xref: /freebsd/sys/kern/kern_descrip.c (revision ab00ac327a66a53edaac95b536b209db3ae2cd9f)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ddb.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 
48 #include <sys/capsicum.h>
49 #include <sys/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/file.h>
52 #include <sys/filedesc.h>
53 #include <sys/filio.h>
54 #include <sys/jail.h>
55 #include <sys/kernel.h>
56 #include <sys/limits.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/selinfo.h>
63 #include <sys/priv.h>
64 #include <sys/proc.h>
65 #include <sys/protosw.h>
66 #include <sys/racct.h>
67 #include <sys/resourcevar.h>
68 #include <sys/sbuf.h>
69 #include <sys/signalvar.h>
70 #include <sys/kdb.h>
71 #include <sys/stat.h>
72 #include <sys/sx.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysctl.h>
75 #include <sys/sysproto.h>
76 #include <sys/unistd.h>
77 #include <sys/user.h>
78 #include <sys/vnode.h>
79 #ifdef KTRACE
80 #include <sys/ktrace.h>
81 #endif
82 
83 #include <net/vnet.h>
84 
85 #include <security/audit/audit.h>
86 
87 #include <vm/uma.h>
88 #include <vm/vm.h>
89 
90 #include <ddb/ddb.h>
91 
92 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
93 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
94     "file desc to leader structures");
95 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
96 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
97 
98 MALLOC_DECLARE(M_FADVISE);
99 
100 static __read_mostly uma_zone_t file_zone;
101 static __read_mostly uma_zone_t filedesc0_zone;
102 
103 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
104 		    struct thread *td, int holdleaders);
105 static int	fd_first_free(struct filedesc *fdp, int low, int size);
106 static int	fd_last_used(struct filedesc *fdp, int size);
107 static void	fdgrowtable(struct filedesc *fdp, int nfd);
108 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
109 static void	fdunused(struct filedesc *fdp, int fd);
110 static void	fdused(struct filedesc *fdp, int fd);
111 static int	getmaxfd(struct thread *td);
112 
113 /*
114  * Each process has:
115  *
116  * - An array of open file descriptors (fd_ofiles)
117  * - An array of file flags (fd_ofileflags)
118  * - A bitmap recording which descriptors are in use (fd_map)
119  *
120  * A process starts out with NDFILE descriptors.  The value of NDFILE has
121  * been selected based the historical limit of 20 open files, and an
122  * assumption that the majority of processes, especially short-lived
123  * processes like shells, will never need more.
124  *
125  * If this initial allocation is exhausted, a larger descriptor table and
126  * map are allocated dynamically, and the pointers in the process's struct
127  * filedesc are updated to point to those.  This is repeated every time
128  * the process runs out of file descriptors (provided it hasn't hit its
129  * resource limit).
130  *
131  * Since threads may hold references to individual descriptor table
132  * entries, the tables are never freed.  Instead, they are placed on a
133  * linked list and freed only when the struct filedesc is released.
134  */
135 #define NDFILE		20
136 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
137 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
138 #define NDSLOT(x)	((x) / NDENTRIES)
139 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
140 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
141 
142 /*
143  * SLIST entry used to keep track of ofiles which must be reclaimed when
144  * the process exits.
145  */
146 struct freetable {
147 	struct fdescenttbl *ft_table;
148 	SLIST_ENTRY(freetable) ft_next;
149 };
150 
151 /*
152  * Initial allocation: a filedesc structure + the head of SLIST used to
153  * keep track of old ofiles + enough space for NDFILE descriptors.
154  */
155 
156 struct fdescenttbl0 {
157 	int	fdt_nfiles;
158 	struct	filedescent fdt_ofiles[NDFILE];
159 };
160 
161 struct filedesc0 {
162 	struct filedesc fd_fd;
163 	SLIST_HEAD(, freetable) fd_free;
164 	struct	fdescenttbl0 fd_dfiles;
165 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
166 };
167 
168 /*
169  * Descriptor management.
170  */
171 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
172 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
173 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
174 
175 /*
176  * If low >= size, just return low. Otherwise find the first zero bit in the
177  * given bitmap, starting at low and not exceeding size - 1. Return size if
178  * not found.
179  */
180 static int
181 fd_first_free(struct filedesc *fdp, int low, int size)
182 {
183 	NDSLOTTYPE *map = fdp->fd_map;
184 	NDSLOTTYPE mask;
185 	int off, maxoff;
186 
187 	if (low >= size)
188 		return (low);
189 
190 	off = NDSLOT(low);
191 	if (low % NDENTRIES) {
192 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
193 		if ((mask &= ~map[off]) != 0UL)
194 			return (off * NDENTRIES + ffsl(mask) - 1);
195 		++off;
196 	}
197 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
198 		if (map[off] != ~0UL)
199 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
200 	return (size);
201 }
202 
203 /*
204  * Find the highest non-zero bit in the given bitmap, starting at 0 and
205  * not exceeding size - 1. Return -1 if not found.
206  */
207 static int
208 fd_last_used(struct filedesc *fdp, int size)
209 {
210 	NDSLOTTYPE *map = fdp->fd_map;
211 	NDSLOTTYPE mask;
212 	int off, minoff;
213 
214 	off = NDSLOT(size);
215 	if (size % NDENTRIES) {
216 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
217 		if ((mask &= map[off]) != 0)
218 			return (off * NDENTRIES + flsl(mask) - 1);
219 		--off;
220 	}
221 	for (minoff = NDSLOT(0); off >= minoff; --off)
222 		if (map[off] != 0)
223 			return (off * NDENTRIES + flsl(map[off]) - 1);
224 	return (-1);
225 }
226 
227 static int
228 fdisused(struct filedesc *fdp, int fd)
229 {
230 
231 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
232 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
233 
234 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
235 }
236 
237 /*
238  * Mark a file descriptor as used.
239  */
240 static void
241 fdused_init(struct filedesc *fdp, int fd)
242 {
243 
244 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
245 
246 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
247 }
248 
249 static void
250 fdused(struct filedesc *fdp, int fd)
251 {
252 
253 	FILEDESC_XLOCK_ASSERT(fdp);
254 
255 	fdused_init(fdp, fd);
256 	if (fd > fdp->fd_lastfile)
257 		fdp->fd_lastfile = fd;
258 	if (fd == fdp->fd_freefile)
259 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
260 }
261 
262 /*
263  * Mark a file descriptor as unused.
264  */
265 static void
266 fdunused(struct filedesc *fdp, int fd)
267 {
268 
269 	FILEDESC_XLOCK_ASSERT(fdp);
270 
271 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
272 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
273 	    ("fd=%d is still in use", fd));
274 
275 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
276 	if (fd < fdp->fd_freefile)
277 		fdp->fd_freefile = fd;
278 	if (fd == fdp->fd_lastfile)
279 		fdp->fd_lastfile = fd_last_used(fdp, fd);
280 }
281 
282 /*
283  * Free a file descriptor.
284  *
285  * Avoid some work if fdp is about to be destroyed.
286  */
287 static inline void
288 fdefree_last(struct filedescent *fde)
289 {
290 
291 	filecaps_free(&fde->fde_caps);
292 }
293 
294 static inline void
295 fdfree(struct filedesc *fdp, int fd)
296 {
297 	struct filedescent *fde;
298 
299 	fde = &fdp->fd_ofiles[fd];
300 #ifdef CAPABILITIES
301 	seq_write_begin(&fde->fde_seq);
302 #endif
303 	fdefree_last(fde);
304 	fde->fde_file = NULL;
305 	fdunused(fdp, fd);
306 #ifdef CAPABILITIES
307 	seq_write_end(&fde->fde_seq);
308 #endif
309 }
310 
311 void
312 pwd_ensure_dirs(void)
313 {
314 	struct filedesc *fdp;
315 
316 	fdp = curproc->p_fd;
317 	FILEDESC_XLOCK(fdp);
318 	if (fdp->fd_cdir == NULL) {
319 		fdp->fd_cdir = rootvnode;
320 		vrefact(rootvnode);
321 	}
322 	if (fdp->fd_rdir == NULL) {
323 		fdp->fd_rdir = rootvnode;
324 		vrefact(rootvnode);
325 	}
326 	FILEDESC_XUNLOCK(fdp);
327 }
328 
329 /*
330  * System calls on descriptors.
331  */
332 #ifndef _SYS_SYSPROTO_H_
333 struct getdtablesize_args {
334 	int	dummy;
335 };
336 #endif
337 /* ARGSUSED */
338 int
339 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
340 {
341 #ifdef	RACCT
342 	uint64_t lim;
343 #endif
344 
345 	td->td_retval[0] =
346 	    min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
347 #ifdef	RACCT
348 	PROC_LOCK(td->td_proc);
349 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
350 	PROC_UNLOCK(td->td_proc);
351 	if (lim < td->td_retval[0])
352 		td->td_retval[0] = lim;
353 #endif
354 	return (0);
355 }
356 
357 /*
358  * Duplicate a file descriptor to a particular value.
359  *
360  * Note: keep in mind that a potential race condition exists when closing
361  * descriptors from a shared descriptor table (via rfork).
362  */
363 #ifndef _SYS_SYSPROTO_H_
364 struct dup2_args {
365 	u_int	from;
366 	u_int	to;
367 };
368 #endif
369 /* ARGSUSED */
370 int
371 sys_dup2(struct thread *td, struct dup2_args *uap)
372 {
373 
374 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
375 }
376 
377 /*
378  * Duplicate a file descriptor.
379  */
380 #ifndef _SYS_SYSPROTO_H_
381 struct dup_args {
382 	u_int	fd;
383 };
384 #endif
385 /* ARGSUSED */
386 int
387 sys_dup(struct thread *td, struct dup_args *uap)
388 {
389 
390 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
391 }
392 
393 /*
394  * The file control system call.
395  */
396 #ifndef _SYS_SYSPROTO_H_
397 struct fcntl_args {
398 	int	fd;
399 	int	cmd;
400 	long	arg;
401 };
402 #endif
403 /* ARGSUSED */
404 int
405 sys_fcntl(struct thread *td, struct fcntl_args *uap)
406 {
407 
408 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
409 }
410 
411 int
412 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
413 {
414 	struct flock fl;
415 	struct __oflock ofl;
416 	intptr_t arg1;
417 	int error, newcmd;
418 
419 	error = 0;
420 	newcmd = cmd;
421 	switch (cmd) {
422 	case F_OGETLK:
423 	case F_OSETLK:
424 	case F_OSETLKW:
425 		/*
426 		 * Convert old flock structure to new.
427 		 */
428 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
429 		fl.l_start = ofl.l_start;
430 		fl.l_len = ofl.l_len;
431 		fl.l_pid = ofl.l_pid;
432 		fl.l_type = ofl.l_type;
433 		fl.l_whence = ofl.l_whence;
434 		fl.l_sysid = 0;
435 
436 		switch (cmd) {
437 		case F_OGETLK:
438 			newcmd = F_GETLK;
439 			break;
440 		case F_OSETLK:
441 			newcmd = F_SETLK;
442 			break;
443 		case F_OSETLKW:
444 			newcmd = F_SETLKW;
445 			break;
446 		}
447 		arg1 = (intptr_t)&fl;
448 		break;
449 	case F_GETLK:
450 	case F_SETLK:
451 	case F_SETLKW:
452 	case F_SETLK_REMOTE:
453 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
454 		arg1 = (intptr_t)&fl;
455 		break;
456 	default:
457 		arg1 = arg;
458 		break;
459 	}
460 	if (error)
461 		return (error);
462 	error = kern_fcntl(td, fd, newcmd, arg1);
463 	if (error)
464 		return (error);
465 	if (cmd == F_OGETLK) {
466 		ofl.l_start = fl.l_start;
467 		ofl.l_len = fl.l_len;
468 		ofl.l_pid = fl.l_pid;
469 		ofl.l_type = fl.l_type;
470 		ofl.l_whence = fl.l_whence;
471 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
472 	} else if (cmd == F_GETLK) {
473 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
474 	}
475 	return (error);
476 }
477 
478 int
479 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
480 {
481 	struct filedesc *fdp;
482 	struct flock *flp;
483 	struct file *fp, *fp2;
484 	struct filedescent *fde;
485 	struct proc *p;
486 	struct vnode *vp;
487 	cap_rights_t rights;
488 	int error, flg, tmp;
489 	uint64_t bsize;
490 	off_t foffset;
491 
492 	error = 0;
493 	flg = F_POSIX;
494 	p = td->td_proc;
495 	fdp = p->p_fd;
496 
497 	AUDIT_ARG_FD(cmd);
498 	AUDIT_ARG_CMD(cmd);
499 	switch (cmd) {
500 	case F_DUPFD:
501 		tmp = arg;
502 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
503 		break;
504 
505 	case F_DUPFD_CLOEXEC:
506 		tmp = arg;
507 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
508 		break;
509 
510 	case F_DUP2FD:
511 		tmp = arg;
512 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
513 		break;
514 
515 	case F_DUP2FD_CLOEXEC:
516 		tmp = arg;
517 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
518 		break;
519 
520 	case F_GETFD:
521 		error = EBADF;
522 		FILEDESC_SLOCK(fdp);
523 		fde = fdeget_locked(fdp, fd);
524 		if (fde != NULL) {
525 			td->td_retval[0] =
526 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
527 			error = 0;
528 		}
529 		FILEDESC_SUNLOCK(fdp);
530 		break;
531 
532 	case F_SETFD:
533 		error = EBADF;
534 		FILEDESC_XLOCK(fdp);
535 		fde = fdeget_locked(fdp, fd);
536 		if (fde != NULL) {
537 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
538 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
539 			error = 0;
540 		}
541 		FILEDESC_XUNLOCK(fdp);
542 		break;
543 
544 	case F_GETFL:
545 		error = fget_fcntl(td, fd,
546 		    cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp);
547 		if (error != 0)
548 			break;
549 		td->td_retval[0] = OFLAGS(fp->f_flag);
550 		fdrop(fp, td);
551 		break;
552 
553 	case F_SETFL:
554 		error = fget_fcntl(td, fd,
555 		    cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp);
556 		if (error != 0)
557 			break;
558 		do {
559 			tmp = flg = fp->f_flag;
560 			tmp &= ~FCNTLFLAGS;
561 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
562 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
563 		tmp = fp->f_flag & FNONBLOCK;
564 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
565 		if (error != 0) {
566 			fdrop(fp, td);
567 			break;
568 		}
569 		tmp = fp->f_flag & FASYNC;
570 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
571 		if (error == 0) {
572 			fdrop(fp, td);
573 			break;
574 		}
575 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
576 		tmp = 0;
577 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
578 		fdrop(fp, td);
579 		break;
580 
581 	case F_GETOWN:
582 		error = fget_fcntl(td, fd,
583 		    cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
587 		if (error == 0)
588 			td->td_retval[0] = tmp;
589 		fdrop(fp, td);
590 		break;
591 
592 	case F_SETOWN:
593 		error = fget_fcntl(td, fd,
594 		    cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp);
595 		if (error != 0)
596 			break;
597 		tmp = arg;
598 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
599 		fdrop(fp, td);
600 		break;
601 
602 	case F_SETLK_REMOTE:
603 		error = priv_check(td, PRIV_NFS_LOCKD);
604 		if (error)
605 			return (error);
606 		flg = F_REMOTE;
607 		goto do_setlk;
608 
609 	case F_SETLKW:
610 		flg |= F_WAIT;
611 		/* FALLTHROUGH F_SETLK */
612 
613 	case F_SETLK:
614 	do_setlk:
615 		cap_rights_init(&rights, CAP_FLOCK);
616 		error = fget_unlocked(fdp, fd, &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 			PROC_LOCK(p->p_leader);
646 			p->p_leader->p_flag |= P_ADVLOCK;
647 			PROC_UNLOCK(p->p_leader);
648 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
649 			    flp, flg);
650 			break;
651 		case F_WRLCK:
652 			if ((fp->f_flag & FWRITE) == 0) {
653 				error = EBADF;
654 				break;
655 			}
656 			PROC_LOCK(p->p_leader);
657 			p->p_leader->p_flag |= P_ADVLOCK;
658 			PROC_UNLOCK(p->p_leader);
659 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
660 			    flp, flg);
661 			break;
662 		case F_UNLCK:
663 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
664 			    flp, flg);
665 			break;
666 		case F_UNLCKSYS:
667 			/*
668 			 * Temporary api for testing remote lock
669 			 * infrastructure.
670 			 */
671 			if (flg != F_REMOTE) {
672 				error = EINVAL;
673 				break;
674 			}
675 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
676 			    F_UNLCKSYS, flp, flg);
677 			break;
678 		default:
679 			error = EINVAL;
680 			break;
681 		}
682 		if (error != 0 || flp->l_type == F_UNLCK ||
683 		    flp->l_type == F_UNLCKSYS) {
684 			fdrop(fp, td);
685 			break;
686 		}
687 
688 		/*
689 		 * Check for a race with close.
690 		 *
691 		 * The vnode is now advisory locked (or unlocked, but this case
692 		 * is not really important) as the caller requested.
693 		 * We had to drop the filedesc lock, so we need to recheck if
694 		 * the descriptor is still valid, because if it was closed
695 		 * in the meantime we need to remove advisory lock from the
696 		 * vnode - close on any descriptor leading to an advisory
697 		 * locked vnode, removes that lock.
698 		 * We will return 0 on purpose in that case, as the result of
699 		 * successful advisory lock might have been externally visible
700 		 * already. This is fine - effectively we pretend to the caller
701 		 * that the closing thread was a bit slower and that the
702 		 * advisory lock succeeded before the close.
703 		 */
704 		error = fget_unlocked(fdp, fd, &rights, &fp2, NULL);
705 		if (error != 0) {
706 			fdrop(fp, td);
707 			break;
708 		}
709 		if (fp != fp2) {
710 			flp->l_whence = SEEK_SET;
711 			flp->l_start = 0;
712 			flp->l_len = 0;
713 			flp->l_type = F_UNLCK;
714 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
715 			    F_UNLCK, flp, F_POSIX);
716 		}
717 		fdrop(fp, td);
718 		fdrop(fp2, td);
719 		break;
720 
721 	case F_GETLK:
722 		error = fget_unlocked(fdp, fd,
723 		    cap_rights_init(&rights, CAP_FLOCK), &fp, NULL);
724 		if (error != 0)
725 			break;
726 		if (fp->f_type != DTYPE_VNODE) {
727 			error = EBADF;
728 			fdrop(fp, td);
729 			break;
730 		}
731 		flp = (struct flock *)arg;
732 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
733 		    flp->l_type != F_UNLCK) {
734 			error = EINVAL;
735 			fdrop(fp, td);
736 			break;
737 		}
738 		if (flp->l_whence == SEEK_CUR) {
739 			foffset = foffset_get(fp);
740 			if ((flp->l_start > 0 &&
741 			    foffset > OFF_MAX - flp->l_start) ||
742 			    (flp->l_start < 0 &&
743 			    foffset < OFF_MIN - flp->l_start)) {
744 				error = EOVERFLOW;
745 				fdrop(fp, td);
746 				break;
747 			}
748 			flp->l_start += foffset;
749 		}
750 		vp = fp->f_vnode;
751 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
752 		    F_POSIX);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_RDAHEAD:
757 		arg = arg ? 128 * 1024: 0;
758 		/* FALLTHROUGH */
759 	case F_READAHEAD:
760 		error = fget_unlocked(fdp, fd,
761 		    cap_rights_init(&rights), &fp, NULL);
762 		if (error != 0)
763 			break;
764 		if (fp->f_type != DTYPE_VNODE) {
765 			fdrop(fp, td);
766 			error = EBADF;
767 			break;
768 		}
769 		vp = fp->f_vnode;
770 		/*
771 		 * Exclusive lock synchronizes against f_seqcount reads and
772 		 * writes in sequential_heuristic().
773 		 */
774 		error = vn_lock(vp, LK_EXCLUSIVE);
775 		if (error != 0) {
776 			fdrop(fp, td);
777 			break;
778 		}
779 		if (arg >= 0) {
780 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
781 			fp->f_seqcount = (arg + bsize - 1) / bsize;
782 			atomic_set_int(&fp->f_flag, FRDAHEAD);
783 		} else {
784 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
785 		}
786 		VOP_UNLOCK(vp, 0);
787 		fdrop(fp, td);
788 		break;
789 
790 	default:
791 		error = EINVAL;
792 		break;
793 	}
794 	return (error);
795 }
796 
797 static int
798 getmaxfd(struct thread *td)
799 {
800 
801 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
802 }
803 
804 /*
805  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
806  */
807 int
808 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
809 {
810 	struct filedesc *fdp;
811 	struct filedescent *oldfde, *newfde;
812 	struct proc *p;
813 	struct file *delfp;
814 	int error, maxfd;
815 
816 	p = td->td_proc;
817 	fdp = p->p_fd;
818 
819 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
820 	MPASS(mode < FDDUP_LASTMODE);
821 
822 	AUDIT_ARG_FD(old);
823 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
824 
825 	/*
826 	 * Verify we have a valid descriptor to dup from and possibly to
827 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
828 	 * return EINVAL when the new descriptor is out of bounds.
829 	 */
830 	if (old < 0)
831 		return (EBADF);
832 	if (new < 0)
833 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
834 	maxfd = getmaxfd(td);
835 	if (new >= maxfd)
836 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
837 
838 	error = EBADF;
839 	FILEDESC_XLOCK(fdp);
840 	if (fget_locked(fdp, old) == NULL)
841 		goto unlock;
842 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
843 		td->td_retval[0] = new;
844 		if (flags & FDDUP_FLAG_CLOEXEC)
845 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
846 		error = 0;
847 		goto unlock;
848 	}
849 
850 	/*
851 	 * If the caller specified a file descriptor, make sure the file
852 	 * table is large enough to hold it, and grab it.  Otherwise, just
853 	 * allocate a new descriptor the usual way.
854 	 */
855 	switch (mode) {
856 	case FDDUP_NORMAL:
857 	case FDDUP_FCNTL:
858 		if ((error = fdalloc(td, new, &new)) != 0)
859 			goto unlock;
860 		break;
861 	case FDDUP_MUSTREPLACE:
862 		/* Target file descriptor must exist. */
863 		if (fget_locked(fdp, new) == NULL)
864 			goto unlock;
865 		break;
866 	case FDDUP_FIXED:
867 		if (new >= fdp->fd_nfiles) {
868 			/*
869 			 * The resource limits are here instead of e.g.
870 			 * fdalloc(), because the file descriptor table may be
871 			 * shared between processes, so we can't really use
872 			 * racct_add()/racct_sub().  Instead of counting the
873 			 * number of actually allocated descriptors, just put
874 			 * the limit on the size of the file descriptor table.
875 			 */
876 #ifdef RACCT
877 			if (racct_enable) {
878 				PROC_LOCK(p);
879 				error = racct_set(p, RACCT_NOFILE, new + 1);
880 				PROC_UNLOCK(p);
881 				if (error != 0) {
882 					error = EMFILE;
883 					goto unlock;
884 				}
885 			}
886 #endif
887 			fdgrowtable_exp(fdp, new + 1);
888 		}
889 		if (!fdisused(fdp, new))
890 			fdused(fdp, new);
891 		break;
892 	default:
893 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
894 	}
895 
896 	KASSERT(old != new, ("new fd is same as old"));
897 
898 	oldfde = &fdp->fd_ofiles[old];
899 	fhold(oldfde->fde_file);
900 	newfde = &fdp->fd_ofiles[new];
901 	delfp = newfde->fde_file;
902 
903 	/*
904 	 * Duplicate the source descriptor.
905 	 */
906 #ifdef CAPABILITIES
907 	seq_write_begin(&newfde->fde_seq);
908 #endif
909 	filecaps_free(&newfde->fde_caps);
910 	memcpy(newfde, oldfde, fde_change_size);
911 	filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
912 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
913 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
914 	else
915 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
916 #ifdef CAPABILITIES
917 	seq_write_end(&newfde->fde_seq);
918 #endif
919 	td->td_retval[0] = new;
920 
921 	error = 0;
922 
923 	if (delfp != NULL) {
924 		(void) closefp(fdp, new, delfp, td, 1);
925 		FILEDESC_UNLOCK_ASSERT(fdp);
926 	} else {
927 unlock:
928 		FILEDESC_XUNLOCK(fdp);
929 	}
930 
931 	return (error);
932 }
933 
934 /*
935  * If sigio is on the list associated with a process or process group,
936  * disable signalling from the device, remove sigio from the list and
937  * free sigio.
938  */
939 void
940 funsetown(struct sigio **sigiop)
941 {
942 	struct sigio *sigio;
943 
944 	if (*sigiop == NULL)
945 		return;
946 	SIGIO_LOCK();
947 	sigio = *sigiop;
948 	if (sigio == NULL) {
949 		SIGIO_UNLOCK();
950 		return;
951 	}
952 	*(sigio->sio_myref) = NULL;
953 	if ((sigio)->sio_pgid < 0) {
954 		struct pgrp *pg = (sigio)->sio_pgrp;
955 		PGRP_LOCK(pg);
956 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
957 			    sigio, sio_pgsigio);
958 		PGRP_UNLOCK(pg);
959 	} else {
960 		struct proc *p = (sigio)->sio_proc;
961 		PROC_LOCK(p);
962 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
963 			    sigio, sio_pgsigio);
964 		PROC_UNLOCK(p);
965 	}
966 	SIGIO_UNLOCK();
967 	crfree(sigio->sio_ucred);
968 	free(sigio, M_SIGIO);
969 }
970 
971 /*
972  * Free a list of sigio structures.
973  * We only need to lock the SIGIO_LOCK because we have made ourselves
974  * inaccessible to callers of fsetown and therefore do not need to lock
975  * the proc or pgrp struct for the list manipulation.
976  */
977 void
978 funsetownlst(struct sigiolst *sigiolst)
979 {
980 	struct proc *p;
981 	struct pgrp *pg;
982 	struct sigio *sigio;
983 
984 	sigio = SLIST_FIRST(sigiolst);
985 	if (sigio == NULL)
986 		return;
987 	p = NULL;
988 	pg = NULL;
989 
990 	/*
991 	 * Every entry of the list should belong
992 	 * to a single proc or pgrp.
993 	 */
994 	if (sigio->sio_pgid < 0) {
995 		pg = sigio->sio_pgrp;
996 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
997 	} else /* if (sigio->sio_pgid > 0) */ {
998 		p = sigio->sio_proc;
999 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1000 	}
1001 
1002 	SIGIO_LOCK();
1003 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1004 		*(sigio->sio_myref) = NULL;
1005 		if (pg != NULL) {
1006 			KASSERT(sigio->sio_pgid < 0,
1007 			    ("Proc sigio in pgrp sigio list"));
1008 			KASSERT(sigio->sio_pgrp == pg,
1009 			    ("Bogus pgrp in sigio list"));
1010 			PGRP_LOCK(pg);
1011 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1012 			    sio_pgsigio);
1013 			PGRP_UNLOCK(pg);
1014 		} else /* if (p != NULL) */ {
1015 			KASSERT(sigio->sio_pgid > 0,
1016 			    ("Pgrp sigio in proc sigio list"));
1017 			KASSERT(sigio->sio_proc == p,
1018 			    ("Bogus proc in sigio list"));
1019 			PROC_LOCK(p);
1020 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1021 			    sio_pgsigio);
1022 			PROC_UNLOCK(p);
1023 		}
1024 		SIGIO_UNLOCK();
1025 		crfree(sigio->sio_ucred);
1026 		free(sigio, M_SIGIO);
1027 		SIGIO_LOCK();
1028 	}
1029 	SIGIO_UNLOCK();
1030 }
1031 
1032 /*
1033  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1034  *
1035  * After permission checking, add a sigio structure to the sigio list for
1036  * the process or process group.
1037  */
1038 int
1039 fsetown(pid_t pgid, struct sigio **sigiop)
1040 {
1041 	struct proc *proc;
1042 	struct pgrp *pgrp;
1043 	struct sigio *sigio;
1044 	int ret;
1045 
1046 	if (pgid == 0) {
1047 		funsetown(sigiop);
1048 		return (0);
1049 	}
1050 
1051 	ret = 0;
1052 
1053 	/* Allocate and fill in the new sigio out of locks. */
1054 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1055 	sigio->sio_pgid = pgid;
1056 	sigio->sio_ucred = crhold(curthread->td_ucred);
1057 	sigio->sio_myref = sigiop;
1058 
1059 	sx_slock(&proctree_lock);
1060 	if (pgid > 0) {
1061 		proc = pfind(pgid);
1062 		if (proc == NULL) {
1063 			ret = ESRCH;
1064 			goto fail;
1065 		}
1066 
1067 		/*
1068 		 * Policy - Don't allow a process to FSETOWN a process
1069 		 * in another session.
1070 		 *
1071 		 * Remove this test to allow maximum flexibility or
1072 		 * restrict FSETOWN to the current process or process
1073 		 * group for maximum safety.
1074 		 */
1075 		PROC_UNLOCK(proc);
1076 		if (proc->p_session != curthread->td_proc->p_session) {
1077 			ret = EPERM;
1078 			goto fail;
1079 		}
1080 
1081 		pgrp = NULL;
1082 	} else /* if (pgid < 0) */ {
1083 		pgrp = pgfind(-pgid);
1084 		if (pgrp == NULL) {
1085 			ret = ESRCH;
1086 			goto fail;
1087 		}
1088 		PGRP_UNLOCK(pgrp);
1089 
1090 		/*
1091 		 * Policy - Don't allow a process to FSETOWN a process
1092 		 * in another session.
1093 		 *
1094 		 * Remove this test to allow maximum flexibility or
1095 		 * restrict FSETOWN to the current process or process
1096 		 * group for maximum safety.
1097 		 */
1098 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1099 			ret = EPERM;
1100 			goto fail;
1101 		}
1102 
1103 		proc = NULL;
1104 	}
1105 	funsetown(sigiop);
1106 	if (pgid > 0) {
1107 		PROC_LOCK(proc);
1108 		/*
1109 		 * Since funsetownlst() is called without the proctree
1110 		 * locked, we need to check for P_WEXIT.
1111 		 * XXX: is ESRCH correct?
1112 		 */
1113 		if ((proc->p_flag & P_WEXIT) != 0) {
1114 			PROC_UNLOCK(proc);
1115 			ret = ESRCH;
1116 			goto fail;
1117 		}
1118 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1119 		sigio->sio_proc = proc;
1120 		PROC_UNLOCK(proc);
1121 	} else {
1122 		PGRP_LOCK(pgrp);
1123 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1124 		sigio->sio_pgrp = pgrp;
1125 		PGRP_UNLOCK(pgrp);
1126 	}
1127 	sx_sunlock(&proctree_lock);
1128 	SIGIO_LOCK();
1129 	*sigiop = sigio;
1130 	SIGIO_UNLOCK();
1131 	return (0);
1132 
1133 fail:
1134 	sx_sunlock(&proctree_lock);
1135 	crfree(sigio->sio_ucred);
1136 	free(sigio, M_SIGIO);
1137 	return (ret);
1138 }
1139 
1140 /*
1141  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1142  */
1143 pid_t
1144 fgetown(sigiop)
1145 	struct sigio **sigiop;
1146 {
1147 	pid_t pgid;
1148 
1149 	SIGIO_LOCK();
1150 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1151 	SIGIO_UNLOCK();
1152 	return (pgid);
1153 }
1154 
1155 /*
1156  * Function drops the filedesc lock on return.
1157  */
1158 static int
1159 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1160     int holdleaders)
1161 {
1162 	int error;
1163 
1164 	FILEDESC_XLOCK_ASSERT(fdp);
1165 
1166 	if (holdleaders) {
1167 		if (td->td_proc->p_fdtol != NULL) {
1168 			/*
1169 			 * Ask fdfree() to sleep to ensure that all relevant
1170 			 * process leaders can be traversed in closef().
1171 			 */
1172 			fdp->fd_holdleaderscount++;
1173 		} else {
1174 			holdleaders = 0;
1175 		}
1176 	}
1177 
1178 	/*
1179 	 * We now hold the fp reference that used to be owned by the
1180 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1181 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1182 	 * added, and deleteing a knote for the new fd.
1183 	 */
1184 	knote_fdclose(td, fd);
1185 
1186 	/*
1187 	 * We need to notify mqueue if the object is of type mqueue.
1188 	 */
1189 	if (fp->f_type == DTYPE_MQUEUE)
1190 		mq_fdclose(td, fd, fp);
1191 	FILEDESC_XUNLOCK(fdp);
1192 
1193 	error = closef(fp, td);
1194 	if (holdleaders) {
1195 		FILEDESC_XLOCK(fdp);
1196 		fdp->fd_holdleaderscount--;
1197 		if (fdp->fd_holdleaderscount == 0 &&
1198 		    fdp->fd_holdleaderswakeup != 0) {
1199 			fdp->fd_holdleaderswakeup = 0;
1200 			wakeup(&fdp->fd_holdleaderscount);
1201 		}
1202 		FILEDESC_XUNLOCK(fdp);
1203 	}
1204 	return (error);
1205 }
1206 
1207 /*
1208  * Close a file descriptor.
1209  */
1210 #ifndef _SYS_SYSPROTO_H_
1211 struct close_args {
1212 	int     fd;
1213 };
1214 #endif
1215 /* ARGSUSED */
1216 int
1217 sys_close(struct thread *td, struct close_args *uap)
1218 {
1219 
1220 	return (kern_close(td, uap->fd));
1221 }
1222 
1223 int
1224 kern_close(struct thread *td, int fd)
1225 {
1226 	struct filedesc *fdp;
1227 	struct file *fp;
1228 
1229 	fdp = td->td_proc->p_fd;
1230 
1231 	AUDIT_SYSCLOSE(td, fd);
1232 
1233 	FILEDESC_XLOCK(fdp);
1234 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1235 		FILEDESC_XUNLOCK(fdp);
1236 		return (EBADF);
1237 	}
1238 	fdfree(fdp, fd);
1239 
1240 	/* closefp() drops the FILEDESC lock for us. */
1241 	return (closefp(fdp, fd, fp, td, 1));
1242 }
1243 
1244 /*
1245  * Close open file descriptors.
1246  */
1247 #ifndef _SYS_SYSPROTO_H_
1248 struct closefrom_args {
1249 	int	lowfd;
1250 };
1251 #endif
1252 /* ARGSUSED */
1253 int
1254 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1255 {
1256 	struct filedesc *fdp;
1257 	int fd;
1258 
1259 	fdp = td->td_proc->p_fd;
1260 	AUDIT_ARG_FD(uap->lowfd);
1261 
1262 	/*
1263 	 * Treat negative starting file descriptor values identical to
1264 	 * closefrom(0) which closes all files.
1265 	 */
1266 	if (uap->lowfd < 0)
1267 		uap->lowfd = 0;
1268 	FILEDESC_SLOCK(fdp);
1269 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1270 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1271 			FILEDESC_SUNLOCK(fdp);
1272 			(void)kern_close(td, fd);
1273 			FILEDESC_SLOCK(fdp);
1274 		}
1275 	}
1276 	FILEDESC_SUNLOCK(fdp);
1277 	return (0);
1278 }
1279 
1280 #if defined(COMPAT_43)
1281 /*
1282  * Return status information about a file descriptor.
1283  */
1284 #ifndef _SYS_SYSPROTO_H_
1285 struct ofstat_args {
1286 	int	fd;
1287 	struct	ostat *sb;
1288 };
1289 #endif
1290 /* ARGSUSED */
1291 int
1292 ofstat(struct thread *td, struct ofstat_args *uap)
1293 {
1294 	struct ostat oub;
1295 	struct stat ub;
1296 	int error;
1297 
1298 	error = kern_fstat(td, uap->fd, &ub);
1299 	if (error == 0) {
1300 		cvtstat(&ub, &oub);
1301 		error = copyout(&oub, uap->sb, sizeof(oub));
1302 	}
1303 	return (error);
1304 }
1305 #endif /* COMPAT_43 */
1306 
1307 #if defined(COMPAT_FREEBSD11)
1308 int
1309 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1310 {
1311 	struct stat sb;
1312 	struct freebsd11_stat osb;
1313 	int error;
1314 
1315 	error = kern_fstat(td, uap->fd, &sb);
1316 	if (error != 0)
1317 		return (error);
1318 	freebsd11_cvtstat(&sb, &osb);
1319 	error = copyout(&osb, uap->sb, sizeof(osb));
1320 	return (error);
1321 }
1322 #endif	/* COMPAT_FREEBSD11 */
1323 
1324 /*
1325  * Return status information about a file descriptor.
1326  */
1327 #ifndef _SYS_SYSPROTO_H_
1328 struct fstat_args {
1329 	int	fd;
1330 	struct	stat *sb;
1331 };
1332 #endif
1333 /* ARGSUSED */
1334 int
1335 sys_fstat(struct thread *td, struct fstat_args *uap)
1336 {
1337 	struct stat ub;
1338 	int error;
1339 
1340 	error = kern_fstat(td, uap->fd, &ub);
1341 	if (error == 0)
1342 		error = copyout(&ub, uap->sb, sizeof(ub));
1343 	return (error);
1344 }
1345 
1346 int
1347 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1348 {
1349 	struct file *fp;
1350 	cap_rights_t rights;
1351 	int error;
1352 
1353 	AUDIT_ARG_FD(fd);
1354 
1355 	error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1356 	if (error != 0)
1357 		return (error);
1358 
1359 	AUDIT_ARG_FILE(td->td_proc, fp);
1360 
1361 	error = fo_stat(fp, sbp, td->td_ucred, td);
1362 	fdrop(fp, td);
1363 #ifdef __STAT_TIME_T_EXT
1364 	if (error == 0) {
1365 		sbp->st_atim_ext = 0;
1366 		sbp->st_mtim_ext = 0;
1367 		sbp->st_ctim_ext = 0;
1368 		sbp->st_btim_ext = 0;
1369 	}
1370 #endif
1371 #ifdef KTRACE
1372 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1373 		ktrstat(sbp);
1374 #endif
1375 	return (error);
1376 }
1377 
1378 #if defined(COMPAT_FREEBSD11)
1379 /*
1380  * Return status information about a file descriptor.
1381  */
1382 #ifndef _SYS_SYSPROTO_H_
1383 struct freebsd11_nfstat_args {
1384 	int	fd;
1385 	struct	nstat *sb;
1386 };
1387 #endif
1388 /* ARGSUSED */
1389 int
1390 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1391 {
1392 	struct nstat nub;
1393 	struct stat ub;
1394 	int error;
1395 
1396 	error = kern_fstat(td, uap->fd, &ub);
1397 	if (error == 0) {
1398 		freebsd11_cvtnstat(&ub, &nub);
1399 		error = copyout(&nub, uap->sb, sizeof(nub));
1400 	}
1401 	return (error);
1402 }
1403 #endif /* COMPAT_FREEBSD11 */
1404 
1405 /*
1406  * Return pathconf information about a file descriptor.
1407  */
1408 #ifndef _SYS_SYSPROTO_H_
1409 struct fpathconf_args {
1410 	int	fd;
1411 	int	name;
1412 };
1413 #endif
1414 /* ARGSUSED */
1415 int
1416 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1417 {
1418 	struct file *fp;
1419 	struct vnode *vp;
1420 	cap_rights_t rights;
1421 	int error;
1422 
1423 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1424 	if (error != 0)
1425 		return (error);
1426 
1427 	if (uap->name == _PC_ASYNC_IO) {
1428 		td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1429 		goto out;
1430 	}
1431 	vp = fp->f_vnode;
1432 	if (vp != NULL) {
1433 		vn_lock(vp, LK_SHARED | LK_RETRY);
1434 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1435 		VOP_UNLOCK(vp, 0);
1436 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1437 		if (uap->name != _PC_PIPE_BUF) {
1438 			error = EINVAL;
1439 		} else {
1440 			td->td_retval[0] = PIPE_BUF;
1441 			error = 0;
1442 		}
1443 	} else {
1444 		error = EOPNOTSUPP;
1445 	}
1446 out:
1447 	fdrop(fp, td);
1448 	return (error);
1449 }
1450 
1451 /*
1452  * Initialize filecaps structure.
1453  */
1454 void
1455 filecaps_init(struct filecaps *fcaps)
1456 {
1457 
1458 	bzero(fcaps, sizeof(*fcaps));
1459 	fcaps->fc_nioctls = -1;
1460 }
1461 
1462 /*
1463  * Copy filecaps structure allocating memory for ioctls array if needed.
1464  *
1465  * The last parameter indicates whether the fdtable is locked. If it is not and
1466  * ioctls are encountered, copying fails and the caller must lock the table.
1467  *
1468  * Note that if the table was not locked, the caller has to check the relevant
1469  * sequence counter to determine whether the operation was successful.
1470  */
1471 int
1472 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1473 {
1474 	size_t size;
1475 
1476 	*dst = *src;
1477 	if (src->fc_ioctls == NULL)
1478 		return (0);
1479 	if (!locked)
1480 		return (1);
1481 
1482 	KASSERT(src->fc_nioctls > 0,
1483 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1484 
1485 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1486 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1487 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1488 	return (0);
1489 }
1490 
1491 /*
1492  * Move filecaps structure to the new place and clear the old place.
1493  */
1494 void
1495 filecaps_move(struct filecaps *src, struct filecaps *dst)
1496 {
1497 
1498 	*dst = *src;
1499 	bzero(src, sizeof(*src));
1500 }
1501 
1502 /*
1503  * Fill the given filecaps structure with full rights.
1504  */
1505 static void
1506 filecaps_fill(struct filecaps *fcaps)
1507 {
1508 
1509 	CAP_ALL(&fcaps->fc_rights);
1510 	fcaps->fc_ioctls = NULL;
1511 	fcaps->fc_nioctls = -1;
1512 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1513 }
1514 
1515 /*
1516  * Free memory allocated within filecaps structure.
1517  */
1518 void
1519 filecaps_free(struct filecaps *fcaps)
1520 {
1521 
1522 	free(fcaps->fc_ioctls, M_FILECAPS);
1523 	bzero(fcaps, sizeof(*fcaps));
1524 }
1525 
1526 /*
1527  * Validate the given filecaps structure.
1528  */
1529 static void
1530 filecaps_validate(const struct filecaps *fcaps, const char *func)
1531 {
1532 
1533 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1534 	    ("%s: invalid rights", func));
1535 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1536 	    ("%s: invalid fcntls", func));
1537 	KASSERT(fcaps->fc_fcntls == 0 ||
1538 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1539 	    ("%s: fcntls without CAP_FCNTL", func));
1540 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1541 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1542 	    ("%s: invalid ioctls", func));
1543 	KASSERT(fcaps->fc_nioctls == 0 ||
1544 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1545 	    ("%s: ioctls without CAP_IOCTL", func));
1546 }
1547 
1548 static void
1549 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1550 {
1551 	int nfd1;
1552 
1553 	FILEDESC_XLOCK_ASSERT(fdp);
1554 
1555 	nfd1 = fdp->fd_nfiles * 2;
1556 	if (nfd1 < nfd)
1557 		nfd1 = nfd;
1558 	fdgrowtable(fdp, nfd1);
1559 }
1560 
1561 /*
1562  * Grow the file table to accommodate (at least) nfd descriptors.
1563  */
1564 static void
1565 fdgrowtable(struct filedesc *fdp, int nfd)
1566 {
1567 	struct filedesc0 *fdp0;
1568 	struct freetable *ft;
1569 	struct fdescenttbl *ntable;
1570 	struct fdescenttbl *otable;
1571 	int nnfiles, onfiles;
1572 	NDSLOTTYPE *nmap, *omap;
1573 
1574 	/*
1575 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1576 	 * growing it to accommodate for the one we are going to copy from. There
1577 	 * is no need to have a lock on this one as it's not visible to anyone.
1578 	 */
1579 	if (fdp->fd_lastfile != -1)
1580 		FILEDESC_XLOCK_ASSERT(fdp);
1581 
1582 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1583 
1584 	/* save old values */
1585 	onfiles = fdp->fd_nfiles;
1586 	otable = fdp->fd_files;
1587 	omap = fdp->fd_map;
1588 
1589 	/* compute the size of the new table */
1590 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1591 	if (nnfiles <= onfiles)
1592 		/* the table is already large enough */
1593 		return;
1594 
1595 	/*
1596 	 * Allocate a new table.  We need enough space for the number of
1597 	 * entries, file entries themselves and the struct freetable we will use
1598 	 * when we decommission the table and place it on the freelist.
1599 	 * We place the struct freetable in the middle so we don't have
1600 	 * to worry about padding.
1601 	 */
1602 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1603 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1604 	    sizeof(struct freetable),
1605 	    M_FILEDESC, M_ZERO | M_WAITOK);
1606 	/* copy the old data */
1607 	ntable->fdt_nfiles = nnfiles;
1608 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1609 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1610 
1611 	/*
1612 	 * Allocate a new map only if the old is not large enough.  It will
1613 	 * grow at a slower rate than the table as it can map more
1614 	 * entries than the table can hold.
1615 	 */
1616 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1617 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1618 		    M_ZERO | M_WAITOK);
1619 		/* copy over the old data and update the pointer */
1620 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1621 		fdp->fd_map = nmap;
1622 	}
1623 
1624 	/*
1625 	 * Make sure that ntable is correctly initialized before we replace
1626 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1627 	 * data.
1628 	 */
1629 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1630 
1631 	/*
1632 	 * Do not free the old file table, as some threads may still
1633 	 * reference entries within it.  Instead, place it on a freelist
1634 	 * which will be processed when the struct filedesc is released.
1635 	 *
1636 	 * Note that if onfiles == NDFILE, we're dealing with the original
1637 	 * static allocation contained within (struct filedesc0 *)fdp,
1638 	 * which must not be freed.
1639 	 */
1640 	if (onfiles > NDFILE) {
1641 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1642 		fdp0 = (struct filedesc0 *)fdp;
1643 		ft->ft_table = otable;
1644 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1645 	}
1646 	/*
1647 	 * The map does not have the same possibility of threads still
1648 	 * holding references to it.  So always free it as long as it
1649 	 * does not reference the original static allocation.
1650 	 */
1651 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1652 		free(omap, M_FILEDESC);
1653 }
1654 
1655 /*
1656  * Allocate a file descriptor for the process.
1657  */
1658 int
1659 fdalloc(struct thread *td, int minfd, int *result)
1660 {
1661 	struct proc *p = td->td_proc;
1662 	struct filedesc *fdp = p->p_fd;
1663 	int fd, maxfd, allocfd;
1664 #ifdef RACCT
1665 	int error;
1666 #endif
1667 
1668 	FILEDESC_XLOCK_ASSERT(fdp);
1669 
1670 	if (fdp->fd_freefile > minfd)
1671 		minfd = fdp->fd_freefile;
1672 
1673 	maxfd = getmaxfd(td);
1674 
1675 	/*
1676 	 * Search the bitmap for a free descriptor starting at minfd.
1677 	 * If none is found, grow the file table.
1678 	 */
1679 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1680 	if (fd >= maxfd)
1681 		return (EMFILE);
1682 	if (fd >= fdp->fd_nfiles) {
1683 		allocfd = min(fd * 2, maxfd);
1684 #ifdef RACCT
1685 		if (racct_enable) {
1686 			PROC_LOCK(p);
1687 			error = racct_set(p, RACCT_NOFILE, allocfd);
1688 			PROC_UNLOCK(p);
1689 			if (error != 0)
1690 				return (EMFILE);
1691 		}
1692 #endif
1693 		/*
1694 		 * fd is already equal to first free descriptor >= minfd, so
1695 		 * we only need to grow the table and we are done.
1696 		 */
1697 		fdgrowtable_exp(fdp, allocfd);
1698 	}
1699 
1700 	/*
1701 	 * Perform some sanity checks, then mark the file descriptor as
1702 	 * used and return it to the caller.
1703 	 */
1704 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1705 	    ("invalid descriptor %d", fd));
1706 	KASSERT(!fdisused(fdp, fd),
1707 	    ("fd_first_free() returned non-free descriptor"));
1708 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1709 	    ("file descriptor isn't free"));
1710 	fdused(fdp, fd);
1711 	*result = fd;
1712 	return (0);
1713 }
1714 
1715 /*
1716  * Allocate n file descriptors for the process.
1717  */
1718 int
1719 fdallocn(struct thread *td, int minfd, int *fds, int n)
1720 {
1721 	struct proc *p = td->td_proc;
1722 	struct filedesc *fdp = p->p_fd;
1723 	int i;
1724 
1725 	FILEDESC_XLOCK_ASSERT(fdp);
1726 
1727 	for (i = 0; i < n; i++)
1728 		if (fdalloc(td, 0, &fds[i]) != 0)
1729 			break;
1730 
1731 	if (i < n) {
1732 		for (i--; i >= 0; i--)
1733 			fdunused(fdp, fds[i]);
1734 		return (EMFILE);
1735 	}
1736 
1737 	return (0);
1738 }
1739 
1740 /*
1741  * Create a new open file structure and allocate a file descriptor for the
1742  * process that refers to it.  We add one reference to the file for the
1743  * descriptor table and one reference for resultfp. This is to prevent us
1744  * being preempted and the entry in the descriptor table closed after we
1745  * release the FILEDESC lock.
1746  */
1747 int
1748 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1749     struct filecaps *fcaps)
1750 {
1751 	struct file *fp;
1752 	int error, fd;
1753 
1754 	error = falloc_noinstall(td, &fp);
1755 	if (error)
1756 		return (error);		/* no reference held on error */
1757 
1758 	error = finstall(td, fp, &fd, flags, fcaps);
1759 	if (error) {
1760 		fdrop(fp, td);		/* one reference (fp only) */
1761 		return (error);
1762 	}
1763 
1764 	if (resultfp != NULL)
1765 		*resultfp = fp;		/* copy out result */
1766 	else
1767 		fdrop(fp, td);		/* release local reference */
1768 
1769 	if (resultfd != NULL)
1770 		*resultfd = fd;
1771 
1772 	return (0);
1773 }
1774 
1775 /*
1776  * Create a new open file structure without allocating a file descriptor.
1777  */
1778 int
1779 falloc_noinstall(struct thread *td, struct file **resultfp)
1780 {
1781 	struct file *fp;
1782 	int maxuserfiles = maxfiles - (maxfiles / 20);
1783 	int openfiles_new;
1784 	static struct timeval lastfail;
1785 	static int curfail;
1786 
1787 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1788 
1789 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1790 	if ((openfiles_new >= maxuserfiles &&
1791 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1792 	    openfiles_new >= maxfiles) {
1793 		atomic_subtract_int(&openfiles, 1);
1794 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1795 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1796 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1797 		}
1798 		return (ENFILE);
1799 	}
1800 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1801 	refcount_init(&fp->f_count, 1);
1802 	fp->f_cred = crhold(td->td_ucred);
1803 	fp->f_ops = &badfileops;
1804 	*resultfp = fp;
1805 	return (0);
1806 }
1807 
1808 /*
1809  * Install a file in a file descriptor table.
1810  */
1811 void
1812 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1813     struct filecaps *fcaps)
1814 {
1815 	struct filedescent *fde;
1816 
1817 	MPASS(fp != NULL);
1818 	if (fcaps != NULL)
1819 		filecaps_validate(fcaps, __func__);
1820 	FILEDESC_XLOCK_ASSERT(fdp);
1821 
1822 	fde = &fdp->fd_ofiles[fd];
1823 #ifdef CAPABILITIES
1824 	seq_write_begin(&fde->fde_seq);
1825 #endif
1826 	fde->fde_file = fp;
1827 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1828 	if (fcaps != NULL)
1829 		filecaps_move(fcaps, &fde->fde_caps);
1830 	else
1831 		filecaps_fill(&fde->fde_caps);
1832 #ifdef CAPABILITIES
1833 	seq_write_end(&fde->fde_seq);
1834 #endif
1835 }
1836 
1837 int
1838 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1839     struct filecaps *fcaps)
1840 {
1841 	struct filedesc *fdp = td->td_proc->p_fd;
1842 	int error;
1843 
1844 	MPASS(fd != NULL);
1845 
1846 	FILEDESC_XLOCK(fdp);
1847 	if ((error = fdalloc(td, 0, fd))) {
1848 		FILEDESC_XUNLOCK(fdp);
1849 		return (error);
1850 	}
1851 	fhold(fp);
1852 	_finstall(fdp, fp, *fd, flags, fcaps);
1853 	FILEDESC_XUNLOCK(fdp);
1854 	return (0);
1855 }
1856 
1857 /*
1858  * Build a new filedesc structure from another.
1859  * Copy the current, root, and jail root vnode references.
1860  *
1861  * If fdp is not NULL, return with it shared locked.
1862  */
1863 struct filedesc *
1864 fdinit(struct filedesc *fdp, bool prepfiles)
1865 {
1866 	struct filedesc0 *newfdp0;
1867 	struct filedesc *newfdp;
1868 
1869 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1870 	newfdp = &newfdp0->fd_fd;
1871 
1872 	/* Create the file descriptor table. */
1873 	FILEDESC_LOCK_INIT(newfdp);
1874 	refcount_init(&newfdp->fd_refcnt, 1);
1875 	refcount_init(&newfdp->fd_holdcnt, 1);
1876 	newfdp->fd_cmask = CMASK;
1877 	newfdp->fd_map = newfdp0->fd_dmap;
1878 	newfdp->fd_lastfile = -1;
1879 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1880 	newfdp->fd_files->fdt_nfiles = NDFILE;
1881 
1882 	if (fdp == NULL)
1883 		return (newfdp);
1884 
1885 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1886 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1887 
1888 	FILEDESC_SLOCK(fdp);
1889 	newfdp->fd_cdir = fdp->fd_cdir;
1890 	if (newfdp->fd_cdir)
1891 		vrefact(newfdp->fd_cdir);
1892 	newfdp->fd_rdir = fdp->fd_rdir;
1893 	if (newfdp->fd_rdir)
1894 		vrefact(newfdp->fd_rdir);
1895 	newfdp->fd_jdir = fdp->fd_jdir;
1896 	if (newfdp->fd_jdir)
1897 		vrefact(newfdp->fd_jdir);
1898 
1899 	if (!prepfiles) {
1900 		FILEDESC_SUNLOCK(fdp);
1901 	} else {
1902 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1903 			FILEDESC_SUNLOCK(fdp);
1904 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1905 			FILEDESC_SLOCK(fdp);
1906 		}
1907 	}
1908 
1909 	return (newfdp);
1910 }
1911 
1912 static struct filedesc *
1913 fdhold(struct proc *p)
1914 {
1915 	struct filedesc *fdp;
1916 
1917 	PROC_LOCK_ASSERT(p, MA_OWNED);
1918 	fdp = p->p_fd;
1919 	if (fdp != NULL)
1920 		refcount_acquire(&fdp->fd_holdcnt);
1921 	return (fdp);
1922 }
1923 
1924 static void
1925 fddrop(struct filedesc *fdp)
1926 {
1927 
1928 	if (fdp->fd_holdcnt > 1) {
1929 		if (refcount_release(&fdp->fd_holdcnt) == 0)
1930 			return;
1931 	}
1932 
1933 	FILEDESC_LOCK_DESTROY(fdp);
1934 	uma_zfree(filedesc0_zone, fdp);
1935 }
1936 
1937 /*
1938  * Share a filedesc structure.
1939  */
1940 struct filedesc *
1941 fdshare(struct filedesc *fdp)
1942 {
1943 
1944 	refcount_acquire(&fdp->fd_refcnt);
1945 	return (fdp);
1946 }
1947 
1948 /*
1949  * Unshare a filedesc structure, if necessary by making a copy
1950  */
1951 void
1952 fdunshare(struct thread *td)
1953 {
1954 	struct filedesc *tmp;
1955 	struct proc *p = td->td_proc;
1956 
1957 	if (p->p_fd->fd_refcnt == 1)
1958 		return;
1959 
1960 	tmp = fdcopy(p->p_fd);
1961 	fdescfree(td);
1962 	p->p_fd = tmp;
1963 }
1964 
1965 void
1966 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1967 {
1968 
1969 	fdescfree(td);
1970 	td->td_proc->p_fd = fdp;
1971 }
1972 
1973 /*
1974  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1975  * this is to ease callers, not catch errors.
1976  */
1977 struct filedesc *
1978 fdcopy(struct filedesc *fdp)
1979 {
1980 	struct filedesc *newfdp;
1981 	struct filedescent *nfde, *ofde;
1982 	int i;
1983 
1984 	MPASS(fdp != NULL);
1985 
1986 	newfdp = fdinit(fdp, true);
1987 	/* copy all passable descriptors (i.e. not kqueue) */
1988 	newfdp->fd_freefile = -1;
1989 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1990 		ofde = &fdp->fd_ofiles[i];
1991 		if (ofde->fde_file == NULL ||
1992 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1993 			if (newfdp->fd_freefile == -1)
1994 				newfdp->fd_freefile = i;
1995 			continue;
1996 		}
1997 		nfde = &newfdp->fd_ofiles[i];
1998 		*nfde = *ofde;
1999 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2000 		fhold(nfde->fde_file);
2001 		fdused_init(newfdp, i);
2002 		newfdp->fd_lastfile = i;
2003 	}
2004 	if (newfdp->fd_freefile == -1)
2005 		newfdp->fd_freefile = i;
2006 	newfdp->fd_cmask = fdp->fd_cmask;
2007 	FILEDESC_SUNLOCK(fdp);
2008 	return (newfdp);
2009 }
2010 
2011 /*
2012  * Copies a filedesc structure, while remapping all file descriptors
2013  * stored inside using a translation table.
2014  *
2015  * File descriptors are copied over to the new file descriptor table,
2016  * regardless of whether the close-on-exec flag is set.
2017  */
2018 int
2019 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2020     struct filedesc **ret)
2021 {
2022 	struct filedesc *newfdp;
2023 	struct filedescent *nfde, *ofde;
2024 	int error, i;
2025 
2026 	MPASS(fdp != NULL);
2027 
2028 	newfdp = fdinit(fdp, true);
2029 	if (nfds > fdp->fd_lastfile + 1) {
2030 		/* New table cannot be larger than the old one. */
2031 		error = E2BIG;
2032 		goto bad;
2033 	}
2034 	/* Copy all passable descriptors (i.e. not kqueue). */
2035 	newfdp->fd_freefile = nfds;
2036 	for (i = 0; i < nfds; ++i) {
2037 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2038 			/* File descriptor out of bounds. */
2039 			error = EBADF;
2040 			goto bad;
2041 		}
2042 		ofde = &fdp->fd_ofiles[fds[i]];
2043 		if (ofde->fde_file == NULL) {
2044 			/* Unused file descriptor. */
2045 			error = EBADF;
2046 			goto bad;
2047 		}
2048 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2049 			/* File descriptor cannot be passed. */
2050 			error = EINVAL;
2051 			goto bad;
2052 		}
2053 		nfde = &newfdp->fd_ofiles[i];
2054 		*nfde = *ofde;
2055 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2056 		fhold(nfde->fde_file);
2057 		fdused_init(newfdp, i);
2058 		newfdp->fd_lastfile = i;
2059 	}
2060 	newfdp->fd_cmask = fdp->fd_cmask;
2061 	FILEDESC_SUNLOCK(fdp);
2062 	*ret = newfdp;
2063 	return (0);
2064 bad:
2065 	FILEDESC_SUNLOCK(fdp);
2066 	fdescfree_remapped(newfdp);
2067 	return (error);
2068 }
2069 
2070 /*
2071  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2072  * one of processes using it exits) and the table used to be shared.
2073  */
2074 static void
2075 fdclearlocks(struct thread *td)
2076 {
2077 	struct filedesc *fdp;
2078 	struct filedesc_to_leader *fdtol;
2079 	struct flock lf;
2080 	struct file *fp;
2081 	struct proc *p;
2082 	struct vnode *vp;
2083 	int i;
2084 
2085 	p = td->td_proc;
2086 	fdp = p->p_fd;
2087 	fdtol = p->p_fdtol;
2088 	MPASS(fdtol != NULL);
2089 
2090 	FILEDESC_XLOCK(fdp);
2091 	KASSERT(fdtol->fdl_refcount > 0,
2092 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2093 	    fdtol->fdl_refcount));
2094 	if (fdtol->fdl_refcount == 1 &&
2095 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2096 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2097 			fp = fdp->fd_ofiles[i].fde_file;
2098 			if (fp == NULL || fp->f_type != DTYPE_VNODE)
2099 				continue;
2100 			fhold(fp);
2101 			FILEDESC_XUNLOCK(fdp);
2102 			lf.l_whence = SEEK_SET;
2103 			lf.l_start = 0;
2104 			lf.l_len = 0;
2105 			lf.l_type = F_UNLCK;
2106 			vp = fp->f_vnode;
2107 			(void) VOP_ADVLOCK(vp,
2108 			    (caddr_t)p->p_leader, F_UNLCK,
2109 			    &lf, F_POSIX);
2110 			FILEDESC_XLOCK(fdp);
2111 			fdrop(fp, td);
2112 		}
2113 	}
2114 retry:
2115 	if (fdtol->fdl_refcount == 1) {
2116 		if (fdp->fd_holdleaderscount > 0 &&
2117 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2118 			/*
2119 			 * close() or kern_dup() has cleared a reference
2120 			 * in a shared file descriptor table.
2121 			 */
2122 			fdp->fd_holdleaderswakeup = 1;
2123 			sx_sleep(&fdp->fd_holdleaderscount,
2124 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2125 			goto retry;
2126 		}
2127 		if (fdtol->fdl_holdcount > 0) {
2128 			/*
2129 			 * Ensure that fdtol->fdl_leader remains
2130 			 * valid in closef().
2131 			 */
2132 			fdtol->fdl_wakeup = 1;
2133 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2134 			    "fdlhold", 0);
2135 			goto retry;
2136 		}
2137 	}
2138 	fdtol->fdl_refcount--;
2139 	if (fdtol->fdl_refcount == 0 &&
2140 	    fdtol->fdl_holdcount == 0) {
2141 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2142 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2143 	} else
2144 		fdtol = NULL;
2145 	p->p_fdtol = NULL;
2146 	FILEDESC_XUNLOCK(fdp);
2147 	if (fdtol != NULL)
2148 		free(fdtol, M_FILEDESC_TO_LEADER);
2149 }
2150 
2151 /*
2152  * Release a filedesc structure.
2153  */
2154 static void
2155 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2156 {
2157 	struct filedesc0 *fdp0;
2158 	struct freetable *ft, *tft;
2159 	struct filedescent *fde;
2160 	struct file *fp;
2161 	int i;
2162 
2163 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2164 		fde = &fdp->fd_ofiles[i];
2165 		fp = fde->fde_file;
2166 		if (fp != NULL) {
2167 			fdefree_last(fde);
2168 			if (needclose)
2169 				(void) closef(fp, td);
2170 			else
2171 				fdrop(fp, td);
2172 		}
2173 	}
2174 
2175 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2176 		free(fdp->fd_map, M_FILEDESC);
2177 	if (fdp->fd_nfiles > NDFILE)
2178 		free(fdp->fd_files, M_FILEDESC);
2179 
2180 	fdp0 = (struct filedesc0 *)fdp;
2181 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2182 		free(ft->ft_table, M_FILEDESC);
2183 
2184 	fddrop(fdp);
2185 }
2186 
2187 void
2188 fdescfree(struct thread *td)
2189 {
2190 	struct proc *p;
2191 	struct filedesc *fdp;
2192 	struct vnode *cdir, *jdir, *rdir;
2193 
2194 	p = td->td_proc;
2195 	fdp = p->p_fd;
2196 	MPASS(fdp != NULL);
2197 
2198 #ifdef RACCT
2199 	if (racct_enable) {
2200 		PROC_LOCK(p);
2201 		racct_set(p, RACCT_NOFILE, 0);
2202 		PROC_UNLOCK(p);
2203 	}
2204 #endif
2205 
2206 	if (p->p_fdtol != NULL)
2207 		fdclearlocks(td);
2208 
2209 	PROC_LOCK(p);
2210 	p->p_fd = NULL;
2211 	PROC_UNLOCK(p);
2212 
2213 	if (refcount_release(&fdp->fd_refcnt) == 0)
2214 		return;
2215 
2216 	FILEDESC_XLOCK(fdp);
2217 	cdir = fdp->fd_cdir;
2218 	fdp->fd_cdir = NULL;
2219 	rdir = fdp->fd_rdir;
2220 	fdp->fd_rdir = NULL;
2221 	jdir = fdp->fd_jdir;
2222 	fdp->fd_jdir = NULL;
2223 	FILEDESC_XUNLOCK(fdp);
2224 
2225 	if (cdir != NULL)
2226 		vrele(cdir);
2227 	if (rdir != NULL)
2228 		vrele(rdir);
2229 	if (jdir != NULL)
2230 		vrele(jdir);
2231 
2232 	fdescfree_fds(td, fdp, 1);
2233 }
2234 
2235 void
2236 fdescfree_remapped(struct filedesc *fdp)
2237 {
2238 
2239 	if (fdp->fd_cdir != NULL)
2240 		vrele(fdp->fd_cdir);
2241 	if (fdp->fd_rdir != NULL)
2242 		vrele(fdp->fd_rdir);
2243 	if (fdp->fd_jdir != NULL)
2244 		vrele(fdp->fd_jdir);
2245 
2246 	fdescfree_fds(curthread, fdp, 0);
2247 }
2248 
2249 /*
2250  * For setugid programs, we don't want to people to use that setugidness
2251  * to generate error messages which write to a file which otherwise would
2252  * otherwise be off-limits to the process.  We check for filesystems where
2253  * the vnode can change out from under us after execve (like [lin]procfs).
2254  *
2255  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2256  * sufficient.  We also don't check for setugidness since we know we are.
2257  */
2258 static bool
2259 is_unsafe(struct file *fp)
2260 {
2261 	struct vnode *vp;
2262 
2263 	if (fp->f_type != DTYPE_VNODE)
2264 		return (false);
2265 
2266 	vp = fp->f_vnode;
2267 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2268 }
2269 
2270 /*
2271  * Make this setguid thing safe, if at all possible.
2272  */
2273 void
2274 fdsetugidsafety(struct thread *td)
2275 {
2276 	struct filedesc *fdp;
2277 	struct file *fp;
2278 	int i;
2279 
2280 	fdp = td->td_proc->p_fd;
2281 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2282 	MPASS(fdp->fd_nfiles >= 3);
2283 	for (i = 0; i <= 2; i++) {
2284 		fp = fdp->fd_ofiles[i].fde_file;
2285 		if (fp != NULL && is_unsafe(fp)) {
2286 			FILEDESC_XLOCK(fdp);
2287 			knote_fdclose(td, i);
2288 			/*
2289 			 * NULL-out descriptor prior to close to avoid
2290 			 * a race while close blocks.
2291 			 */
2292 			fdfree(fdp, i);
2293 			FILEDESC_XUNLOCK(fdp);
2294 			(void) closef(fp, td);
2295 		}
2296 	}
2297 }
2298 
2299 /*
2300  * If a specific file object occupies a specific file descriptor, close the
2301  * file descriptor entry and drop a reference on the file object.  This is a
2302  * convenience function to handle a subsequent error in a function that calls
2303  * falloc() that handles the race that another thread might have closed the
2304  * file descriptor out from under the thread creating the file object.
2305  */
2306 void
2307 fdclose(struct thread *td, struct file *fp, int idx)
2308 {
2309 	struct filedesc *fdp = td->td_proc->p_fd;
2310 
2311 	FILEDESC_XLOCK(fdp);
2312 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2313 		fdfree(fdp, idx);
2314 		FILEDESC_XUNLOCK(fdp);
2315 		fdrop(fp, td);
2316 	} else
2317 		FILEDESC_XUNLOCK(fdp);
2318 }
2319 
2320 /*
2321  * Close any files on exec?
2322  */
2323 void
2324 fdcloseexec(struct thread *td)
2325 {
2326 	struct filedesc *fdp;
2327 	struct filedescent *fde;
2328 	struct file *fp;
2329 	int i;
2330 
2331 	fdp = td->td_proc->p_fd;
2332 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2333 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2334 		fde = &fdp->fd_ofiles[i];
2335 		fp = fde->fde_file;
2336 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2337 		    (fde->fde_flags & UF_EXCLOSE))) {
2338 			FILEDESC_XLOCK(fdp);
2339 			fdfree(fdp, i);
2340 			(void) closefp(fdp, i, fp, td, 0);
2341 			FILEDESC_UNLOCK_ASSERT(fdp);
2342 		}
2343 	}
2344 }
2345 
2346 /*
2347  * It is unsafe for set[ug]id processes to be started with file
2348  * descriptors 0..2 closed, as these descriptors are given implicit
2349  * significance in the Standard C library.  fdcheckstd() will create a
2350  * descriptor referencing /dev/null for each of stdin, stdout, and
2351  * stderr that is not already open.
2352  */
2353 int
2354 fdcheckstd(struct thread *td)
2355 {
2356 	struct filedesc *fdp;
2357 	register_t save;
2358 	int i, error, devnull;
2359 
2360 	fdp = td->td_proc->p_fd;
2361 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2362 	MPASS(fdp->fd_nfiles >= 3);
2363 	devnull = -1;
2364 	for (i = 0; i <= 2; i++) {
2365 		if (fdp->fd_ofiles[i].fde_file != NULL)
2366 			continue;
2367 
2368 		save = td->td_retval[0];
2369 		if (devnull != -1) {
2370 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2371 		} else {
2372 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2373 			    UIO_SYSSPACE, O_RDWR, 0);
2374 			if (error == 0) {
2375 				devnull = td->td_retval[0];
2376 				KASSERT(devnull == i, ("we didn't get our fd"));
2377 			}
2378 		}
2379 		td->td_retval[0] = save;
2380 		if (error != 0)
2381 			return (error);
2382 	}
2383 	return (0);
2384 }
2385 
2386 /*
2387  * Internal form of close.  Decrement reference count on file structure.
2388  * Note: td may be NULL when closing a file that was being passed in a
2389  * message.
2390  *
2391  * XXXRW: Giant is not required for the caller, but often will be held; this
2392  * makes it moderately likely the Giant will be recursed in the VFS case.
2393  */
2394 int
2395 closef(struct file *fp, struct thread *td)
2396 {
2397 	struct vnode *vp;
2398 	struct flock lf;
2399 	struct filedesc_to_leader *fdtol;
2400 	struct filedesc *fdp;
2401 
2402 	/*
2403 	 * POSIX record locking dictates that any close releases ALL
2404 	 * locks owned by this process.  This is handled by setting
2405 	 * a flag in the unlock to free ONLY locks obeying POSIX
2406 	 * semantics, and not to free BSD-style file locks.
2407 	 * If the descriptor was in a message, POSIX-style locks
2408 	 * aren't passed with the descriptor, and the thread pointer
2409 	 * will be NULL.  Callers should be careful only to pass a
2410 	 * NULL thread pointer when there really is no owning
2411 	 * context that might have locks, or the locks will be
2412 	 * leaked.
2413 	 */
2414 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2415 		vp = fp->f_vnode;
2416 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2417 			lf.l_whence = SEEK_SET;
2418 			lf.l_start = 0;
2419 			lf.l_len = 0;
2420 			lf.l_type = F_UNLCK;
2421 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2422 			    F_UNLCK, &lf, F_POSIX);
2423 		}
2424 		fdtol = td->td_proc->p_fdtol;
2425 		if (fdtol != NULL) {
2426 			/*
2427 			 * Handle special case where file descriptor table is
2428 			 * shared between multiple process leaders.
2429 			 */
2430 			fdp = td->td_proc->p_fd;
2431 			FILEDESC_XLOCK(fdp);
2432 			for (fdtol = fdtol->fdl_next;
2433 			    fdtol != td->td_proc->p_fdtol;
2434 			    fdtol = fdtol->fdl_next) {
2435 				if ((fdtol->fdl_leader->p_flag &
2436 				    P_ADVLOCK) == 0)
2437 					continue;
2438 				fdtol->fdl_holdcount++;
2439 				FILEDESC_XUNLOCK(fdp);
2440 				lf.l_whence = SEEK_SET;
2441 				lf.l_start = 0;
2442 				lf.l_len = 0;
2443 				lf.l_type = F_UNLCK;
2444 				vp = fp->f_vnode;
2445 				(void) VOP_ADVLOCK(vp,
2446 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2447 				    F_POSIX);
2448 				FILEDESC_XLOCK(fdp);
2449 				fdtol->fdl_holdcount--;
2450 				if (fdtol->fdl_holdcount == 0 &&
2451 				    fdtol->fdl_wakeup != 0) {
2452 					fdtol->fdl_wakeup = 0;
2453 					wakeup(fdtol);
2454 				}
2455 			}
2456 			FILEDESC_XUNLOCK(fdp);
2457 		}
2458 	}
2459 	return (fdrop(fp, td));
2460 }
2461 
2462 /*
2463  * Initialize the file pointer with the specified properties.
2464  *
2465  * The ops are set with release semantics to be certain that the flags, type,
2466  * and data are visible when ops is.  This is to prevent ops methods from being
2467  * called with bad data.
2468  */
2469 void
2470 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2471 {
2472 	fp->f_data = data;
2473 	fp->f_flag = flag;
2474 	fp->f_type = type;
2475 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2476 }
2477 
2478 int
2479 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2480     struct file **fpp, struct filecaps *havecapsp)
2481 {
2482 	struct filedescent *fde;
2483 	int error;
2484 
2485 	FILEDESC_LOCK_ASSERT(fdp);
2486 
2487 	fde = fdeget_locked(fdp, fd);
2488 	if (fde == NULL) {
2489 		error = EBADF;
2490 		goto out;
2491 	}
2492 
2493 #ifdef CAPABILITIES
2494 	error = cap_check(cap_rights_fde(fde), needrightsp);
2495 	if (error != 0)
2496 		goto out;
2497 #endif
2498 
2499 	if (havecapsp != NULL)
2500 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2501 
2502 	*fpp = fde->fde_file;
2503 
2504 	error = 0;
2505 out:
2506 	return (error);
2507 }
2508 
2509 int
2510 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2511     struct file **fpp, struct filecaps *havecapsp)
2512 {
2513 	struct filedesc *fdp = td->td_proc->p_fd;
2514 	int error;
2515 #ifndef CAPABILITIES
2516 	error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2517 	if (error == 0 && havecapsp != NULL)
2518 		filecaps_fill(havecapsp);
2519 #else
2520 	struct file *fp;
2521 	seq_t seq;
2522 
2523 	for (;;) {
2524 		error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2525 		if (error != 0)
2526 			return (error);
2527 
2528 		if (havecapsp != NULL) {
2529 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2530 			    havecapsp, false)) {
2531 				fdrop(fp, td);
2532 				goto get_locked;
2533 			}
2534 		}
2535 
2536 		if (!fd_modified(fdp, fd, seq))
2537 			break;
2538 		fdrop(fp, td);
2539 	}
2540 
2541 	*fpp = fp;
2542 	return (0);
2543 
2544 get_locked:
2545 	FILEDESC_SLOCK(fdp);
2546 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2547 	if (error == 0)
2548 		fhold(*fpp);
2549 	FILEDESC_SUNLOCK(fdp);
2550 #endif
2551 	return (error);
2552 }
2553 
2554 int
2555 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2556     struct file **fpp, seq_t *seqp)
2557 {
2558 #ifdef CAPABILITIES
2559 	struct filedescent *fde;
2560 #endif
2561 	struct fdescenttbl *fdt;
2562 	struct file *fp;
2563 	u_int count;
2564 #ifdef CAPABILITIES
2565 	seq_t seq;
2566 	cap_rights_t haverights;
2567 	int error;
2568 #endif
2569 
2570 	fdt = fdp->fd_files;
2571 	if ((u_int)fd >= fdt->fdt_nfiles)
2572 		return (EBADF);
2573 	/*
2574 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2575 	 * never raising a refcount above 0.  To accomplish this we have
2576 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2577 	 * must be re-verified once we acquire a reference to be certain
2578 	 * that the identity is still correct and we did not lose a race
2579 	 * due to preemption.
2580 	 */
2581 	for (;;) {
2582 #ifdef CAPABILITIES
2583 		seq = seq_read(fd_seq(fdt, fd));
2584 		fde = &fdt->fdt_ofiles[fd];
2585 		haverights = *cap_rights_fde(fde);
2586 		fp = fde->fde_file;
2587 		if (!seq_consistent(fd_seq(fdt, fd), seq))
2588 			continue;
2589 #else
2590 		fp = fdt->fdt_ofiles[fd].fde_file;
2591 #endif
2592 		if (fp == NULL)
2593 			return (EBADF);
2594 #ifdef CAPABILITIES
2595 		error = cap_check(&haverights, needrightsp);
2596 		if (error != 0)
2597 			return (error);
2598 #endif
2599 		count = fp->f_count;
2600 	retry:
2601 		if (count == 0) {
2602 			/*
2603 			 * Force a reload. Other thread could reallocate the
2604 			 * table before this fd was closed, so it possible that
2605 			 * there is a stale fp pointer in cached version.
2606 			 */
2607 			fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2608 			continue;
2609 		}
2610 		/*
2611 		 * Use an acquire barrier to force re-reading of fdt so it is
2612 		 * refreshed for verification.
2613 		 */
2614 		if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2615 			goto retry;
2616 		fdt = fdp->fd_files;
2617 #ifdef	CAPABILITIES
2618 		if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2619 #else
2620 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2621 #endif
2622 			break;
2623 		fdrop(fp, curthread);
2624 	}
2625 	*fpp = fp;
2626 	if (seqp != NULL) {
2627 #ifdef CAPABILITIES
2628 		*seqp = seq;
2629 #endif
2630 	}
2631 	return (0);
2632 }
2633 
2634 /*
2635  * Extract the file pointer associated with the specified descriptor for the
2636  * current user process.
2637  *
2638  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2639  * returned.
2640  *
2641  * File's rights will be checked against the capability rights mask.
2642  *
2643  * If an error occurred the non-zero error is returned and *fpp is set to
2644  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2645  * responsible for fdrop().
2646  */
2647 static __inline int
2648 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2649     cap_rights_t *needrightsp, seq_t *seqp)
2650 {
2651 	struct filedesc *fdp;
2652 	struct file *fp;
2653 	int error;
2654 
2655 	*fpp = NULL;
2656 	fdp = td->td_proc->p_fd;
2657 	error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2658 	if (error != 0)
2659 		return (error);
2660 	if (fp->f_ops == &badfileops) {
2661 		fdrop(fp, td);
2662 		return (EBADF);
2663 	}
2664 
2665 	/*
2666 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2667 	 */
2668 	error = 0;
2669 	switch (flags) {
2670 	case FREAD:
2671 	case FWRITE:
2672 		if ((fp->f_flag & flags) == 0)
2673 			error = EBADF;
2674 		break;
2675 	case FEXEC:
2676 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2677 		    ((fp->f_flag & FWRITE) != 0))
2678 			error = EBADF;
2679 		break;
2680 	case 0:
2681 		break;
2682 	default:
2683 		KASSERT(0, ("wrong flags"));
2684 	}
2685 
2686 	if (error != 0) {
2687 		fdrop(fp, td);
2688 		return (error);
2689 	}
2690 
2691 	*fpp = fp;
2692 	return (0);
2693 }
2694 
2695 int
2696 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2697 {
2698 
2699 	return (_fget(td, fd, fpp, 0, rightsp, NULL));
2700 }
2701 
2702 int
2703 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2704     struct file **fpp)
2705 {
2706 	int error;
2707 #ifndef CAPABILITIES
2708 	error = _fget(td, fd, fpp, 0, rightsp, NULL);
2709 	if (maxprotp != NULL)
2710 		*maxprotp = VM_PROT_ALL;
2711 #else
2712 	struct filedesc *fdp = td->td_proc->p_fd;
2713 	seq_t seq;
2714 
2715 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2716 	for (;;) {
2717 		error = _fget(td, fd, fpp, 0, rightsp, &seq);
2718 		if (error != 0)
2719 			return (error);
2720 		/*
2721 		 * If requested, convert capability rights to access flags.
2722 		 */
2723 		if (maxprotp != NULL)
2724 			*maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2725 		if (!fd_modified(fdp, fd, seq))
2726 			break;
2727 		fdrop(*fpp, td);
2728 	}
2729 #endif
2730 	return (error);
2731 }
2732 
2733 int
2734 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2735 {
2736 
2737 	return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2738 }
2739 
2740 int
2741 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2742 {
2743 
2744 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2745 }
2746 
2747 int
2748 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2749     struct file **fpp)
2750 {
2751 	struct filedesc *fdp = td->td_proc->p_fd;
2752 #ifndef CAPABILITIES
2753 	return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2754 #else
2755 	int error;
2756 	seq_t seq;
2757 
2758 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2759 	for (;;) {
2760 		error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2761 		if (error != 0)
2762 			return (error);
2763 		error = cap_fcntl_check(fdp, fd, needfcntl);
2764 		if (!fd_modified(fdp, fd, seq))
2765 			break;
2766 		fdrop(*fpp, td);
2767 	}
2768 	if (error != 0) {
2769 		fdrop(*fpp, td);
2770 		*fpp = NULL;
2771 	}
2772 	return (error);
2773 #endif
2774 }
2775 
2776 /*
2777  * Like fget() but loads the underlying vnode, or returns an error if the
2778  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2779  * never have VM objects.  The returned vnode will be vref()'d.
2780  *
2781  * XXX: what about the unused flags ?
2782  */
2783 static __inline int
2784 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2785     struct vnode **vpp)
2786 {
2787 	struct file *fp;
2788 	int error;
2789 
2790 	*vpp = NULL;
2791 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2792 	if (error != 0)
2793 		return (error);
2794 	if (fp->f_vnode == NULL) {
2795 		error = EINVAL;
2796 	} else {
2797 		*vpp = fp->f_vnode;
2798 		vrefact(*vpp);
2799 	}
2800 	fdrop(fp, td);
2801 
2802 	return (error);
2803 }
2804 
2805 int
2806 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2807 {
2808 
2809 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2810 }
2811 
2812 int
2813 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2814     struct filecaps *havecaps, struct vnode **vpp)
2815 {
2816 	struct filedesc *fdp;
2817 	struct filecaps caps;
2818 	struct file *fp;
2819 	int error;
2820 
2821 	fdp = td->td_proc->p_fd;
2822 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2823 	if (error != 0)
2824 		return (error);
2825 	if (fp->f_ops == &badfileops) {
2826 		error = EBADF;
2827 		goto out;
2828 	}
2829 	if (fp->f_vnode == NULL) {
2830 		error = EINVAL;
2831 		goto out;
2832 	}
2833 
2834 	*havecaps = caps;
2835 	*vpp = fp->f_vnode;
2836 	vrefact(*vpp);
2837 
2838 	return (0);
2839 out:
2840 	filecaps_free(&caps);
2841 	return (error);
2842 }
2843 
2844 int
2845 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2846 {
2847 
2848 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2849 }
2850 
2851 int
2852 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2853 {
2854 
2855 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2856 }
2857 
2858 #ifdef notyet
2859 int
2860 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2861     struct vnode **vpp)
2862 {
2863 
2864 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2865 }
2866 #endif
2867 
2868 /*
2869  * Handle the last reference to a file being closed.
2870  */
2871 int
2872 _fdrop(struct file *fp, struct thread *td)
2873 {
2874 	int error;
2875 
2876 	if (fp->f_count != 0)
2877 		panic("fdrop: count %d", fp->f_count);
2878 	error = fo_close(fp, td);
2879 	atomic_subtract_int(&openfiles, 1);
2880 	crfree(fp->f_cred);
2881 	free(fp->f_advice, M_FADVISE);
2882 	uma_zfree(file_zone, fp);
2883 
2884 	return (error);
2885 }
2886 
2887 /*
2888  * Apply an advisory lock on a file descriptor.
2889  *
2890  * Just attempt to get a record lock of the requested type on the entire file
2891  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2892  */
2893 #ifndef _SYS_SYSPROTO_H_
2894 struct flock_args {
2895 	int	fd;
2896 	int	how;
2897 };
2898 #endif
2899 /* ARGSUSED */
2900 int
2901 sys_flock(struct thread *td, struct flock_args *uap)
2902 {
2903 	struct file *fp;
2904 	struct vnode *vp;
2905 	struct flock lf;
2906 	cap_rights_t rights;
2907 	int error;
2908 
2909 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2910 	if (error != 0)
2911 		return (error);
2912 	if (fp->f_type != DTYPE_VNODE) {
2913 		fdrop(fp, td);
2914 		return (EOPNOTSUPP);
2915 	}
2916 
2917 	vp = fp->f_vnode;
2918 	lf.l_whence = SEEK_SET;
2919 	lf.l_start = 0;
2920 	lf.l_len = 0;
2921 	if (uap->how & LOCK_UN) {
2922 		lf.l_type = F_UNLCK;
2923 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2924 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2925 		goto done2;
2926 	}
2927 	if (uap->how & LOCK_EX)
2928 		lf.l_type = F_WRLCK;
2929 	else if (uap->how & LOCK_SH)
2930 		lf.l_type = F_RDLCK;
2931 	else {
2932 		error = EBADF;
2933 		goto done2;
2934 	}
2935 	atomic_set_int(&fp->f_flag, FHASLOCK);
2936 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2937 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2938 done2:
2939 	fdrop(fp, td);
2940 	return (error);
2941 }
2942 /*
2943  * Duplicate the specified descriptor to a free descriptor.
2944  */
2945 int
2946 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2947     int openerror, int *indxp)
2948 {
2949 	struct filedescent *newfde, *oldfde;
2950 	struct file *fp;
2951 	int error, indx;
2952 
2953 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2954 	    ("unexpected error %d in %s", openerror, __func__));
2955 
2956 	/*
2957 	 * If the to-be-dup'd fd number is greater than the allowed number
2958 	 * of file descriptors, or the fd to be dup'd has already been
2959 	 * closed, then reject.
2960 	 */
2961 	FILEDESC_XLOCK(fdp);
2962 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2963 		FILEDESC_XUNLOCK(fdp);
2964 		return (EBADF);
2965 	}
2966 
2967 	error = fdalloc(td, 0, &indx);
2968 	if (error != 0) {
2969 		FILEDESC_XUNLOCK(fdp);
2970 		return (error);
2971 	}
2972 
2973 	/*
2974 	 * There are two cases of interest here.
2975 	 *
2976 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2977 	 *
2978 	 * For ENXIO steal away the file structure from (dfd) and store it in
2979 	 * (indx).  (dfd) is effectively closed by this operation.
2980 	 */
2981 	switch (openerror) {
2982 	case ENODEV:
2983 		/*
2984 		 * Check that the mode the file is being opened for is a
2985 		 * subset of the mode of the existing descriptor.
2986 		 */
2987 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2988 			fdunused(fdp, indx);
2989 			FILEDESC_XUNLOCK(fdp);
2990 			return (EACCES);
2991 		}
2992 		fhold(fp);
2993 		newfde = &fdp->fd_ofiles[indx];
2994 		oldfde = &fdp->fd_ofiles[dfd];
2995 #ifdef CAPABILITIES
2996 		seq_write_begin(&newfde->fde_seq);
2997 #endif
2998 		memcpy(newfde, oldfde, fde_change_size);
2999 		filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
3000 #ifdef CAPABILITIES
3001 		seq_write_end(&newfde->fde_seq);
3002 #endif
3003 		break;
3004 	case ENXIO:
3005 		/*
3006 		 * Steal away the file pointer from dfd and stuff it into indx.
3007 		 */
3008 		newfde = &fdp->fd_ofiles[indx];
3009 		oldfde = &fdp->fd_ofiles[dfd];
3010 #ifdef CAPABILITIES
3011 		seq_write_begin(&newfde->fde_seq);
3012 #endif
3013 		memcpy(newfde, oldfde, fde_change_size);
3014 		oldfde->fde_file = NULL;
3015 		fdunused(fdp, dfd);
3016 #ifdef CAPABILITIES
3017 		seq_write_end(&newfde->fde_seq);
3018 #endif
3019 		break;
3020 	}
3021 	FILEDESC_XUNLOCK(fdp);
3022 	*indxp = indx;
3023 	return (0);
3024 }
3025 
3026 /*
3027  * This sysctl determines if we will allow a process to chroot(2) if it
3028  * has a directory open:
3029  *	0: disallowed for all processes.
3030  *	1: allowed for processes that were not already chroot(2)'ed.
3031  *	2: allowed for all processes.
3032  */
3033 
3034 static int chroot_allow_open_directories = 1;
3035 
3036 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3037     &chroot_allow_open_directories, 0,
3038     "Allow a process to chroot(2) if it has a directory open");
3039 
3040 /*
3041  * Helper function for raised chroot(2) security function:  Refuse if
3042  * any filedescriptors are open directories.
3043  */
3044 static int
3045 chroot_refuse_vdir_fds(struct filedesc *fdp)
3046 {
3047 	struct vnode *vp;
3048 	struct file *fp;
3049 	int fd;
3050 
3051 	FILEDESC_LOCK_ASSERT(fdp);
3052 
3053 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3054 		fp = fget_locked(fdp, fd);
3055 		if (fp == NULL)
3056 			continue;
3057 		if (fp->f_type == DTYPE_VNODE) {
3058 			vp = fp->f_vnode;
3059 			if (vp->v_type == VDIR)
3060 				return (EPERM);
3061 		}
3062 	}
3063 	return (0);
3064 }
3065 
3066 /*
3067  * Common routine for kern_chroot() and jail_attach().  The caller is
3068  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3069  * authorize this operation.
3070  */
3071 int
3072 pwd_chroot(struct thread *td, struct vnode *vp)
3073 {
3074 	struct filedesc *fdp;
3075 	struct vnode *oldvp;
3076 	int error;
3077 
3078 	fdp = td->td_proc->p_fd;
3079 	FILEDESC_XLOCK(fdp);
3080 	if (chroot_allow_open_directories == 0 ||
3081 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3082 		error = chroot_refuse_vdir_fds(fdp);
3083 		if (error != 0) {
3084 			FILEDESC_XUNLOCK(fdp);
3085 			return (error);
3086 		}
3087 	}
3088 	oldvp = fdp->fd_rdir;
3089 	vrefact(vp);
3090 	fdp->fd_rdir = vp;
3091 	if (fdp->fd_jdir == NULL) {
3092 		vrefact(vp);
3093 		fdp->fd_jdir = vp;
3094 	}
3095 	FILEDESC_XUNLOCK(fdp);
3096 	vrele(oldvp);
3097 	return (0);
3098 }
3099 
3100 void
3101 pwd_chdir(struct thread *td, struct vnode *vp)
3102 {
3103 	struct filedesc *fdp;
3104 	struct vnode *oldvp;
3105 
3106 	fdp = td->td_proc->p_fd;
3107 	FILEDESC_XLOCK(fdp);
3108 	VNASSERT(vp->v_usecount > 0, vp,
3109 	    ("chdir to a vnode with zero usecount"));
3110 	oldvp = fdp->fd_cdir;
3111 	fdp->fd_cdir = vp;
3112 	FILEDESC_XUNLOCK(fdp);
3113 	vrele(oldvp);
3114 }
3115 
3116 /*
3117  * Scan all active processes and prisons to see if any of them have a current
3118  * or root directory of `olddp'. If so, replace them with the new mount point.
3119  */
3120 void
3121 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3122 {
3123 	struct filedesc *fdp;
3124 	struct prison *pr;
3125 	struct proc *p;
3126 	int nrele;
3127 
3128 	if (vrefcnt(olddp) == 1)
3129 		return;
3130 	nrele = 0;
3131 	sx_slock(&allproc_lock);
3132 	FOREACH_PROC_IN_SYSTEM(p) {
3133 		PROC_LOCK(p);
3134 		fdp = fdhold(p);
3135 		PROC_UNLOCK(p);
3136 		if (fdp == NULL)
3137 			continue;
3138 		FILEDESC_XLOCK(fdp);
3139 		if (fdp->fd_cdir == olddp) {
3140 			vrefact(newdp);
3141 			fdp->fd_cdir = newdp;
3142 			nrele++;
3143 		}
3144 		if (fdp->fd_rdir == olddp) {
3145 			vrefact(newdp);
3146 			fdp->fd_rdir = newdp;
3147 			nrele++;
3148 		}
3149 		if (fdp->fd_jdir == olddp) {
3150 			vrefact(newdp);
3151 			fdp->fd_jdir = newdp;
3152 			nrele++;
3153 		}
3154 		FILEDESC_XUNLOCK(fdp);
3155 		fddrop(fdp);
3156 	}
3157 	sx_sunlock(&allproc_lock);
3158 	if (rootvnode == olddp) {
3159 		vrefact(newdp);
3160 		rootvnode = newdp;
3161 		nrele++;
3162 	}
3163 	mtx_lock(&prison0.pr_mtx);
3164 	if (prison0.pr_root == olddp) {
3165 		vrefact(newdp);
3166 		prison0.pr_root = newdp;
3167 		nrele++;
3168 	}
3169 	mtx_unlock(&prison0.pr_mtx);
3170 	sx_slock(&allprison_lock);
3171 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3172 		mtx_lock(&pr->pr_mtx);
3173 		if (pr->pr_root == olddp) {
3174 			vrefact(newdp);
3175 			pr->pr_root = newdp;
3176 			nrele++;
3177 		}
3178 		mtx_unlock(&pr->pr_mtx);
3179 	}
3180 	sx_sunlock(&allprison_lock);
3181 	while (nrele--)
3182 		vrele(olddp);
3183 }
3184 
3185 struct filedesc_to_leader *
3186 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3187 {
3188 	struct filedesc_to_leader *fdtol;
3189 
3190 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3191 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3192 	fdtol->fdl_refcount = 1;
3193 	fdtol->fdl_holdcount = 0;
3194 	fdtol->fdl_wakeup = 0;
3195 	fdtol->fdl_leader = leader;
3196 	if (old != NULL) {
3197 		FILEDESC_XLOCK(fdp);
3198 		fdtol->fdl_next = old->fdl_next;
3199 		fdtol->fdl_prev = old;
3200 		old->fdl_next = fdtol;
3201 		fdtol->fdl_next->fdl_prev = fdtol;
3202 		FILEDESC_XUNLOCK(fdp);
3203 	} else {
3204 		fdtol->fdl_next = fdtol;
3205 		fdtol->fdl_prev = fdtol;
3206 	}
3207 	return (fdtol);
3208 }
3209 
3210 static int
3211 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3212 {
3213 	struct filedesc *fdp;
3214 	int i, count, slots;
3215 
3216 	if (*(int *)arg1 != 0)
3217 		return (EINVAL);
3218 
3219 	fdp = curproc->p_fd;
3220 	count = 0;
3221 	FILEDESC_SLOCK(fdp);
3222 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3223 	for (i = 0; i < slots; i++)
3224 		count += bitcountl(fdp->fd_map[i]);
3225 	FILEDESC_SUNLOCK(fdp);
3226 
3227 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3228 }
3229 
3230 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3231     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3232     "Number of open file descriptors");
3233 
3234 /*
3235  * Get file structures globally.
3236  */
3237 static int
3238 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3239 {
3240 	struct xfile xf;
3241 	struct filedesc *fdp;
3242 	struct file *fp;
3243 	struct proc *p;
3244 	int error, n;
3245 
3246 	error = sysctl_wire_old_buffer(req, 0);
3247 	if (error != 0)
3248 		return (error);
3249 	if (req->oldptr == NULL) {
3250 		n = 0;
3251 		sx_slock(&allproc_lock);
3252 		FOREACH_PROC_IN_SYSTEM(p) {
3253 			PROC_LOCK(p);
3254 			if (p->p_state == PRS_NEW) {
3255 				PROC_UNLOCK(p);
3256 				continue;
3257 			}
3258 			fdp = fdhold(p);
3259 			PROC_UNLOCK(p);
3260 			if (fdp == NULL)
3261 				continue;
3262 			/* overestimates sparse tables. */
3263 			if (fdp->fd_lastfile > 0)
3264 				n += fdp->fd_lastfile;
3265 			fddrop(fdp);
3266 		}
3267 		sx_sunlock(&allproc_lock);
3268 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3269 	}
3270 	error = 0;
3271 	bzero(&xf, sizeof(xf));
3272 	xf.xf_size = sizeof(xf);
3273 	sx_slock(&allproc_lock);
3274 	FOREACH_PROC_IN_SYSTEM(p) {
3275 		PROC_LOCK(p);
3276 		if (p->p_state == PRS_NEW) {
3277 			PROC_UNLOCK(p);
3278 			continue;
3279 		}
3280 		if (p_cansee(req->td, p) != 0) {
3281 			PROC_UNLOCK(p);
3282 			continue;
3283 		}
3284 		xf.xf_pid = p->p_pid;
3285 		xf.xf_uid = p->p_ucred->cr_uid;
3286 		fdp = fdhold(p);
3287 		PROC_UNLOCK(p);
3288 		if (fdp == NULL)
3289 			continue;
3290 		FILEDESC_SLOCK(fdp);
3291 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3292 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3293 				continue;
3294 			xf.xf_fd = n;
3295 			xf.xf_file = fp;
3296 			xf.xf_data = fp->f_data;
3297 			xf.xf_vnode = fp->f_vnode;
3298 			xf.xf_type = fp->f_type;
3299 			xf.xf_count = fp->f_count;
3300 			xf.xf_msgcount = 0;
3301 			xf.xf_offset = foffset_get(fp);
3302 			xf.xf_flag = fp->f_flag;
3303 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3304 			if (error)
3305 				break;
3306 		}
3307 		FILEDESC_SUNLOCK(fdp);
3308 		fddrop(fdp);
3309 		if (error)
3310 			break;
3311 	}
3312 	sx_sunlock(&allproc_lock);
3313 	return (error);
3314 }
3315 
3316 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3317     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3318 
3319 #ifdef KINFO_FILE_SIZE
3320 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3321 #endif
3322 
3323 static int
3324 xlate_fflags(int fflags)
3325 {
3326 	static const struct {
3327 		int	fflag;
3328 		int	kf_fflag;
3329 	} fflags_table[] = {
3330 		{ FAPPEND, KF_FLAG_APPEND },
3331 		{ FASYNC, KF_FLAG_ASYNC },
3332 		{ FFSYNC, KF_FLAG_FSYNC },
3333 		{ FHASLOCK, KF_FLAG_HASLOCK },
3334 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3335 		{ FREAD, KF_FLAG_READ },
3336 		{ FWRITE, KF_FLAG_WRITE },
3337 		{ O_CREAT, KF_FLAG_CREAT },
3338 		{ O_DIRECT, KF_FLAG_DIRECT },
3339 		{ O_EXCL, KF_FLAG_EXCL },
3340 		{ O_EXEC, KF_FLAG_EXEC },
3341 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3342 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3343 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3344 		{ O_TRUNC, KF_FLAG_TRUNC }
3345 	};
3346 	unsigned int i;
3347 	int kflags;
3348 
3349 	kflags = 0;
3350 	for (i = 0; i < nitems(fflags_table); i++)
3351 		if (fflags & fflags_table[i].fflag)
3352 			kflags |=  fflags_table[i].kf_fflag;
3353 	return (kflags);
3354 }
3355 
3356 /* Trim unused data from kf_path by truncating the structure size. */
3357 static void
3358 pack_kinfo(struct kinfo_file *kif)
3359 {
3360 
3361 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3362 	    strlen(kif->kf_path) + 1;
3363 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3364 }
3365 
3366 static void
3367 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3368     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3369 {
3370 	int error;
3371 
3372 	bzero(kif, sizeof(*kif));
3373 
3374 	/* Set a default type to allow for empty fill_kinfo() methods. */
3375 	kif->kf_type = KF_TYPE_UNKNOWN;
3376 	kif->kf_flags = xlate_fflags(fp->f_flag);
3377 	if (rightsp != NULL)
3378 		kif->kf_cap_rights = *rightsp;
3379 	else
3380 		cap_rights_init(&kif->kf_cap_rights);
3381 	kif->kf_fd = fd;
3382 	kif->kf_ref_count = fp->f_count;
3383 	kif->kf_offset = foffset_get(fp);
3384 
3385 	/*
3386 	 * This may drop the filedesc lock, so the 'fp' cannot be
3387 	 * accessed after this call.
3388 	 */
3389 	error = fo_fill_kinfo(fp, kif, fdp);
3390 	if (error == 0)
3391 		kif->kf_status |= KF_ATTR_VALID;
3392 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3393 		pack_kinfo(kif);
3394 	else
3395 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3396 }
3397 
3398 static void
3399 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3400     struct kinfo_file *kif, int flags)
3401 {
3402 	int error;
3403 
3404 	bzero(kif, sizeof(*kif));
3405 
3406 	kif->kf_type = KF_TYPE_VNODE;
3407 	error = vn_fill_kinfo_vnode(vp, kif);
3408 	if (error == 0)
3409 		kif->kf_status |= KF_ATTR_VALID;
3410 	kif->kf_flags = xlate_fflags(fflags);
3411 	cap_rights_init(&kif->kf_cap_rights);
3412 	kif->kf_fd = fd;
3413 	kif->kf_ref_count = -1;
3414 	kif->kf_offset = -1;
3415 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3416 		pack_kinfo(kif);
3417 	else
3418 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3419 	vrele(vp);
3420 }
3421 
3422 struct export_fd_buf {
3423 	struct filedesc		*fdp;
3424 	struct sbuf 		*sb;
3425 	ssize_t			remainder;
3426 	struct kinfo_file	kif;
3427 	int			flags;
3428 };
3429 
3430 static int
3431 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3432 {
3433 	struct kinfo_file *kif;
3434 
3435 	kif = &efbuf->kif;
3436 	if (efbuf->remainder != -1) {
3437 		if (efbuf->remainder < kif->kf_structsize) {
3438 			/* Terminate export. */
3439 			efbuf->remainder = 0;
3440 			return (0);
3441 		}
3442 		efbuf->remainder -= kif->kf_structsize;
3443 	}
3444 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3445 }
3446 
3447 static int
3448 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3449     struct export_fd_buf *efbuf)
3450 {
3451 	int error;
3452 
3453 	if (efbuf->remainder == 0)
3454 		return (0);
3455 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3456 	    efbuf->flags);
3457 	FILEDESC_SUNLOCK(efbuf->fdp);
3458 	error = export_kinfo_to_sb(efbuf);
3459 	FILEDESC_SLOCK(efbuf->fdp);
3460 	return (error);
3461 }
3462 
3463 static int
3464 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3465     struct export_fd_buf *efbuf)
3466 {
3467 	int error;
3468 
3469 	if (efbuf->remainder == 0)
3470 		return (0);
3471 	if (efbuf->fdp != NULL)
3472 		FILEDESC_SUNLOCK(efbuf->fdp);
3473 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3474 	error = export_kinfo_to_sb(efbuf);
3475 	if (efbuf->fdp != NULL)
3476 		FILEDESC_SLOCK(efbuf->fdp);
3477 	return (error);
3478 }
3479 
3480 /*
3481  * Store a process file descriptor information to sbuf.
3482  *
3483  * Takes a locked proc as argument, and returns with the proc unlocked.
3484  */
3485 int
3486 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3487     int flags)
3488 {
3489 	struct file *fp;
3490 	struct filedesc *fdp;
3491 	struct export_fd_buf *efbuf;
3492 	struct vnode *cttyvp, *textvp, *tracevp;
3493 	int error, i;
3494 	cap_rights_t rights;
3495 
3496 	PROC_LOCK_ASSERT(p, MA_OWNED);
3497 
3498 	/* ktrace vnode */
3499 	tracevp = p->p_tracevp;
3500 	if (tracevp != NULL)
3501 		vrefact(tracevp);
3502 	/* text vnode */
3503 	textvp = p->p_textvp;
3504 	if (textvp != NULL)
3505 		vrefact(textvp);
3506 	/* Controlling tty. */
3507 	cttyvp = NULL;
3508 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3509 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3510 		if (cttyvp != NULL)
3511 			vrefact(cttyvp);
3512 	}
3513 	fdp = fdhold(p);
3514 	PROC_UNLOCK(p);
3515 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3516 	efbuf->fdp = NULL;
3517 	efbuf->sb = sb;
3518 	efbuf->remainder = maxlen;
3519 	efbuf->flags = flags;
3520 	if (tracevp != NULL)
3521 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3522 		    efbuf);
3523 	if (textvp != NULL)
3524 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3525 	if (cttyvp != NULL)
3526 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3527 		    efbuf);
3528 	error = 0;
3529 	if (fdp == NULL)
3530 		goto fail;
3531 	efbuf->fdp = fdp;
3532 	FILEDESC_SLOCK(fdp);
3533 	/* working directory */
3534 	if (fdp->fd_cdir != NULL) {
3535 		vrefact(fdp->fd_cdir);
3536 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3537 	}
3538 	/* root directory */
3539 	if (fdp->fd_rdir != NULL) {
3540 		vrefact(fdp->fd_rdir);
3541 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3542 	}
3543 	/* jail directory */
3544 	if (fdp->fd_jdir != NULL) {
3545 		vrefact(fdp->fd_jdir);
3546 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3547 	}
3548 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3549 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3550 			continue;
3551 #ifdef CAPABILITIES
3552 		rights = *cap_rights(fdp, i);
3553 #else /* !CAPABILITIES */
3554 		cap_rights_init(&rights);
3555 #endif
3556 		/*
3557 		 * Create sysctl entry.  It is OK to drop the filedesc
3558 		 * lock inside of export_file_to_sb() as we will
3559 		 * re-validate and re-evaluate its properties when the
3560 		 * loop continues.
3561 		 */
3562 		error = export_file_to_sb(fp, i, &rights, efbuf);
3563 		if (error != 0 || efbuf->remainder == 0)
3564 			break;
3565 	}
3566 	FILEDESC_SUNLOCK(fdp);
3567 	fddrop(fdp);
3568 fail:
3569 	free(efbuf, M_TEMP);
3570 	return (error);
3571 }
3572 
3573 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3574 
3575 /*
3576  * Get per-process file descriptors for use by procstat(1), et al.
3577  */
3578 static int
3579 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3580 {
3581 	struct sbuf sb;
3582 	struct proc *p;
3583 	ssize_t maxlen;
3584 	int error, error2, *name;
3585 
3586 	name = (int *)arg1;
3587 
3588 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3589 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3590 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3591 	if (error != 0) {
3592 		sbuf_delete(&sb);
3593 		return (error);
3594 	}
3595 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3596 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3597 	    KERN_FILEDESC_PACK_KINFO);
3598 	error2 = sbuf_finish(&sb);
3599 	sbuf_delete(&sb);
3600 	return (error != 0 ? error : error2);
3601 }
3602 
3603 #ifdef COMPAT_FREEBSD7
3604 #ifdef KINFO_OFILE_SIZE
3605 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3606 #endif
3607 
3608 static void
3609 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3610 {
3611 
3612 	okif->kf_structsize = sizeof(*okif);
3613 	okif->kf_type = kif->kf_type;
3614 	okif->kf_fd = kif->kf_fd;
3615 	okif->kf_ref_count = kif->kf_ref_count;
3616 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3617 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3618 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3619 	okif->kf_offset = kif->kf_offset;
3620 	if (kif->kf_type == KF_TYPE_VNODE)
3621 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3622 	else
3623 		okif->kf_vnode_type = KF_VTYPE_VNON;
3624 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3625 	if (kif->kf_type == KF_TYPE_SOCKET) {
3626 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3627 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3628 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3629 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3630 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3631 	} else {
3632 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3633 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3634 	}
3635 }
3636 
3637 static int
3638 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3639     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3640 {
3641 	int error;
3642 
3643 	vrefact(vp);
3644 	FILEDESC_SUNLOCK(fdp);
3645 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3646 	kinfo_to_okinfo(kif, okif);
3647 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3648 	FILEDESC_SLOCK(fdp);
3649 	return (error);
3650 }
3651 
3652 /*
3653  * Get per-process file descriptors for use by procstat(1), et al.
3654  */
3655 static int
3656 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3657 {
3658 	struct kinfo_ofile *okif;
3659 	struct kinfo_file *kif;
3660 	struct filedesc *fdp;
3661 	int error, i, *name;
3662 	struct file *fp;
3663 	struct proc *p;
3664 
3665 	name = (int *)arg1;
3666 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3667 	if (error != 0)
3668 		return (error);
3669 	fdp = fdhold(p);
3670 	PROC_UNLOCK(p);
3671 	if (fdp == NULL)
3672 		return (ENOENT);
3673 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3674 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3675 	FILEDESC_SLOCK(fdp);
3676 	if (fdp->fd_cdir != NULL)
3677 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3678 		    okif, fdp, req);
3679 	if (fdp->fd_rdir != NULL)
3680 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3681 		    okif, fdp, req);
3682 	if (fdp->fd_jdir != NULL)
3683 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3684 		    okif, fdp, req);
3685 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3686 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3687 			continue;
3688 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3689 		    KERN_FILEDESC_PACK_KINFO);
3690 		FILEDESC_SUNLOCK(fdp);
3691 		kinfo_to_okinfo(kif, okif);
3692 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3693 		FILEDESC_SLOCK(fdp);
3694 		if (error)
3695 			break;
3696 	}
3697 	FILEDESC_SUNLOCK(fdp);
3698 	fddrop(fdp);
3699 	free(kif, M_TEMP);
3700 	free(okif, M_TEMP);
3701 	return (0);
3702 }
3703 
3704 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3705     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3706     "Process ofiledesc entries");
3707 #endif	/* COMPAT_FREEBSD7 */
3708 
3709 int
3710 vntype_to_kinfo(int vtype)
3711 {
3712 	struct {
3713 		int	vtype;
3714 		int	kf_vtype;
3715 	} vtypes_table[] = {
3716 		{ VBAD, KF_VTYPE_VBAD },
3717 		{ VBLK, KF_VTYPE_VBLK },
3718 		{ VCHR, KF_VTYPE_VCHR },
3719 		{ VDIR, KF_VTYPE_VDIR },
3720 		{ VFIFO, KF_VTYPE_VFIFO },
3721 		{ VLNK, KF_VTYPE_VLNK },
3722 		{ VNON, KF_VTYPE_VNON },
3723 		{ VREG, KF_VTYPE_VREG },
3724 		{ VSOCK, KF_VTYPE_VSOCK }
3725 	};
3726 	unsigned int i;
3727 
3728 	/*
3729 	 * Perform vtype translation.
3730 	 */
3731 	for (i = 0; i < nitems(vtypes_table); i++)
3732 		if (vtypes_table[i].vtype == vtype)
3733 			return (vtypes_table[i].kf_vtype);
3734 
3735 	return (KF_VTYPE_UNKNOWN);
3736 }
3737 
3738 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3739     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3740     "Process filedesc entries");
3741 
3742 /*
3743  * Store a process current working directory information to sbuf.
3744  *
3745  * Takes a locked proc as argument, and returns with the proc unlocked.
3746  */
3747 int
3748 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3749 {
3750 	struct filedesc *fdp;
3751 	struct export_fd_buf *efbuf;
3752 	int error;
3753 
3754 	PROC_LOCK_ASSERT(p, MA_OWNED);
3755 
3756 	fdp = fdhold(p);
3757 	PROC_UNLOCK(p);
3758 	if (fdp == NULL)
3759 		return (EINVAL);
3760 
3761 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3762 	efbuf->fdp = fdp;
3763 	efbuf->sb = sb;
3764 	efbuf->remainder = maxlen;
3765 
3766 	FILEDESC_SLOCK(fdp);
3767 	if (fdp->fd_cdir == NULL)
3768 		error = EINVAL;
3769 	else {
3770 		vrefact(fdp->fd_cdir);
3771 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3772 		    FREAD, efbuf);
3773 	}
3774 	FILEDESC_SUNLOCK(fdp);
3775 	fddrop(fdp);
3776 	free(efbuf, M_TEMP);
3777 	return (error);
3778 }
3779 
3780 /*
3781  * Get per-process current working directory.
3782  */
3783 static int
3784 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3785 {
3786 	struct sbuf sb;
3787 	struct proc *p;
3788 	ssize_t maxlen;
3789 	int error, error2, *name;
3790 
3791 	name = (int *)arg1;
3792 
3793 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3794 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3795 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3796 	if (error != 0) {
3797 		sbuf_delete(&sb);
3798 		return (error);
3799 	}
3800 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3801 	error = kern_proc_cwd_out(p, &sb, maxlen);
3802 	error2 = sbuf_finish(&sb);
3803 	sbuf_delete(&sb);
3804 	return (error != 0 ? error : error2);
3805 }
3806 
3807 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3808     sysctl_kern_proc_cwd, "Process current working directory");
3809 
3810 #ifdef DDB
3811 /*
3812  * For the purposes of debugging, generate a human-readable string for the
3813  * file type.
3814  */
3815 static const char *
3816 file_type_to_name(short type)
3817 {
3818 
3819 	switch (type) {
3820 	case 0:
3821 		return ("zero");
3822 	case DTYPE_VNODE:
3823 		return ("vnod");
3824 	case DTYPE_SOCKET:
3825 		return ("sock");
3826 	case DTYPE_PIPE:
3827 		return ("pipe");
3828 	case DTYPE_FIFO:
3829 		return ("fifo");
3830 	case DTYPE_KQUEUE:
3831 		return ("kque");
3832 	case DTYPE_CRYPTO:
3833 		return ("crpt");
3834 	case DTYPE_MQUEUE:
3835 		return ("mque");
3836 	case DTYPE_SHM:
3837 		return ("shm");
3838 	case DTYPE_SEM:
3839 		return ("ksem");
3840 	default:
3841 		return ("unkn");
3842 	}
3843 }
3844 
3845 /*
3846  * For the purposes of debugging, identify a process (if any, perhaps one of
3847  * many) that references the passed file in its file descriptor array. Return
3848  * NULL if none.
3849  */
3850 static struct proc *
3851 file_to_first_proc(struct file *fp)
3852 {
3853 	struct filedesc *fdp;
3854 	struct proc *p;
3855 	int n;
3856 
3857 	FOREACH_PROC_IN_SYSTEM(p) {
3858 		if (p->p_state == PRS_NEW)
3859 			continue;
3860 		fdp = p->p_fd;
3861 		if (fdp == NULL)
3862 			continue;
3863 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3864 			if (fp == fdp->fd_ofiles[n].fde_file)
3865 				return (p);
3866 		}
3867 	}
3868 	return (NULL);
3869 }
3870 
3871 static void
3872 db_print_file(struct file *fp, int header)
3873 {
3874 	struct proc *p;
3875 
3876 	if (header)
3877 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3878 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3879 		    "MCount", "Vnode", "FPID", "FCmd");
3880 	p = file_to_first_proc(fp);
3881 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3882 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3883 	    0, fp->f_count, 0, fp->f_vnode,
3884 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3885 }
3886 
3887 DB_SHOW_COMMAND(file, db_show_file)
3888 {
3889 	struct file *fp;
3890 
3891 	if (!have_addr) {
3892 		db_printf("usage: show file <addr>\n");
3893 		return;
3894 	}
3895 	fp = (struct file *)addr;
3896 	db_print_file(fp, 1);
3897 }
3898 
3899 DB_SHOW_COMMAND(files, db_show_files)
3900 {
3901 	struct filedesc *fdp;
3902 	struct file *fp;
3903 	struct proc *p;
3904 	int header;
3905 	int n;
3906 
3907 	header = 1;
3908 	FOREACH_PROC_IN_SYSTEM(p) {
3909 		if (p->p_state == PRS_NEW)
3910 			continue;
3911 		if ((fdp = p->p_fd) == NULL)
3912 			continue;
3913 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
3914 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3915 				continue;
3916 			db_print_file(fp, header);
3917 			header = 0;
3918 		}
3919 	}
3920 }
3921 #endif
3922 
3923 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3924     &maxfilesperproc, 0, "Maximum files allowed open per process");
3925 
3926 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3927     &maxfiles, 0, "Maximum number of files");
3928 
3929 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3930     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3931 
3932 /* ARGSUSED*/
3933 static void
3934 filelistinit(void *dummy)
3935 {
3936 
3937 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3938 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3939 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3940 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3941 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3942 }
3943 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3944 
3945 /*-------------------------------------------------------------------*/
3946 
3947 static int
3948 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3949     int flags, struct thread *td)
3950 {
3951 
3952 	return (EBADF);
3953 }
3954 
3955 static int
3956 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3957     struct thread *td)
3958 {
3959 
3960 	return (EINVAL);
3961 }
3962 
3963 static int
3964 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3965     struct thread *td)
3966 {
3967 
3968 	return (EBADF);
3969 }
3970 
3971 static int
3972 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3973     struct thread *td)
3974 {
3975 
3976 	return (0);
3977 }
3978 
3979 static int
3980 badfo_kqfilter(struct file *fp, struct knote *kn)
3981 {
3982 
3983 	return (EBADF);
3984 }
3985 
3986 static int
3987 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3988     struct thread *td)
3989 {
3990 
3991 	return (EBADF);
3992 }
3993 
3994 static int
3995 badfo_close(struct file *fp, struct thread *td)
3996 {
3997 
3998 	return (0);
3999 }
4000 
4001 static int
4002 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4003     struct thread *td)
4004 {
4005 
4006 	return (EBADF);
4007 }
4008 
4009 static int
4010 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4011     struct thread *td)
4012 {
4013 
4014 	return (EBADF);
4015 }
4016 
4017 static int
4018 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4019     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4020     struct thread *td)
4021 {
4022 
4023 	return (EBADF);
4024 }
4025 
4026 static int
4027 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4028 {
4029 
4030 	return (0);
4031 }
4032 
4033 struct fileops badfileops = {
4034 	.fo_read = badfo_readwrite,
4035 	.fo_write = badfo_readwrite,
4036 	.fo_truncate = badfo_truncate,
4037 	.fo_ioctl = badfo_ioctl,
4038 	.fo_poll = badfo_poll,
4039 	.fo_kqfilter = badfo_kqfilter,
4040 	.fo_stat = badfo_stat,
4041 	.fo_close = badfo_close,
4042 	.fo_chmod = badfo_chmod,
4043 	.fo_chown = badfo_chown,
4044 	.fo_sendfile = badfo_sendfile,
4045 	.fo_fill_kinfo = badfo_fill_kinfo,
4046 };
4047 
4048 int
4049 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4050     int flags, struct thread *td)
4051 {
4052 
4053 	return (EOPNOTSUPP);
4054 }
4055 
4056 int
4057 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4058     struct thread *td)
4059 {
4060 
4061 	return (EINVAL);
4062 }
4063 
4064 int
4065 invfo_ioctl(struct file *fp, u_long com, void *data,
4066     struct ucred *active_cred, struct thread *td)
4067 {
4068 
4069 	return (ENOTTY);
4070 }
4071 
4072 int
4073 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4074     struct thread *td)
4075 {
4076 
4077 	return (poll_no_poll(events));
4078 }
4079 
4080 int
4081 invfo_kqfilter(struct file *fp, struct knote *kn)
4082 {
4083 
4084 	return (EINVAL);
4085 }
4086 
4087 int
4088 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4089     struct thread *td)
4090 {
4091 
4092 	return (EINVAL);
4093 }
4094 
4095 int
4096 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4097     struct thread *td)
4098 {
4099 
4100 	return (EINVAL);
4101 }
4102 
4103 int
4104 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4105     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4106     struct thread *td)
4107 {
4108 
4109 	return (EINVAL);
4110 }
4111 
4112 /*-------------------------------------------------------------------*/
4113 
4114 /*
4115  * File Descriptor pseudo-device driver (/dev/fd/).
4116  *
4117  * Opening minor device N dup()s the file (if any) connected to file
4118  * descriptor N belonging to the calling process.  Note that this driver
4119  * consists of only the ``open()'' routine, because all subsequent
4120  * references to this file will be direct to the other driver.
4121  *
4122  * XXX: we could give this one a cloning event handler if necessary.
4123  */
4124 
4125 /* ARGSUSED */
4126 static int
4127 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4128 {
4129 
4130 	/*
4131 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4132 	 * the file descriptor being sought for duplication. The error
4133 	 * return ensures that the vnode for this device will be released
4134 	 * by vn_open. Open will detect this special error and take the
4135 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4136 	 * will simply report the error.
4137 	 */
4138 	td->td_dupfd = dev2unit(dev);
4139 	return (ENODEV);
4140 }
4141 
4142 static struct cdevsw fildesc_cdevsw = {
4143 	.d_version =	D_VERSION,
4144 	.d_open =	fdopen,
4145 	.d_name =	"FD",
4146 };
4147 
4148 static void
4149 fildesc_drvinit(void *unused)
4150 {
4151 	struct cdev *dev;
4152 
4153 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4154 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4155 	make_dev_alias(dev, "stdin");
4156 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4157 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4158 	make_dev_alias(dev, "stdout");
4159 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4160 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4161 	make_dev_alias(dev, "stderr");
4162 }
4163 
4164 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4165