < n) {
1801 		for (i--; i >= 0; i--)
1802 			fdunused(fdp, fds[i]);
1803 		return (EMFILE);
1804 	}
1805 
1806 	return (0);
1807 }
1808 
1809 /*
1810  * Create a new open file structure and allocate a file descriptor for the
1811  * process that refers to it.  We add one reference to the file for the
1812  * descriptor table and one reference for resultfp. This is to prevent us
1813  * being preempted and the entry in the descriptor table closed after we
1814  * release the FILEDESC lock.
1815  */
1816 int
1817 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1818     struct filecaps *fcaps)
1819 {
1820 	struct file *fp;
1821 	int error, fd;
1822 
1823 	error = falloc_noinstall(td, &fp);
1824 	if (error)
1825 		return (error);		/* no reference held on error */
1826 
1827 	error = finstall(td, fp, &fd, flags, fcaps);
1828 	if (error) {
1829 		fdrop(fp, td);		/* one reference (fp only) */
1830 		return (error);
1831 	}
1832 
1833 	if (resultfp != NULL)
1834 		*resultfp = fp;		/* copy out result */
1835 	else
1836 		fdrop(fp, td);		/* release local reference */
1837 
1838 	if (resultfd != NULL)
1839 		*resultfd = fd;
1840 
1841 	return (0);
1842 }
1843 
1844 /*
1845  * Create a new open file structure without allocating a file descriptor.
1846  */
1847 int
1848 falloc_noinstall(struct thread *td, struct file **resultfp)
1849 {
1850 	struct file *fp;
1851 	int maxuserfiles = maxfiles - (maxfiles / 20);
1852 	int openfiles_new;
1853 	static struct timeval lastfail;
1854 	static int curfail;
1855 
1856 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1857 
1858 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1859 	if ((openfiles_new >= maxuserfiles &&
1860 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1861 	    openfiles_new >= maxfiles) {
1862 		atomic_subtract_int(&openfiles, 1);
1863 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1864 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1865 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1866 		}
1867 		return (ENFILE);
1868 	}
1869 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1870 	refcount_init(&fp->f_count, 1);
1871 	fp->f_cred = crhold(td->td_ucred);
1872 	fp->f_ops = &badfileops;
1873 	*resultfp = fp;
1874 	return (0);
1875 }
1876 
1877 /*
1878  * Install a file in a file descriptor table.
1879  */
1880 void
1881 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1882     struct filecaps *fcaps)
1883 {
1884 	struct filedescent *fde;
1885 
1886 	MPASS(fp != NULL);
1887 	if (fcaps != NULL)
1888 		filecaps_validate(fcaps, __func__);
1889 	FILEDESC_XLOCK_ASSERT(fdp);
1890 
1891 	fde = &fdp->fd_ofiles[fd];
1892 #ifdef CAPABILITIES
1893 	seq_write_begin(&fde->fde_seq);
1894 #endif
1895 	fde->fde_file = fp;
1896 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1897 	if (fcaps != NULL)
1898 		filecaps_move(fcaps, &fde->fde_caps);
1899 	else
1900 		filecaps_fill(&fde->fde_caps);
1901 #ifdef CAPABILITIES
1902 	seq_write_end(&fde->fde_seq);
1903 #endif
1904 }
1905 
1906 int
1907 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1908     struct filecaps *fcaps)
1909 {
1910 	struct filedesc *fdp = td->td_proc->p_fd;
1911 	int error;
1912 
1913 	MPASS(fd != NULL);
1914 
1915 	FILEDESC_XLOCK(fdp);
1916 	if ((error = fdalloc(td, 0, fd))) {
1917 		FILEDESC_XUNLOCK(fdp);
1918 		return (error);
1919 	}
1920 	fhold(fp);
1921 	_finstall(fdp, fp, *fd, flags, fcaps);
1922 	FILEDESC_XUNLOCK(fdp);
1923 	return (0);
1924 }
1925 
1926 /*
1927  * Build a new filedesc structure from another.
1928  * Copy the current, root, and jail root vnode references.
1929  *
1930  * If fdp is not NULL, return with it shared locked.
1931  */
1932 struct filedesc *
1933 fdinit(struct filedesc *fdp, bool prepfiles)
1934 {
1935 	struct filedesc0 *newfdp0;
1936 	struct filedesc *newfdp;
1937 
1938 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1939 	newfdp = &newfdp0->fd_fd;
1940 
1941 	/* Create the file descriptor table. */
1942 	FILEDESC_LOCK_INIT(newfdp);
1943 	refcount_init(&newfdp->fd_refcnt, 1);
1944 	refcount_init(&newfdp->fd_holdcnt, 1);
1945 	newfdp->fd_cmask = CMASK;
1946 	newfdp->fd_map = newfdp0->fd_dmap;
1947 	newfdp->fd_lastfile = -1;
1948 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1949 	newfdp->fd_files->fdt_nfiles = NDFILE;
1950 
1951 	if (fdp == NULL)
1952 		return (newfdp);
1953 
1954 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1955 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1956 
1957 	FILEDESC_SLOCK(fdp);
1958 	newfdp->fd_cdir = fdp->fd_cdir;
1959 	if (newfdp->fd_cdir)
1960 		vrefact(newfdp->fd_cdir);
1961 	newfdp->fd_rdir = fdp->fd_rdir;
1962 	if (newfdp->fd_rdir)
1963 		vrefact(newfdp->fd_rdir);
1964 	newfdp->fd_jdir = fdp->fd_jdir;
1965 	if (newfdp->fd_jdir)
1966 		vrefact(newfdp->fd_jdir);
1967 
1968 	if (!prepfiles) {
1969 		FILEDESC_SUNLOCK(fdp);
1970 	} else {
1971 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1972 			FILEDESC_SUNLOCK(fdp);
1973 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1974 			FILEDESC_SLOCK(fdp);
1975 		}
1976 	}
1977 
1978 	return (newfdp);
1979 }
1980 
1981 static struct filedesc *
1982 fdhold(struct proc *p)
1983 {
1984 	struct filedesc *fdp;
1985 
1986 	PROC_LOCK_ASSERT(p, MA_OWNED);
1987 	fdp = p->p_fd;
1988 	if (fdp != NULL)
1989 		refcount_acquire(&fdp->fd_holdcnt);
1990 	return (fdp);
1991 }
1992 
1993 static void
1994 fddrop(struct filedesc *fdp)
1995 {
1996 
1997 	if (fdp->fd_holdcnt > 1) {
1998 		if (refcount_release(&fdp->fd_holdcnt) == 0)
1999 			return;
2000 	}
2001 
2002 	FILEDESC_LOCK_DESTROY(fdp);
2003 	uma_zfree(filedesc0_zone, fdp);
2004 }
2005 
2006 /*
2007  * Share a filedesc structure.
2008  */
2009 struct filedesc *
2010 fdshare(struct filedesc *fdp)
2011 {
2012 
2013 	refcount_acquire(&fdp->fd_refcnt);
2014 	return (fdp);
2015 }
2016 
2017 /*
2018  * Unshare a filedesc structure, if necessary by making a copy
2019  */
2020 void
2021 fdunshare(struct thread *td)
2022 {
2023 	struct filedesc *tmp;
2024 	struct proc *p = td->td_proc;
2025 
2026 	if (p->p_fd->fd_refcnt == 1)
2027 		return;
2028 
2029 	tmp = fdcopy(p->p_fd);
2030 	fdescfree(td);
2031 	p->p_fd = tmp;
2032 }
2033 
2034 void
2035 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2036 {
2037 
2038 	fdescfree(td);
2039 	td->td_proc->p_fd = fdp;
2040 }
2041 
2042 /*
2043  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2044  * this is to ease callers, not catch errors.
2045  */
2046 struct filedesc *
2047 fdcopy(struct filedesc *fdp)
2048 {
2049 	struct filedesc *newfdp;
2050 	struct filedescent *nfde, *ofde;
2051 	int i;
2052 
2053 	MPASS(fdp != NULL);
2054 
2055 	newfdp = fdinit(fdp, true);
2056 	/* copy all passable descriptors (i.e. not kqueue) */
2057 	newfdp->fd_freefile = -1;
2058 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
2059 		ofde = &fdp->fd_ofiles[i];
2060 		if (ofde->fde_file == NULL ||
2061 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2062 			if (newfdp->fd_freefile == -1)
2063 				newfdp->fd_freefile = i;
2064 			continue;
2065 		}
2066 		nfde = &newfdp->fd_ofiles[i];
2067 		*nfde = *ofde;
2068 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2069 		fhold(nfde->fde_file);
2070 		fdused_init(newfdp, i);
2071 		newfdp->fd_lastfile = i;
2072 	}
2073 	if (newfdp->fd_freefile == -1)
2074 		newfdp->fd_freefile = i;
2075 	newfdp->fd_cmask = fdp->fd_cmask;
2076 	FILEDESC_SUNLOCK(fdp);
2077 	return (newfdp);
2078 }
2079 
2080 /*
2081  * Copies a filedesc structure, while remapping all file descriptors
2082  * stored inside using a translation table.
2083  *
2084  * File descriptors are copied over to the new file descriptor table,
2085  * regardless of whether the close-on-exec flag is set.
2086  */
2087 int
2088 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2089     struct filedesc **ret)
2090 {
2091 	struct filedesc *newfdp;
2092 	struct filedescent *nfde, *ofde;
2093 	int error, i;
2094 
2095 	MPASS(fdp != NULL);
2096 
2097 	newfdp = fdinit(fdp, true);
2098 	if (nfds > fdp->fd_lastfile + 1) {
2099 		/* New table cannot be larger than the old one. */
2100 		error = E2BIG;
2101 		goto bad;
2102 	}
2103 	/* Copy all passable descriptors (i.e. not kqueue). */
2104 	newfdp->fd_freefile = nfds;
2105 	for (i = 0; i < nfds; ++i) {
2106 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2107 			/* File descriptor out of bounds. */
2108 			error = EBADF;
2109 			goto bad;
2110 		}
2111 		ofde = &fdp->fd_ofiles[fds[i]];
2112 		if (ofde->fde_file == NULL) {
2113 			/* Unused file descriptor. */
2114 			error = EBADF;
2115 			goto bad;
2116 		}
2117 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2118 			/* File descriptor cannot be passed. */
2119 			error = EINVAL;
2120 			goto bad;
2121 		}
2122 		nfde = &newfdp->fd_ofiles[i];
2123 		*nfde = *ofde;
2124 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2125 		fhold(nfde->fde_file);
2126 		fdused_init(newfdp, i);
2127 		newfdp->fd_lastfile = i;
2128 	}
2129 	newfdp->fd_cmask = fdp->fd_cmask;
2130 	FILEDESC_SUNLOCK(fdp);
2131 	*ret = newfdp;
2132 	return (0);
2133 bad:
2134 	FILEDESC_SUNLOCK(fdp);
2135 	fdescfree_remapped(newfdp);
2136 	return (error);
2137 }
2138 
2139 /*
2140  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2141  * one of processes using it exits) and the table used to be shared.
2142  */
2143 static void
2144 fdclearlocks(struct thread *td)
2145 {
2146 	struct filedesc *fdp;
2147 	struct filedesc_to_leader *fdtol;
2148 	struct flock lf;
2149 	struct file *fp;
2150 	struct proc *p;
2151 	struct vnode *vp;
2152 	int i;
2153 
2154 	p = td->td_proc;
2155 	fdp = p->p_fd;
2156 	fdtol = p->p_fdtol;
2157 	MPASS(fdtol != NULL);
2158 
2159 	FILEDESC_XLOCK(fdp);
2160 	KASSERT(fdtol->fdl_refcount > 0,
2161 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2162 	    fdtol->fdl_refcount));
2163 	if (fdtol->fdl_refcount == 1 &&
2164 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2165 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2166 			fp = fdp->fd_ofiles[i].fde_file;
2167 			if (fp == NULL || fp->f_type != DTYPE_VNODE)
2168 				continue;
2169 			fhold(fp);
2170 			FILEDESC_XUNLOCK(fdp);
2171 			lf.l_whence = SEEK_SET;
2172 			lf.l_start = 0;
2173 			lf.l_len = 0;
2174 			lf.l_type = F_UNLCK;
2175 			vp = fp->f_vnode;
2176 			(void) VOP_ADVLOCK(vp,
2177 			    (caddr_t)p->p_leader, F_UNLCK,
2178 			    &lf, F_POSIX);
2179 			FILEDESC_XLOCK(fdp);
2180 			fdrop(fp, td);
2181 		}
2182 	}
2183 retry:
2184 	if (fdtol->fdl_refcount == 1) {
2185 		if (fdp->fd_holdleaderscount > 0 &&
2186 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2187 			/*
2188 			 * close() or kern_dup() has cleared a reference
2189 			 * in a shared file descriptor table.
2190 			 */
2191 			fdp->fd_holdleaderswakeup = 1;
2192 			sx_sleep(&fdp->fd_holdleaderscount,
2193 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2194 			goto retry;
2195 		}
2196 		if (fdtol->fdl_holdcount > 0) {
2197 			/*
2198 			 * Ensure that fdtol->fdl_leader remains
2199 			 * valid in closef().
2200 			 */
2201 			fdtol->fdl_wakeup = 1;
2202 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2203 			    "fdlhold", 0);
2204 			goto retry;
2205 		}
2206 	}
2207 	fdtol->fdl_refcount--;
2208 	if (fdtol->fdl_refcount == 0 &&
2209 	    fdtol->fdl_holdcount == 0) {
2210 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2211 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2212 	} else
2213 		fdtol = NULL;
2214 	p->p_fdtol = NULL;
2215 	FILEDESC_XUNLOCK(fdp);
2216 	if (fdtol != NULL)
2217 		free(fdtol, M_FILEDESC_TO_LEADER);
2218 }
2219 
2220 /*
2221  * Release a filedesc structure.
2222  */
2223 static void
2224 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2225 {
2226 	struct filedesc0 *fdp0;
2227 	struct freetable *ft, *tft;
2228 	struct filedescent *fde;
2229 	struct file *fp;
2230 	int i;
2231 
2232 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2233 		fde = &fdp->fd_ofiles[i];
2234 		fp = fde->fde_file;
2235 		if (fp != NULL) {
2236 			fdefree_last(fde);
2237 			if (needclose)
2238 				(void) closef(fp, td);
2239 			else
2240 				fdrop(fp, td);
2241 		}
2242 	}
2243 
2244 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2245 		free(fdp->fd_map, M_FILEDESC);
2246 	if (fdp->fd_nfiles > NDFILE)
2247 		free(fdp->fd_files, M_FILEDESC);
2248 
2249 	fdp0 = (struct filedesc0 *)fdp;
2250 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2251 		free(ft->ft_table, M_FILEDESC);
2252 
2253 	fddrop(fdp);
2254 }
2255 
2256 void
2257 fdescfree(struct thread *td)
2258 {
2259 	struct proc *p;
2260 	struct filedesc *fdp;
2261 	struct vnode *cdir, *jdir, *rdir;
2262 
2263 	p = td->td_proc;
2264 	fdp = p->p_fd;
2265 	MPASS(fdp != NULL);
2266 
2267 #ifdef RACCT
2268 	if (racct_enable) {
2269 		PROC_LOCK(p);
2270 		racct_set(p, RACCT_NOFILE, 0);
2271 		PROC_UNLOCK(p);
2272 	}
2273 #endif
2274 
2275 	if (p->p_fdtol != NULL)
2276 		fdclearlocks(td);
2277 
2278 	PROC_LOCK(p);
2279 	p->p_fd = NULL;
2280 	PROC_UNLOCK(p);
2281 
2282 	if (refcount_release(&fdp->fd_refcnt) == 0)
2283 		return;
2284 
2285 	FILEDESC_XLOCK(fdp);
2286 	cdir = fdp->fd_cdir;
2287 	fdp->fd_cdir = NULL;
2288 	rdir = fdp->fd_rdir;
2289 	fdp->fd_rdir = NULL;
2290 	jdir = fdp->fd_jdir;
2291 	fdp->fd_jdir = NULL;
2292 	FILEDESC_XUNLOCK(fdp);
2293 
2294 	if (cdir != NULL)
2295 		vrele(cdir);
2296 	if (rdir != NULL)
2297 		vrele(rdir);
2298 	if (jdir != NULL)
2299 		vrele(jdir);
2300 
2301 	fdescfree_fds(td, fdp, 1);
2302 }
2303 
2304 void
2305 fdescfree_remapped(struct filedesc *fdp)
2306 {
2307 
2308 	if (fdp->fd_cdir != NULL)
2309 		vrele(fdp->fd_cdir);
2310 	if (fdp->fd_rdir != NULL)
2311 		vrele(fdp->fd_rdir);
2312 	if (fdp->fd_jdir != NULL)
2313 		vrele(fdp->fd_jdir);
2314 
2315 	fdescfree_fds(curthread, fdp, 0);
2316 }
2317 
2318 /*
2319  * For setugid programs, we don't want to people to use that setugidness
2320  * to generate error messages which write to a file which otherwise would
2321  * otherwise be off-limits to the process.  We check for filesystems where
2322  * the vnode can change out from under us after execve (like [lin]procfs).
2323  *
2324  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2325  * sufficient.  We also don't check for setugidness since we know we are.
2326  */
2327 static bool
2328 is_unsafe(struct file *fp)
2329 {
2330 	struct vnode *vp;
2331 
2332 	if (fp->f_type != DTYPE_VNODE)
2333 		return (false);
2334 
2335 	vp = fp->f_vnode;
2336 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2337 }
2338 
2339 /*
2340  * Make this setguid thing safe, if at all possible.
2341  */
2342 void
2343 fdsetugidsafety(struct thread *td)
2344 {
2345 	struct filedesc *fdp;
2346 	struct file *fp;
2347 	int i;
2348 
2349 	fdp = td->td_proc->p_fd;
2350 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2351 	MPASS(fdp->fd_nfiles >= 3);
2352 	for (i = 0; i <= 2; i++) {
2353 		fp = fdp->fd_ofiles[i].fde_file;
2354 		if (fp != NULL && is_unsafe(fp)) {
2355 			FILEDESC_XLOCK(fdp);
2356 			knote_fdclose(td, i);
2357 			/*
2358 			 * NULL-out descriptor prior to close to avoid
2359 			 * a race while close blocks.
2360 			 */
2361 			fdfree(fdp, i);
2362 			FILEDESC_XUNLOCK(fdp);
2363 			(void) closef(fp, td);
2364 		}
2365 	}
2366 }
2367 
2368 /*
2369  * If a specific file object occupies a specific file descriptor, close the
2370  * file descriptor entry and drop a reference on the file object.  This is a
2371  * convenience function to handle a subsequent error in a function that calls
2372  * falloc() that handles the race that another thread might have closed the
2373  * file descriptor out from under the thread creating the file object.
2374  */
2375 void
2376 fdclose(struct thread *td, struct file *fp, int idx)
2377 {
2378 	struct filedesc *fdp = td->td_proc->p_fd;
2379 
2380 	FILEDESC_XLOCK(fdp);
2381 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2382 		fdfree(fdp, idx);
2383 		FILEDESC_XUNLOCK(fdp);
2384 		fdrop(fp, td);
2385 	} else
2386 		FILEDESC_XUNLOCK(fdp);
2387 }
2388 
2389 /*
2390  * Close any files on exec?
2391  */
2392 void
2393 fdcloseexec(struct thread *td)
2394 {
2395 	struct filedesc *fdp;
2396 	struct filedescent *fde;
2397 	struct file *fp;
2398 	int i;
2399 
2400 	fdp = td->td_proc->p_fd;
2401 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2402 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2403 		fde = &fdp->fd_ofiles[i];
2404 		fp = fde->fde_file;
2405 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2406 		    (fde->fde_flags & UF_EXCLOSE))) {
2407 			FILEDESC_XLOCK(fdp);
2408 			fdfree(fdp, i);
2409 			(void) closefp(fdp, i, fp, td, 0);
2410 			FILEDESC_UNLOCK_ASSERT(fdp);
2411 		}
2412 	}
2413 }
2414 
2415 /*
2416  * It is unsafe for set[ug]id processes to be started with file
2417  * descriptors 0..2 closed, as these descriptors are given implicit
2418  * significance in the Standard C library.  fdcheckstd() will create a
2419  * descriptor referencing /dev/null for each of stdin, stdout, and
2420  * stderr that is not already open.
2421  */
2422 int
2423 fdcheckstd(struct thread *td)
2424 {
2425 	struct filedesc *fdp;
2426 	register_t save;
2427 	int i, error, devnull;
2428 
2429 	fdp = td->td_proc->p_fd;
2430 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2431 	MPASS(fdp->fd_nfiles >= 3);
2432 	devnull = -1;
2433 	for (i = 0; i <= 2; i++) {
2434 		if (fdp->fd_ofiles[i].fde_file != NULL)
2435 			continue;
2436 
2437 		save = td->td_retval[0];
2438 		if (devnull != -1) {
2439 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2440 		} else {
2441 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2442 			    UIO_SYSSPACE, O_RDWR, 0);
2443 			if (error == 0) {
2444 				devnull = td->td_retval[0];
2445 				KASSERT(devnull == i, ("we didn't get our fd"));
2446 			}
2447 		}
2448 		td->td_retval[0] = save;
2449 		if (error != 0)
2450 			return (error);
2451 	}
2452 	return (0);
2453 }
2454 
2455 /*
2456  * Internal form of close.  Decrement reference count on file structure.
2457  * Note: td may be NULL when closing a file that was being passed in a
2458  * message.
2459  *
2460  * XXXRW: Giant is not required for the caller, but often will be held; this
2461  * makes it moderately likely the Giant will be recursed in the VFS case.
2462  */
2463 int
2464 closef(struct file *fp, struct thread *td)
2465 {
2466 	struct vnode *vp;
2467 	struct flock lf;
2468 	struct filedesc_to_leader *fdtol;
2469 	struct filedesc *fdp;
2470 
2471 	/*
2472 	 * POSIX record locking dictates that any close releases ALL
2473 	 * locks owned by this process.  This is handled by setting
2474 	 * a flag in the unlock to free ONLY locks obeying POSIX
2475 	 * semantics, and not to free BSD-style file locks.
2476 	 * If the descriptor was in a message, POSIX-style locks
2477 	 * aren't passed with the descriptor, and the thread pointer
2478 	 * will be NULL.  Callers should be careful only to pass a
2479 	 * NULL thread pointer when there really is no owning
2480 	 * context that might have locks, or the locks will be
2481 	 * leaked.
2482 	 */
2483 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2484 		vp = fp->f_vnode;
2485 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2486 			lf.l_whence = SEEK_SET;
2487 			lf.l_start = 0;
2488 			lf.l_len = 0;
2489 			lf.l_type = F_UNLCK;
2490 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2491 			    F_UNLCK, &lf, F_POSIX);
2492 		}
2493 		fdtol = td->td_proc->p_fdtol;
2494 		if (fdtol != NULL) {
2495 			/*
2496 			 * Handle special case where file descriptor table is
2497 			 * shared between multiple process leaders.
2498 			 */
2499 			fdp = td->td_proc->p_fd;
2500 			FILEDESC_XLOCK(fdp);
2501 			for (fdtol = fdtol->fdl_next;
2502 			    fdtol != td->td_proc->p_fdtol;
2503 			    fdtol = fdtol->fdl_next) {
2504 				if ((fdtol->fdl_leader->p_flag &
2505 				    P_ADVLOCK) == 0)
2506 					continue;
2507 				fdtol->fdl_holdcount++;
2508 				FILEDESC_XUNLOCK(fdp);
2509 				lf.l_whence = SEEK_SET;
2510 				lf.l_start = 0;
2511 				lf.l_len = 0;
2512 				lf.l_type = F_UNLCK;
2513 				vp = fp->f_vnode;
2514 				(void) VOP_ADVLOCK(vp,
2515 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2516 				    F_POSIX);
2517 				FILEDESC_XLOCK(fdp);
2518 				fdtol->fdl_holdcount--;
2519 				if (fdtol->fdl_holdcount == 0 &&
2520 				    fdtol->fdl_wakeup != 0) {
2521 					fdtol->fdl_wakeup = 0;
2522 					wakeup(fdtol);
2523 				}
2524 			}
2525 			FILEDESC_XUNLOCK(fdp);
2526 		}
2527 	}
2528 	return (fdrop(fp, td));
2529 }
2530 
2531 /*
2532  * Initialize the file pointer with the specified properties.
2533  *
2534  * The ops are set with release semantics to be certain that the flags, type,
2535  * and data are visible when ops is.  This is to prevent ops methods from being
2536  * called with bad data.
2537  */
2538 void
2539 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2540 {
2541 	fp->f_data = data;
2542 	fp->f_flag = flag;
2543 	fp->f_type = type;
2544 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2545 }
2546 
2547 int
2548 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2549     struct file **fpp, struct filecaps *havecapsp)
2550 {
2551 	struct filedescent *fde;
2552 	int error;
2553 
2554 	FILEDESC_LOCK_ASSERT(fdp);
2555 
2556 	fde = fdeget_locked(fdp, fd);
2557 	if (fde == NULL) {
2558 		error = EBADF;
2559 		goto out;
2560 	}
2561 
2562 #ifdef CAPABILITIES
2563 	error = cap_check(cap_rights_fde(fde), needrightsp);
2564 	if (error != 0)
2565 		goto out;
2566 #endif
2567 
2568 	if (havecapsp != NULL)
2569 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2570 
2571 	*fpp = fde->fde_file;
2572 
2573 	error = 0;
2574 out:
2575 	return (error);
2576 }
2577 
2578 int
2579 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2580     struct file **fpp, struct filecaps *havecapsp)
2581 {
2582 	struct filedesc *fdp = td->td_proc->p_fd;
2583 	int error;
2584 #ifndef CAPABILITIES
2585 	error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2586 	if (error == 0 && havecapsp != NULL)
2587 		filecaps_fill(havecapsp);
2588 #else
2589 	struct file *fp;
2590 	seq_t seq;
2591 
2592 	for (;;) {
2593 		error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2594 		if (error != 0)
2595 			return (error);
2596 
2597 		if (havecapsp != NULL) {
2598 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2599 			    havecapsp, false)) {
2600 				fdrop(fp, td);
2601 				goto get_locked;
2602 			}
2603 		}
2604 
2605 		if (!fd_modified(fdp, fd, seq))
2606 			break;
2607 		fdrop(fp, td);
2608 	}
2609 
2610 	*fpp = fp;
2611 	return (0);
2612 
2613 get_locked:
2614 	FILEDESC_SLOCK(fdp);
2615 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2616 	if (error == 0)
2617 		fhold(*fpp);
2618 	FILEDESC_SUNLOCK(fdp);
2619 #endif
2620 	return (error);
2621 }
2622 
2623 int
2624 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2625     struct file **fpp, seq_t *seqp)
2626 {
2627 #ifdef CAPABILITIES
2628 	struct filedescent *fde;
2629 #endif
2630 	struct fdescenttbl *fdt;
2631 	struct file *fp;
2632 	u_int count;
2633 #ifdef CAPABILITIES
2634 	seq_t seq;
2635 	cap_rights_t haverights;
2636 	int error;
2637 #endif
2638 
2639 	fdt = fdp->fd_files;
2640 	if ((u_int)fd >= fdt->fdt_nfiles)
2641 		return (EBADF);
2642 	/*
2643 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2644 	 * never raising a refcount above 0.  To accomplish this we have
2645 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2646 	 * must be re-verified once we acquire a reference to be certain
2647 	 * that the identity is still correct and we did not lose a race
2648 	 * due to preemption.
2649 	 */
2650 	for (;;) {
2651 #ifdef CAPABILITIES
2652 		seq = seq_read(fd_seq(fdt, fd));
2653 		fde = &fdt->fdt_ofiles[fd];
2654 		haverights = *cap_rights_fde(fde);
2655 		fp = fde->fde_file;
2656 		if (!seq_consistent(fd_seq(fdt, fd), seq))
2657 			continue;
2658 #else
2659 		fp = fdt->fdt_ofiles[fd].fde_file;
2660 #endif
2661 		if (fp == NULL)
2662 			return (EBADF);
2663 #ifdef CAPABILITIES
2664 		error = cap_check(&haverights, needrightsp);
2665 		if (error != 0)
2666 			return (error);
2667 #endif
2668 		count = fp->f_count;
2669 	retry:
2670 		if (count == 0) {
2671 			/*
2672 			 * Force a reload. Other thread could reallocate the
2673 			 * table before this fd was closed, so it possible that
2674 			 * there is a stale fp pointer in cached version.
2675 			 */
2676 			fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2677 			continue;
2678 		}
2679 		/*
2680 		 * Use an acquire barrier to force re-reading of fdt so it is
2681 		 * refreshed for verification.
2682 		 */
2683 		if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2684 			goto retry;
2685 		fdt = fdp->fd_files;
2686 #ifdef	CAPABILITIES
2687 		if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2688 #else
2689 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2690 #endif
2691 			break;
2692 		fdrop(fp, curthread);
2693 	}
2694 	*fpp = fp;
2695 	if (seqp != NULL) {
2696 #ifdef CAPABILITIES
2697 		*seqp = seq;
2698 #endif
2699 	}
2700 	return (0);
2701 }
2702 
2703 /*
2704  * Extract the file pointer associated with the specified descriptor for the
2705  * current user process.
2706  *
2707  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2708  * returned.
2709  *
2710  * File's rights will be checked against the capability rights mask.
2711  *
2712  * If an error occurred the non-zero error is returned and *fpp is set to
2713  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2714  * responsible for fdrop().
2715  */
2716 static __inline int
2717 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2718     cap_rights_t *needrightsp, seq_t *seqp)
2719 {
2720 	struct filedesc *fdp;
2721 	struct file *fp;
2722 	int error;
2723 
2724 	*fpp = NULL;
2725 	fdp = td->td_proc->p_fd;
2726 	error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2727 	if (error != 0)
2728 		return (error);
2729 	if (fp->f_ops == &badfileops) {
2730 		fdrop(fp, td);
2731 		return (EBADF);
2732 	}
2733 
2734 	/*
2735 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2736 	 */
2737 	error = 0;
2738 	switch (flags) {
2739 	case FREAD:
2740 	case FWRITE:
2741 		if ((fp->f_flag & flags) == 0)
2742 			error = EBADF;
2743 		break;
2744 	case FEXEC:
2745 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2746 		    ((fp->f_flag & FWRITE) != 0))
2747 			error = EBADF;
2748 		break;
2749 	case 0:
2750 		break;
2751 	default:
2752 		KASSERT(0, ("wrong flags"));
2753 	}
2754 
2755 	if (error != 0) {
2756 		fdrop(fp, td);
2757 		return (error);
2758 	}
2759 
2760 	*fpp = fp;
2761 	return (0);
2762 }
2763 
2764 int
2765 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2766 {
2767 
2768 	return (_fget(td, fd, fpp, 0, rightsp, NULL));
2769 }
2770 
2771 int
2772 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2773     struct file **fpp)
2774 {
2775 	int error;
2776 #ifndef CAPABILITIES
2777 	error = _fget(td, fd, fpp, 0, rightsp, NULL);
2778 	if (maxprotp != NULL)
2779 		*maxprotp = VM_PROT_ALL;
2780 #else
2781 	struct filedesc *fdp = td->td_proc->p_fd;
2782 	seq_t seq;
2783 
2784 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2785 	for (;;) {
2786 		error = _fget(td, fd, fpp, 0, rightsp, &seq);
2787 		if (error != 0)
2788 			return (error);
2789 		/*
2790 		 * If requested, convert capability rights to access flags.
2791 		 */
2792 		if (maxprotp != NULL)
2793 			*maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2794 		if (!fd_modified(fdp, fd, seq))
2795 			break;
2796 		fdrop(*fpp, td);
2797 	}
2798 #endif
2799 	return (error);
2800 }
2801 
2802 int
2803 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2804 {
2805 
2806 	return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2807 }
2808 
2809 int
2810 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2811 {
2812 
2813 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2814 }
2815 
2816 int
2817 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2818     struct file **fpp)
2819 {
2820 	struct filedesc *fdp = td->td_proc->p_fd;
2821 #ifndef CAPABILITIES
2822 	return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2823 #else
2824 	int error;
2825 	seq_t seq;
2826 
2827 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2828 	for (;;) {
2829 		error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2830 		if (error != 0)
2831 			return (error);
2832 		error = cap_fcntl_check(fdp, fd, needfcntl);
2833 		if (!fd_modified(fdp, fd, seq))
2834 			break;
2835 		fdrop(*fpp, td);
2836 	}
2837 	if (error != 0) {
2838 		fdrop(*fpp, td);
2839 		*fpp = NULL;
2840 	}
2841 	return (error);
2842 #endif
2843 }
2844 
2845 /*
2846  * Like fget() but loads the underlying vnode, or returns an error if the
2847  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2848  * never have VM objects.  The returned vnode will be vref()'d.
2849  *
2850  * XXX: what about the unused flags ?
2851  */
2852 static __inline int
2853 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2854     struct vnode **vpp)
2855 {
2856 	struct file *fp;
2857 	int error;
2858 
2859 	*vpp = NULL;
2860 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2861 	if (error != 0)
2862 		return (error);
2863 	if (fp->f_vnode == NULL) {
2864 		error = EINVAL;
2865 	} else {
2866 		*vpp = fp->f_vnode;
2867 		vrefact(*vpp);
2868 	}
2869 	fdrop(fp, td);
2870 
2871 	return (error);
2872 }
2873 
2874 int
2875 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2876 {
2877 
2878 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2879 }
2880 
2881 int
2882 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2883     struct filecaps *havecaps, struct vnode **vpp)
2884 {
2885 	struct filedesc *fdp;
2886 	struct filecaps caps;
2887 	struct file *fp;
2888 	int error;
2889 
2890 	fdp = td->td_proc->p_fd;
2891 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2892 	if (error != 0)
2893 		return (error);
2894 	if (fp->f_ops == &badfileops) {
2895 		error = EBADF;
2896 		goto out;
2897 	}
2898 	if (fp->f_vnode == NULL) {
2899 		error = EINVAL;
2900 		goto out;
2901 	}
2902 
2903 	*havecaps = caps;
2904 	*vpp = fp->f_vnode;
2905 	vrefact(*vpp);
2906 
2907 	return (0);
2908 out:
2909 	filecaps_free(&caps);
2910 	return (error);
2911 }
2912 
2913 int
2914 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2915 {
2916 
2917 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2918 }
2919 
2920 int
2921 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2922 {
2923 
2924 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2925 }
2926 
2927 #ifdef notyet
2928 int
2929 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2930     struct vnode **vpp)
2931 {
2932 
2933 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2934 }
2935 #endif
2936 
2937 /*
2938  * Handle the last reference to a file being closed.
2939  */
2940 int
2941 _fdrop(struct file *fp, struct thread *td)
2942 {
2943 	int error;
2944 
2945 	if (fp->f_count != 0)
2946 		panic("fdrop: count %d", fp->f_count);
2947 	error = fo_close(fp, td);
2948 	atomic_subtract_int(&openfiles, 1);
2949 	crfree(fp->f_cred);
2950 	free(fp->f_advice, M_FADVISE);
2951 	uma_zfree(file_zone, fp);
2952 
2953 	return (error);
2954 }
2955 
2956 /*
2957  * Apply an advisory lock on a file descriptor.
2958  *
2959  * Just attempt to get a record lock of the requested type on the entire file
2960  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2961  */
2962 #ifndef _SYS_SYSPROTO_H_
2963 struct flock_args {
2964 	int	fd;
2965 	int	how;
2966 };
2967 #endif
2968 /* ARGSUSED */
2969 int
2970 sys_flock(struct thread *td, struct flock_args *uap)
2971 {
2972 	struct file *fp;
2973 	struct vnode *vp;
2974 	struct flock lf;
2975 	int error;
2976 
2977 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
2978 	if (error != 0)
2979 		return (error);
2980 	if (fp->f_type != DTYPE_VNODE) {
2981 		fdrop(fp, td);
2982 		return (EOPNOTSUPP);
2983 	}
2984 
2985 	vp = fp->f_vnode;
2986 	lf.l_whence = SEEK_SET;
2987 	lf.l_start = 0;
2988 	lf.l_len = 0;
2989 	if (uap->how & LOCK_UN) {
2990 		lf.l_type = F_UNLCK;
2991 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2992 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2993 		goto done2;
2994 	}
2995 	if (uap->how & LOCK_EX)
2996 		lf.l_type = F_WRLCK;
2997 	else if (uap->how & LOCK_SH)
2998 		lf.l_type = F_RDLCK;
2999 	else {
3000 		error = EBADF;
3001 		goto done2;
3002 	}
3003 	atomic_set_int(&fp->f_flag, FHASLOCK);
3004 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3005 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3006 done2:
3007 	fdrop(fp, td);
3008 	return (error);
3009 }
3010 /*
3011  * Duplicate the specified descriptor to a free descriptor.
3012  */
3013 int
3014 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3015     int openerror, int *indxp)
3016 {
3017 	struct filedescent *newfde, *oldfde;
3018 	struct file *fp;
3019 	u_long *ioctls;
3020 	int error, indx;
3021 
3022 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3023 	    ("unexpected error %d in %s", openerror, __func__));
3024 
3025 	/*
3026 	 * If the to-be-dup'd fd number is greater than the allowed number
3027 	 * of file descriptors, or the fd to be dup'd has already been
3028 	 * closed, then reject.
3029 	 */
3030 	FILEDESC_XLOCK(fdp);
3031 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3032 		FILEDESC_XUNLOCK(fdp);
3033 		return (EBADF);
3034 	}
3035 
3036 	error = fdalloc(td, 0, &indx);
3037 	if (error != 0) {
3038 		FILEDESC_XUNLOCK(fdp);
3039 		return (error);
3040 	}
3041 
3042 	/*
3043 	 * There are two cases of interest here.
3044 	 *
3045 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3046 	 *
3047 	 * For ENXIO steal away the file structure from (dfd) and store it in
3048 	 * (indx).  (dfd) is effectively closed by this operation.
3049 	 */
3050 	switch (openerror) {
3051 	case ENODEV:
3052 		/*
3053 		 * Check that the mode the file is being opened for is a
3054 		 * subset of the mode of the existing descriptor.
3055 		 */
3056 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3057 			fdunused(fdp, indx);
3058 			FILEDESC_XUNLOCK(fdp);
3059 			return (EACCES);
3060 		}
3061 		fhold(fp);
3062 		newfde = &fdp->fd_ofiles[indx];
3063 		oldfde = &fdp->fd_ofiles[dfd];
3064 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3065 #ifdef CAPABILITIES
3066 		seq_write_begin(&newfde->fde_seq);
3067 #endif
3068 		memcpy(newfde, oldfde, fde_change_size);
3069 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3070 		    ioctls);
3071 #ifdef CAPABILITIES
3072 		seq_write_end(&newfde->fde_seq);
3073 #endif
3074 		break;
3075 	case ENXIO:
3076 		/*
3077 		 * Steal away the file pointer from dfd and stuff it into indx.
3078 		 */
3079 		newfde = &fdp->fd_ofiles[indx];
3080 		oldfde = &fdp->fd_ofiles[dfd];
3081 #ifdef CAPABILITIES
3082 		seq_write_begin(&newfde->fde_seq);
3083 #endif
3084 		memcpy(newfde, oldfde, fde_change_size);
3085 		oldfde->fde_file = NULL;
3086 		fdunused(fdp, dfd);
3087 #ifdef CAPABILITIES
3088 		seq_write_end(&newfde->fde_seq);
3089 #endif
3090 		break;
3091 	}
3092 	FILEDESC_XUNLOCK(fdp);
3093 	*indxp = indx;
3094 	return (0);
3095 }
3096 
3097 /*
3098  * This sysctl determines if we will allow a process to chroot(2) if it
3099  * has a directory open:
3100  *	0: disallowed for all processes.
3101  *	1: allowed for processes that were not already chroot(2)'ed.
3102  *	2: allowed for all processes.
3103  */
3104 
3105 static int chroot_allow_open_directories = 1;
3106 
3107 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3108     &chroot_allow_open_directories, 0,
3109     "Allow a process to chroot(2) if it has a directory open");
3110 
3111 /*
3112  * Helper function for raised chroot(2) security function:  Refuse if
3113  * any filedescriptors are open directories.
3114  */
3115 static int
3116 chroot_refuse_vdir_fds(struct filedesc *fdp)
3117 {
3118 	struct vnode *vp;
3119 	struct file *fp;
3120 	int fd;
3121 
3122 	FILEDESC_LOCK_ASSERT(fdp);
3123 
3124 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3125 		fp = fget_locked(fdp, fd);
3126 		if (fp == NULL)
3127 			continue;
3128 		if (fp->f_type == DTYPE_VNODE) {
3129 			vp = fp->f_vnode;
3130 			if (vp->v_type == VDIR)
3131 				return (EPERM);
3132 		}
3133 	}
3134 	return (0);
3135 }
3136 
3137 /*
3138  * Common routine for kern_chroot() and jail_attach().  The caller is
3139  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3140  * authorize this operation.
3141  */
3142 int
3143 pwd_chroot(struct thread *td, struct vnode *vp)
3144 {
3145 	struct filedesc *fdp;
3146 	struct vnode *oldvp;
3147 	int error;
3148 
3149 	fdp = td->td_proc->p_fd;
3150 	FILEDESC_XLOCK(fdp);
3151 	if (chroot_allow_open_directories == 0 ||
3152 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3153 		error = chroot_refuse_vdir_fds(fdp);
3154 		if (error != 0) {
3155 			FILEDESC_XUNLOCK(fdp);
3156 			return (error);
3157 		}
3158 	}
3159 	oldvp = fdp->fd_rdir;
3160 	vrefact(vp);
3161 	fdp->fd_rdir = vp;
3162 	if (fdp->fd_jdir == NULL) {
3163 		vrefact(vp);
3164 		fdp->fd_jdir = vp;
3165 	}
3166 	FILEDESC_XUNLOCK(fdp);
3167 	vrele(oldvp);
3168 	return (0);
3169 }
3170 
3171 void
3172 pwd_chdir(struct thread *td, struct vnode *vp)
3173 {
3174 	struct filedesc *fdp;
3175 	struct vnode *oldvp;
3176 
3177 	fdp = td->td_proc->p_fd;
3178 	FILEDESC_XLOCK(fdp);
3179 	VNASSERT(vp->v_usecount > 0, vp,
3180 	    ("chdir to a vnode with zero usecount"));
3181 	oldvp = fdp->fd_cdir;
3182 	fdp->fd_cdir = vp;
3183 	FILEDESC_XUNLOCK(fdp);
3184 	vrele(oldvp);
3185 }
3186 
3187 /*
3188  * Scan all active processes and prisons to see if any of them have a current
3189  * or root directory of `olddp'. If so, replace them with the new mount point.
3190  */
3191 void
3192 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3193 {
3194 	struct filedesc *fdp;
3195 	struct prison *pr;
3196 	struct proc *p;
3197 	int nrele;
3198 
3199 	if (vrefcnt(olddp) == 1)
3200 		return;
3201 	nrele = 0;
3202 	sx_slock(&allproc_lock);
3203 	FOREACH_PROC_IN_SYSTEM(p) {
3204 		PROC_LOCK(p);
3205 		fdp = fdhold(p);
3206 		PROC_UNLOCK(p);
3207 		if (fdp == NULL)
3208 			continue;
3209 		FILEDESC_XLOCK(fdp);
3210 		if (fdp->fd_cdir == olddp) {
3211 			vrefact(newdp);
3212 			fdp->fd_cdir = newdp;
3213 			nrele++;
3214 		}
3215 		if (fdp->fd_rdir == olddp) {
3216 			vrefact(newdp);
3217 			fdp->fd_rdir = newdp;
3218 			nrele++;
3219 		}
3220 		if (fdp->fd_jdir == olddp) {
3221 			vrefact(newdp);
3222 			fdp->fd_jdir = newdp;
3223 			nrele++;
3224 		}
3225 		FILEDESC_XUNLOCK(fdp);
3226 		fddrop(fdp);
3227 	}
3228 	sx_sunlock(&allproc_lock);
3229 	if (rootvnode == olddp) {
3230 		vrefact(newdp);
3231 		rootvnode = newdp;
3232 		nrele++;
3233 	}
3234 	mtx_lock(&prison0.pr_mtx);
3235 	if (prison0.pr_root == olddp) {
3236 		vrefact(newdp);
3237 		prison0.pr_root = newdp;
3238 		nrele++;
3239 	}
3240 	mtx_unlock(&prison0.pr_mtx);
3241 	sx_slock(&allprison_lock);
3242 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3243 		mtx_lock(&pr->pr_mtx);
3244 		if (pr->pr_root == olddp) {
3245 			vrefact(newdp);
3246 			pr->pr_root = newdp;
3247 			nrele++;
3248 		}
3249 		mtx_unlock(&pr->pr_mtx);
3250 	}
3251 	sx_sunlock(&allprison_lock);
3252 	while (nrele--)
3253 		vrele(olddp);
3254 }
3255 
3256 struct filedesc_to_leader *
3257 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3258 {
3259 	struct filedesc_to_leader *fdtol;
3260 
3261 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3262 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3263 	fdtol->fdl_refcount = 1;
3264 	fdtol->fdl_holdcount = 0;
3265 	fdtol->fdl_wakeup = 0;
3266 	fdtol->fdl_leader = leader;
3267 	if (old != NULL) {
3268 		FILEDESC_XLOCK(fdp);
3269 		fdtol->fdl_next = old->fdl_next;
3270 		fdtol->fdl_prev = old;
3271 		old->fdl_next = fdtol;
3272 		fdtol->fdl_next->fdl_prev = fdtol;
3273 		FILEDESC_XUNLOCK(fdp);
3274 	} else {
3275 		fdtol->fdl_next = fdtol;
3276 		fdtol->fdl_prev = fdtol;
3277 	}
3278 	return (fdtol);
3279 }
3280 
3281 static int
3282 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3283 {
3284 	struct filedesc *fdp;
3285 	int i, count, slots;
3286 
3287 	if (*(int *)arg1 != 0)
3288 		return (EINVAL);
3289 
3290 	fdp = curproc->p_fd;
3291 	count = 0;
3292 	FILEDESC_SLOCK(fdp);
3293 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3294 	for (i = 0; i < slots; i++)
3295 		count += bitcountl(fdp->fd_map[i]);
3296 	FILEDESC_SUNLOCK(fdp);
3297 
3298 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3299 }
3300 
3301 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3302     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3303     "Number of open file descriptors");
3304 
3305 /*
3306  * Get file structures globally.
3307  */
3308 static int
3309 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3310 {
3311 	struct xfile xf;
3312 	struct filedesc *fdp;
3313 	struct file *fp;
3314 	struct proc *p;
3315 	int error, n;
3316 
3317 	error = sysctl_wire_old_buffer(req, 0);
3318 	if (error != 0)
3319 		return (error);
3320 	if (req->oldptr == NULL) {
3321 		n = 0;
3322 		sx_slock(&allproc_lock);
3323 		FOREACH_PROC_IN_SYSTEM(p) {
3324 			PROC_LOCK(p);
3325 			if (p->p_state == PRS_NEW) {
3326 				PROC_UNLOCK(p);
3327 				continue;
3328 			}
3329 			fdp = fdhold(p);
3330 			PROC_UNLOCK(p);
3331 			if (fdp == NULL)
3332 				continue;
3333 			/* overestimates sparse tables. */
3334 			if (fdp->fd_lastfile > 0)
3335 				n += fdp->fd_lastfile;
3336 			fddrop(fdp);
3337 		}
3338 		sx_sunlock(&allproc_lock);
3339 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3340 	}
3341 	error = 0;
3342 	bzero(&xf, sizeof(xf));
3343 	xf.xf_size = sizeof(xf);
3344 	sx_slock(&allproc_lock);
3345 	FOREACH_PROC_IN_SYSTEM(p) {
3346 		PROC_LOCK(p);
3347 		if (p->p_state == PRS_NEW) {
3348 			PROC_UNLOCK(p);
3349 			continue;
3350 		}
3351 		if (p_cansee(req->td, p) != 0) {
3352 			PROC_UNLOCK(p);
3353 			continue;
3354 		}
3355 		xf.xf_pid = p->p_pid;
3356 		xf.xf_uid = p->p_ucred->cr_uid;
3357 		fdp = fdhold(p);
3358 		PROC_UNLOCK(p);
3359 		if (fdp == NULL)
3360 			continue;
3361 		FILEDESC_SLOCK(fdp);
3362 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3363 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3364 				continue;
3365 			xf.xf_fd = n;
3366 			xf.xf_file = fp;
3367 			xf.xf_data = fp->f_data;
3368 			xf.xf_vnode = fp->f_vnode;
3369 			xf.xf_type = fp->f_type;
3370 			xf.xf_count = fp->f_count;
3371 			xf.xf_msgcount = 0;
3372 			xf.xf_offset = foffset_get(fp);
3373 			xf.xf_flag = fp->f_flag;
3374 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3375 			if (error)
3376 				break;
3377 		}
3378 		FILEDESC_SUNLOCK(fdp);
3379 		fddrop(fdp);
3380 		if (error)
3381 			break;
3382 	}
3383 	sx_sunlock(&allproc_lock);
3384 	return (error);
3385 }
3386 
3387 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3388     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3389 
3390 #ifdef KINFO_FILE_SIZE
3391 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3392 #endif
3393 
3394 static int
3395 xlate_fflags(int fflags)
3396 {
3397 	static const struct {
3398 		int	fflag;
3399 		int	kf_fflag;
3400 	} fflags_table[] = {
3401 		{ FAPPEND, KF_FLAG_APPEND },
3402 		{ FASYNC, KF_FLAG_ASYNC },
3403 		{ FFSYNC, KF_FLAG_FSYNC },
3404 		{ FHASLOCK, KF_FLAG_HASLOCK },
3405 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3406 		{ FREAD, KF_FLAG_READ },
3407 		{ FWRITE, KF_FLAG_WRITE },
3408 		{ O_CREAT, KF_FLAG_CREAT },
3409 		{ O_DIRECT, KF_FLAG_DIRECT },
3410 		{ O_EXCL, KF_FLAG_EXCL },
3411 		{ O_EXEC, KF_FLAG_EXEC },
3412 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3413 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3414 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3415 		{ O_TRUNC, KF_FLAG_TRUNC }
3416 	};
3417 	unsigned int i;
3418 	int kflags;
3419 
3420 	kflags = 0;
3421 	for (i = 0; i < nitems(fflags_table); i++)
3422 		if (fflags & fflags_table[i].fflag)
3423 			kflags |=  fflags_table[i].kf_fflag;
3424 	return (kflags);
3425 }
3426 
3427 /* Trim unused data from kf_path by truncating the structure size. */
3428 static void
3429 pack_kinfo(struct kinfo_file *kif)
3430 {
3431 
3432 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3433 	    strlen(kif->kf_path) + 1;
3434 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3435 }
3436 
3437 static void
3438 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3439     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3440 {
3441 	int error;
3442 
3443 	bzero(kif, sizeof(*kif));
3444 
3445 	/* Set a default type to allow for empty fill_kinfo() methods. */
3446 	kif->kf_type = KF_TYPE_UNKNOWN;
3447 	kif->kf_flags = xlate_fflags(fp->f_flag);
3448 	if (rightsp != NULL)
3449 		kif->kf_cap_rights = *rightsp;
3450 	else
3451 		cap_rights_init(&kif->kf_cap_rights);
3452 	kif->kf_fd = fd;
3453 	kif->kf_ref_count = fp->f_count;
3454 	kif->kf_offset = foffset_get(fp);
3455 
3456 	/*
3457 	 * This may drop the filedesc lock, so the 'fp' cannot be
3458 	 * accessed after this call.
3459 	 */
3460 	error = fo_fill_kinfo(fp, kif, fdp);
3461 	if (error == 0)
3462 		kif->kf_status |= KF_ATTR_VALID;
3463 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3464 		pack_kinfo(kif);
3465 	else
3466 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3467 }
3468 
3469 static void
3470 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3471     struct kinfo_file *kif, int flags)
3472 {
3473 	int error;
3474 
3475 	bzero(kif, sizeof(*kif));
3476 
3477 	kif->kf_type = KF_TYPE_VNODE;
3478 	error = vn_fill_kinfo_vnode(vp, kif);
3479 	if (error == 0)
3480 		kif->kf_status |= KF_ATTR_VALID;
3481 	kif->kf_flags = xlate_fflags(fflags);
3482 	cap_rights_init(&kif->kf_cap_rights);
3483 	kif->kf_fd = fd;
3484 	kif->kf_ref_count = -1;
3485 	kif->kf_offset = -1;
3486 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3487 		pack_kinfo(kif);
3488 	else
3489 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3490 	vrele(vp);
3491 }
3492 
3493 struct export_fd_buf {
3494 	struct filedesc		*fdp;
3495 	struct sbuf 		*sb;
3496 	ssize_t			remainder;
3497 	struct kinfo_file	kif;
3498 	int			flags;
3499 };
3500 
3501 static int
3502 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3503 {
3504 	struct kinfo_file *kif;
3505 
3506 	kif = &efbuf->kif;
3507 	if (efbuf->remainder != -1) {
3508 		if (efbuf->remainder < kif->kf_structsize) {
3509 			/* Terminate export. */
3510 			efbuf->remainder = 0;
3511 			return (0);
3512 		}
3513 		efbuf->remainder -= kif->kf_structsize;
3514 	}
3515 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3516 }
3517 
3518 static int
3519 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3520     struct export_fd_buf *efbuf)
3521 {
3522 	int error;
3523 
3524 	if (efbuf->remainder == 0)
3525 		return (0);
3526 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3527 	    efbuf->flags);
3528 	FILEDESC_SUNLOCK(efbuf->fdp);
3529 	error = export_kinfo_to_sb(efbuf);
3530 	FILEDESC_SLOCK(efbuf->fdp);
3531 	return (error);
3532 }
3533 
3534 static int
3535 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3536     struct export_fd_buf *efbuf)
3537 {
3538 	int error;
3539 
3540 	if (efbuf->remainder == 0)
3541 		return (0);
3542 	if (efbuf->fdp != NULL)
3543 		FILEDESC_SUNLOCK(efbuf->fdp);
3544 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3545 	error = export_kinfo_to_sb(efbuf);
3546 	if (efbuf->fdp != NULL)
3547 		FILEDESC_SLOCK(efbuf->fdp);
3548 	return (error);
3549 }
3550 
3551 /*
3552  * Store a process file descriptor information to sbuf.
3553  *
3554  * Takes a locked proc as argument, and returns with the proc unlocked.
3555  */
3556 int
3557 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3558     int flags)
3559 {
3560 	struct file *fp;
3561 	struct filedesc *fdp;
3562 	struct export_fd_buf *efbuf;
3563 	struct vnode *cttyvp, *textvp, *tracevp;
3564 	int error, i;
3565 	cap_rights_t rights;
3566 
3567 	PROC_LOCK_ASSERT(p, MA_OWNED);
3568 
3569 	/* ktrace vnode */
3570 	tracevp = p->p_tracevp;
3571 	if (tracevp != NULL)
3572 		vrefact(tracevp);
3573 	/* text vnode */
3574 	textvp = p->p_textvp;
3575 	if (textvp != NULL)
3576 		vrefact(textvp);
3577 	/* Controlling tty. */
3578 	cttyvp = NULL;
3579 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3580 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3581 		if (cttyvp != NULL)
3582 			vrefact(cttyvp);
3583 	}
3584 	fdp = fdhold(p);
3585 	PROC_UNLOCK(p);
3586 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3587 	efbuf->fdp = NULL;
3588 	efbuf->sb = sb;
3589 	efbuf->remainder = maxlen;
3590 	efbuf->flags = flags;
3591 	if (tracevp != NULL)
3592 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3593 		    efbuf);
3594 	if (textvp != NULL)
3595 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3596 	if (cttyvp != NULL)
3597 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3598 		    efbuf);
3599 	error = 0;
3600 	if (fdp == NULL)
3601 		goto fail;
3602 	efbuf->fdp = fdp;
3603 	FILEDESC_SLOCK(fdp);
3604 	/* working directory */
3605 	if (fdp->fd_cdir != NULL) {
3606 		vrefact(fdp->fd_cdir);
3607 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3608 	}
3609 	/* root directory */
3610 	if (fdp->fd_rdir != NULL) {
3611 		vrefact(fdp->fd_rdir);
3612 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3613 	}
3614 	/* jail directory */
3615 	if (fdp->fd_jdir != NULL) {
3616 		vrefact(fdp->fd_jdir);
3617 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3618 	}
3619 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3620 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3621 			continue;
3622 #ifdef CAPABILITIES
3623 		rights = *cap_rights(fdp, i);
3624 #else /* !CAPABILITIES */
3625 		rights = cap_no_rights;
3626 #endif
3627 		/*
3628 		 * Create sysctl entry.  It is OK to drop the filedesc
3629 		 * lock inside of export_file_to_sb() as we will
3630 		 * re-validate and re-evaluate its properties when the
3631 		 * loop continues.
3632 		 */
3633 		error = export_file_to_sb(fp, i, &rights, efbuf);
3634 		if (error != 0 || efbuf->remainder == 0)
3635 			break;
3636 	}
3637 	FILEDESC_SUNLOCK(fdp);
3638 	fddrop(fdp);
3639 fail:
3640 	free(efbuf, M_TEMP);
3641 	return (error);
3642 }
3643 
3644 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3645 
3646 /*
3647  * Get per-process file descriptors for use by procstat(1), et al.
3648  */
3649 static int
3650 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3651 {
3652 	struct sbuf sb;
3653 	struct proc *p;
3654 	ssize_t maxlen;
3655 	int error, error2, *name;
3656 
3657 	name = (int *)arg1;
3658 
3659 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3660 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3661 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3662 	if (error != 0) {
3663 		sbuf_delete(&sb);
3664 		return (error);
3665 	}
3666 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3667 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3668 	    KERN_FILEDESC_PACK_KINFO);
3669 	error2 = sbuf_finish(&sb);
3670 	sbuf_delete(&sb);
3671 	return (error != 0 ? error : error2);
3672 }
3673 
3674 #ifdef COMPAT_FREEBSD7
3675 #ifdef KINFO_OFILE_SIZE
3676 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3677 #endif
3678 
3679 static void
3680 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3681 {
3682 
3683 	okif->kf_structsize = sizeof(*okif);
3684 	okif->kf_type = kif->kf_type;
3685 	okif->kf_fd = kif->kf_fd;
3686 	okif->kf_ref_count = kif->kf_ref_count;
3687 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3688 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3689 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3690 	okif->kf_offset = kif->kf_offset;
3691 	if (kif->kf_type == KF_TYPE_VNODE)
3692 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3693 	else
3694 		okif->kf_vnode_type = KF_VTYPE_VNON;
3695 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3696 	if (kif->kf_type == KF_TYPE_SOCKET) {
3697 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3698 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3699 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3700 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3701 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3702 	} else {
3703 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3704 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3705 	}
3706 }
3707 
3708 static int
3709 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3710     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3711 {
3712 	int error;
3713 
3714 	vrefact(vp);
3715 	FILEDESC_SUNLOCK(fdp);
3716 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3717 	kinfo_to_okinfo(kif, okif);
3718 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3719 	FILEDESC_SLOCK(fdp);
3720 	return (error);
3721 }
3722 
3723 /*
3724  * Get per-process file descriptors for use by procstat(1), et al.
3725  */
3726 static int
3727 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3728 {
3729 	struct kinfo_ofile *okif;
3730 	struct kinfo_file *kif;
3731 	struct filedesc *fdp;
3732 	int error, i, *name;
3733 	struct file *fp;
3734 	struct proc *p;
3735 
3736 	name = (int *)arg1;
3737 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3738 	if (error != 0)
3739 		return (error);
3740 	fdp = fdhold(p);
3741 	PROC_UNLOCK(p);
3742 	if (fdp == NULL)
3743 		return (ENOENT);
3744 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3745 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3746 	FILEDESC_SLOCK(fdp);
3747 	if (fdp->fd_cdir != NULL)
3748 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3749 		    okif, fdp, req);
3750 	if (fdp->fd_rdir != NULL)
3751 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3752 		    okif, fdp, req);
3753 	if (fdp->fd_jdir != NULL)
3754 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3755 		    okif, fdp, req);
3756 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3757 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3758 			continue;
3759 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3760 		    KERN_FILEDESC_PACK_KINFO);
3761 		FILEDESC_SUNLOCK(fdp);
3762 		kinfo_to_okinfo(kif, okif);
3763 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3764 		FILEDESC_SLOCK(fdp);
3765 		if (error)
3766 			break;
3767 	}
3768 	FILEDESC_SUNLOCK(fdp);
3769 	fddrop(fdp);
3770 	free(kif, M_TEMP);
3771 	free(okif, M_TEMP);
3772 	return (0);
3773 }
3774 
3775 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3776     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3777     "Process ofiledesc entries");
3778 #endif	/* COMPAT_FREEBSD7 */
3779 
3780 int
3781 vntype_to_kinfo(int vtype)
3782 {
3783 	struct {
3784 		int	vtype;
3785 		int	kf_vtype;
3786 	} vtypes_table[] = {
3787 		{ VBAD, KF_VTYPE_VBAD },
3788 		{ VBLK, KF_VTYPE_VBLK },
3789 		{ VCHR, KF_VTYPE_VCHR },
3790 		{ VDIR, KF_VTYPE_VDIR },
3791 		{ VFIFO, KF_VTYPE_VFIFO },
3792 		{ VLNK, KF_VTYPE_VLNK },
3793 		{ VNON, KF_VTYPE_VNON },
3794 		{ VREG, KF_VTYPE_VREG },
3795 		{ VSOCK, KF_VTYPE_VSOCK }
3796 	};
3797 	unsigned int i;
3798 
3799 	/*
3800 	 * Perform vtype translation.
3801 	 */
3802 	for (i = 0; i < nitems(vtypes_table); i++)
3803 		if (vtypes_table[i].vtype == vtype)
3804 			return (vtypes_table[i].kf_vtype);
3805 
3806 	return (KF_VTYPE_UNKNOWN);
3807 }
3808 
3809 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3810     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3811     "Process filedesc entries");
3812 
3813 /*
3814  * Store a process current working directory information to sbuf.
3815  *
3816  * Takes a locked proc as argument, and returns with the proc unlocked.
3817  */
3818 int
3819 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3820 {
3821 	struct filedesc *fdp;
3822 	struct export_fd_buf *efbuf;
3823 	int error;
3824 
3825 	PROC_LOCK_ASSERT(p, MA_OWNED);
3826 
3827 	fdp = fdhold(p);
3828 	PROC_UNLOCK(p);
3829 	if (fdp == NULL)
3830 		return (EINVAL);
3831 
3832 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3833 	efbuf->fdp = fdp;
3834 	efbuf->sb = sb;
3835 	efbuf->remainder = maxlen;
3836 
3837 	FILEDESC_SLOCK(fdp);
3838 	if (fdp->fd_cdir == NULL)
3839 		error = EINVAL;
3840 	else {
3841 		vrefact(fdp->fd_cdir);
3842 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3843 		    FREAD, efbuf);
3844 	}
3845 	FILEDESC_SUNLOCK(fdp);
3846 	fddrop(fdp);
3847 	free(efbuf, M_TEMP);
3848 	return (error);
3849 }
3850 
3851 /*
3852  * Get per-process current working directory.
3853  */
3854 static int
3855 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3856 {
3857 	struct sbuf sb;
3858 	struct proc *p;
3859 	ssize_t maxlen;
3860 	int error, error2, *name;
3861 
3862 	name = (int *)arg1;
3863 
3864 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3865 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3866 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3867 	if (error != 0) {
3868 		sbuf_delete(&sb);
3869 		return (error);
3870 	}
3871 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3872 	error = kern_proc_cwd_out(p, &sb, maxlen);
3873 	error2 = sbuf_finish(&sb);
3874 	sbuf_delete(&sb);
3875 	return (error != 0 ? error : error2);
3876 }
3877 
3878 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3879     sysctl_kern_proc_cwd, "Process current working directory");
3880 
3881 #ifdef DDB
3882 /*
3883  * For the purposes of debugging, generate a human-readable string for the
3884  * file type.
3885  */
3886 static const char *
3887 file_type_to_name(short type)
3888 {
3889 
3890 	switch (type) {
3891 	case 0:
3892 		return ("zero");
3893 	case DTYPE_VNODE:
3894 		return ("vnode");
3895 	case DTYPE_SOCKET:
3896 		return ("socket");
3897 	case DTYPE_PIPE:
3898 		return ("pipe");
3899 	case DTYPE_FIFO:
3900 		return ("fifo");
3901 	case DTYPE_KQUEUE:
3902 		return ("kqueue");
3903 	case DTYPE_CRYPTO:
3904 		return ("crypto");
3905 	case DTYPE_MQUEUE:
3906 		return ("mqueue");
3907 	case DTYPE_SHM:
3908 		return ("shm");
3909 	case DTYPE_SEM:
3910 		return ("ksem");
3911 	case DTYPE_PTS:
3912 		return ("pts");
3913 	case DTYPE_DEV:
3914 		return ("dev");
3915 	case DTYPE_PROCDESC:
3916 		return ("proc");
3917 	case DTYPE_LINUXEFD:
3918 		return ("levent");
3919 	case DTYPE_LINUXTFD:
3920 		return ("ltimer");
3921 	default:
3922 		return ("unkn");
3923 	}
3924 }
3925 
3926 /*
3927  * For the purposes of debugging, identify a process (if any, perhaps one of
3928  * many) that references the passed file in its file descriptor array. Return
3929  * NULL if none.
3930  */
3931 static struct proc *
3932 file_to_first_proc(struct file *fp)
3933 {
3934 	struct filedesc *fdp;
3935 	struct proc *p;
3936 	int n;
3937 
3938 	FOREACH_PROC_IN_SYSTEM(p) {
3939 		if (p->p_state == PRS_NEW)
3940 			continue;
3941 		fdp = p->p_fd;
3942 		if (fdp == NULL)
3943 			continue;
3944 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3945 			if (fp == fdp->fd_ofiles[n].fde_file)
3946 				return (p);
3947 		}
3948 	}
3949 	return (NULL);
3950 }
3951 
3952 static void
3953 db_print_file(struct file *fp, int header)
3954 {
3955 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3956 	struct proc *p;
3957 
3958 	if (header)
3959 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3960 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3961 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3962 		    "FCmd");
3963 	p = file_to_first_proc(fp);
3964 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3965 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3966 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3967 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3968 
3969 #undef XPTRWIDTH
3970 }
3971 
3972 DB_SHOW_COMMAND(file, db_show_file)
3973 {
3974 	struct file *fp;
3975 
3976 	if (!have_addr) {
3977 		db_printf("usage: show file <addr>\n");
3978 		return;
3979 	}
3980 	fp = (struct file *)addr;
3981 	db_print_file(fp, 1);
3982 }
3983 
3984 DB_SHOW_COMMAND(files, db_show_files)
3985 {
3986 	struct filedesc *fdp;
3987 	struct file *fp;
3988 	struct proc *p;
3989 	int header;
3990 	int n;
3991 
3992 	header = 1;
3993 	FOREACH_PROC_IN_SYSTEM(p) {
3994 		if (p->p_state == PRS_NEW)
3995 			continue;
3996 		if ((fdp = p->p_fd) == NULL)
3997 			continue;
3998 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
3999 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4000 				continue;
4001 			db_print_file(fp, header);
4002 			header = 0;
4003 		}
4004 	}
4005 }
4006 #endif
4007 
4008 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4009     &maxfilesperproc, 0, "Maximum files allowed open per process");
4010 
4011 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4012     &maxfiles, 0, "Maximum number of files");
4013 
4014 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4015     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4016 
4017 /* ARGSUSED*/
4018 static void
4019 filelistinit(void *dummy)
4020 {
4021 
4022 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4023 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4024 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4025 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4026 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4027 }
4028 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4029 
4030 /*-------------------------------------------------------------------*/
4031 
4032 static int
4033 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4034     int flags, struct thread *td)
4035 {
4036 
4037 	return (EBADF);
4038 }
4039 
4040 static int
4041 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4042     struct thread *td)
4043 {
4044 
4045 	return (EINVAL);
4046 }
4047 
4048 static int
4049 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4050     struct thread *td)
4051 {
4052 
4053 	return (EBADF);
4054 }
4055 
4056 static int
4057 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4058     struct thread *td)
4059 {
4060 
4061 	return (0);
4062 }
4063 
4064 static int
4065 badfo_kqfilter(struct file *fp, struct knote *kn)
4066 {
4067 
4068 	return (EBADF);
4069 }
4070 
4071 static int
4072 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4073     struct thread *td)
4074 {
4075 
4076 	return (EBADF);
4077 }
4078 
4079 static int
4080 badfo_close(struct file *fp, struct thread *td)
4081 {
4082 
4083 	return (0);
4084 }
4085 
4086 static int
4087 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4088     struct thread *td)
4089 {
4090 
4091 	return (EBADF);
4092 }
4093 
4094 static int
4095 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4096     struct thread *td)
4097 {
4098 
4099 	return (EBADF);
4100 }
4101 
4102 static int
4103 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4104     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4105     struct thread *td)
4106 {
4107 
4108 	return (EBADF);
4109 }
4110 
4111 static int
4112 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4113 {
4114 
4115 	return (0);
4116 }
4117 
4118 struct fileops badfileops = {
4119 	.fo_read = badfo_readwrite,
4120 	.fo_write = badfo_readwrite,
4121 	.fo_truncate = badfo_truncate,
4122 	.fo_ioctl = badfo_ioctl,
4123 	.fo_poll = badfo_poll,
4124 	.fo_kqfilter = badfo_kqfilter,
4125 	.fo_stat = badfo_stat,
4126 	.fo_close = badfo_close,
4127 	.fo_chmod = badfo_chmod,
4128 	.fo_chown = badfo_chown,
4129 	.fo_sendfile = badfo_sendfile,
4130 	.fo_fill_kinfo = badfo_fill_kinfo,
4131 };
4132 
4133 int
4134 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4135     int flags, struct thread *td)
4136 {
4137 
4138 	return (EOPNOTSUPP);
4139 }
4140 
4141 int
4142 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4143     struct thread *td)
4144 {
4145 
4146 	return (EINVAL);
4147 }
4148 
4149 int
4150 invfo_ioctl(struct file *fp, u_long com, void *data,
4151     struct ucred *active_cred, struct thread *td)
4152 {
4153 
4154 	return (ENOTTY);
4155 }
4156 
4157 int
4158 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4159     struct thread *td)
4160 {
4161 
4162 	return (poll_no_poll(events));
4163 }
4164 
4165 int
4166 invfo_kqfilter(struct file *fp, struct knote *kn)
4167 {
4168 
4169 	return (EINVAL);
4170 }
4171 
4172 int
4173 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4174     struct thread *td)
4175 {
4176 
4177 	return (EINVAL);
4178 }
4179 
4180 int
4181 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4182     struct thread *td)
4183 {
4184 
4185 	return (EINVAL);
4186 }
4187 
4188 int
4189 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4190     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4191     struct thread *td)
4192 {
4193 
4194 	return (EINVAL);
4195 }
4196 
4197 /*-------------------------------------------------------------------*/
4198 
4199 /*
4200  * File Descriptor pseudo-device driver (/dev/fd/).
4201  *
4202  * Opening minor device N dup()s the file (if any) connected to file
4203  * descriptor N belonging to the calling process.  Note that this driver
4204  * consists of only the ``open()'' routine, because all subsequent
4205  * references to this file will be direct to the other driver.
4206  *
4207  * XXX: we could give this one a cloning event handler if necessary.
4208  */
4209 
4210 /* ARGSUSED */
4211 static int
4212 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4213 {
4214 
4215 	/*
4216 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4217 	 * the file descriptor being sought for duplication. The error
4218 	 * return ensures that the vnode for this device will be released
4219 	 * by vn_open. Open will detect this special error and take the
4220 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4221 	 * will simply report the error.
4222 	 */
4223 	td->td_dupfd = dev2unit(dev);
4224 	return (ENODEV);
4225 }
4226 
4227 static struct cdevsw fildesc_cdevsw = {
4228 	.d_version =	D_VERSION,
4229 	.d_open =	fdopen,
4230 	.d_name =	"FD",
4231 };
4232 
4233 static void
4234 fildesc_drvinit(void *unused)
4235 {
4236 	struct cdev *dev;
4237 
4238 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4239 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4240 	make_dev_alias(dev, "stdin");
4241 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4242 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4243 	make_dev_alias(dev, "stdout");
4244 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4245 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4246 	make_dev_alias(dev, "stderr");
4247 }
4248 
4249 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4250