xref: /freebsd/sys/kern/kern_descrip.c (revision a8675d927bbde29d5a4dc6efddf2f0dc6d6d6983)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97     "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
100 
101 MALLOC_DECLARE(M_FADVISE);
102 
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 static __read_mostly uma_zone_t pwd_zone;
106 static __read_mostly smr_t pwd_smr;
107 
108 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
109 		    struct thread *td, int holdleaders);
110 static int	fd_first_free(struct filedesc *fdp, int low, int size);
111 static int	fd_last_used(struct filedesc *fdp, int size);
112 static void	fdgrowtable(struct filedesc *fdp, int nfd);
113 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
114 static void	fdunused(struct filedesc *fdp, int fd);
115 static void	fdused(struct filedesc *fdp, int fd);
116 static int	getmaxfd(struct thread *td);
117 static u_long	*filecaps_copy_prep(const struct filecaps *src);
118 static void	filecaps_copy_finish(const struct filecaps *src,
119 		    struct filecaps *dst, u_long *ioctls);
120 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
121 static void	filecaps_free_finish(u_long *ioctls);
122 
123 static struct pwd *pwd_alloc(void);
124 
125 /*
126  * Each process has:
127  *
128  * - An array of open file descriptors (fd_ofiles)
129  * - An array of file flags (fd_ofileflags)
130  * - A bitmap recording which descriptors are in use (fd_map)
131  *
132  * A process starts out with NDFILE descriptors.  The value of NDFILE has
133  * been selected based the historical limit of 20 open files, and an
134  * assumption that the majority of processes, especially short-lived
135  * processes like shells, will never need more.
136  *
137  * If this initial allocation is exhausted, a larger descriptor table and
138  * map are allocated dynamically, and the pointers in the process's struct
139  * filedesc are updated to point to those.  This is repeated every time
140  * the process runs out of file descriptors (provided it hasn't hit its
141  * resource limit).
142  *
143  * Since threads may hold references to individual descriptor table
144  * entries, the tables are never freed.  Instead, they are placed on a
145  * linked list and freed only when the struct filedesc is released.
146  */
147 #define NDFILE		20
148 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
149 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
150 #define NDSLOT(x)	((x) / NDENTRIES)
151 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
152 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
153 
154 /*
155  * SLIST entry used to keep track of ofiles which must be reclaimed when
156  * the process exits.
157  */
158 struct freetable {
159 	struct fdescenttbl *ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Initial allocation: a filedesc structure + the head of SLIST used to
165  * keep track of old ofiles + enough space for NDFILE descriptors.
166  */
167 
168 struct fdescenttbl0 {
169 	int	fdt_nfiles;
170 	struct	filedescent fdt_ofiles[NDFILE];
171 };
172 
173 struct filedesc0 {
174 	struct filedesc fd_fd;
175 	SLIST_HEAD(, freetable) fd_free;
176 	struct	fdescenttbl0 fd_dfiles;
177 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
178 };
179 
180 /*
181  * Descriptor management.
182  */
183 static int __exclusive_cache_line openfiles; /* actual number of open files */
184 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
185 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
186 
187 /*
188  * If low >= size, just return low. Otherwise find the first zero bit in the
189  * given bitmap, starting at low and not exceeding size - 1. Return size if
190  * not found.
191  */
192 static int
193 fd_first_free(struct filedesc *fdp, int low, int size)
194 {
195 	NDSLOTTYPE *map = fdp->fd_map;
196 	NDSLOTTYPE mask;
197 	int off, maxoff;
198 
199 	if (low >= size)
200 		return (low);
201 
202 	off = NDSLOT(low);
203 	if (low % NDENTRIES) {
204 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
205 		if ((mask &= ~map[off]) != 0UL)
206 			return (off * NDENTRIES + ffsl(mask) - 1);
207 		++off;
208 	}
209 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
210 		if (map[off] != ~0UL)
211 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
212 	return (size);
213 }
214 
215 /*
216  * Find the highest non-zero bit in the given bitmap, starting at 0 and
217  * not exceeding size - 1. Return -1 if not found.
218  */
219 static int
220 fd_last_used(struct filedesc *fdp, int size)
221 {
222 	NDSLOTTYPE *map = fdp->fd_map;
223 	NDSLOTTYPE mask;
224 	int off, minoff;
225 
226 	off = NDSLOT(size);
227 	if (size % NDENTRIES) {
228 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
229 		if ((mask &= map[off]) != 0)
230 			return (off * NDENTRIES + flsl(mask) - 1);
231 		--off;
232 	}
233 	for (minoff = NDSLOT(0); off >= minoff; --off)
234 		if (map[off] != 0)
235 			return (off * NDENTRIES + flsl(map[off]) - 1);
236 	return (-1);
237 }
238 
239 static int
240 fdisused(struct filedesc *fdp, int fd)
241 {
242 
243 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
244 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
245 
246 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
247 }
248 
249 /*
250  * Mark a file descriptor as used.
251  */
252 static void
253 fdused_init(struct filedesc *fdp, int fd)
254 {
255 
256 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
257 
258 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
259 }
260 
261 static void
262 fdused(struct filedesc *fdp, int fd)
263 {
264 
265 	FILEDESC_XLOCK_ASSERT(fdp);
266 
267 	fdused_init(fdp, fd);
268 	if (fd > fdp->fd_lastfile)
269 		fdp->fd_lastfile = fd;
270 	if (fd == fdp->fd_freefile)
271 		fdp->fd_freefile++;
272 }
273 
274 /*
275  * Mark a file descriptor as unused.
276  */
277 static void
278 fdunused(struct filedesc *fdp, int fd)
279 {
280 
281 	FILEDESC_XLOCK_ASSERT(fdp);
282 
283 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
284 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
285 	    ("fd=%d is still in use", fd));
286 
287 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
288 	if (fd < fdp->fd_freefile)
289 		fdp->fd_freefile = fd;
290 	if (fd == fdp->fd_lastfile)
291 		fdp->fd_lastfile = fd_last_used(fdp, fd);
292 }
293 
294 /*
295  * Free a file descriptor.
296  *
297  * Avoid some work if fdp is about to be destroyed.
298  */
299 static inline void
300 fdefree_last(struct filedescent *fde)
301 {
302 
303 	filecaps_free(&fde->fde_caps);
304 }
305 
306 static inline void
307 fdfree(struct filedesc *fdp, int fd)
308 {
309 	struct filedescent *fde;
310 
311 	fde = &fdp->fd_ofiles[fd];
312 #ifdef CAPABILITIES
313 	seqc_write_begin(&fde->fde_seqc);
314 #endif
315 	fde->fde_file = NULL;
316 #ifdef CAPABILITIES
317 	seqc_write_end(&fde->fde_seqc);
318 #endif
319 	fdefree_last(fde);
320 	fdunused(fdp, fd);
321 }
322 
323 /*
324  * System calls on descriptors.
325  */
326 #ifndef _SYS_SYSPROTO_H_
327 struct getdtablesize_args {
328 	int	dummy;
329 };
330 #endif
331 /* ARGSUSED */
332 int
333 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
334 {
335 #ifdef	RACCT
336 	uint64_t lim;
337 #endif
338 
339 	td->td_retval[0] = getmaxfd(td);
340 #ifdef	RACCT
341 	PROC_LOCK(td->td_proc);
342 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
343 	PROC_UNLOCK(td->td_proc);
344 	if (lim < td->td_retval[0])
345 		td->td_retval[0] = lim;
346 #endif
347 	return (0);
348 }
349 
350 /*
351  * Duplicate a file descriptor to a particular value.
352  *
353  * Note: keep in mind that a potential race condition exists when closing
354  * descriptors from a shared descriptor table (via rfork).
355  */
356 #ifndef _SYS_SYSPROTO_H_
357 struct dup2_args {
358 	u_int	from;
359 	u_int	to;
360 };
361 #endif
362 /* ARGSUSED */
363 int
364 sys_dup2(struct thread *td, struct dup2_args *uap)
365 {
366 
367 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
368 }
369 
370 /*
371  * Duplicate a file descriptor.
372  */
373 #ifndef _SYS_SYSPROTO_H_
374 struct dup_args {
375 	u_int	fd;
376 };
377 #endif
378 /* ARGSUSED */
379 int
380 sys_dup(struct thread *td, struct dup_args *uap)
381 {
382 
383 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
384 }
385 
386 /*
387  * The file control system call.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct fcntl_args {
391 	int	fd;
392 	int	cmd;
393 	long	arg;
394 };
395 #endif
396 /* ARGSUSED */
397 int
398 sys_fcntl(struct thread *td, struct fcntl_args *uap)
399 {
400 
401 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
402 }
403 
404 int
405 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
406 {
407 	struct flock fl;
408 	struct __oflock ofl;
409 	intptr_t arg1;
410 	int error, newcmd;
411 
412 	error = 0;
413 	newcmd = cmd;
414 	switch (cmd) {
415 	case F_OGETLK:
416 	case F_OSETLK:
417 	case F_OSETLKW:
418 		/*
419 		 * Convert old flock structure to new.
420 		 */
421 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
422 		fl.l_start = ofl.l_start;
423 		fl.l_len = ofl.l_len;
424 		fl.l_pid = ofl.l_pid;
425 		fl.l_type = ofl.l_type;
426 		fl.l_whence = ofl.l_whence;
427 		fl.l_sysid = 0;
428 
429 		switch (cmd) {
430 		case F_OGETLK:
431 			newcmd = F_GETLK;
432 			break;
433 		case F_OSETLK:
434 			newcmd = F_SETLK;
435 			break;
436 		case F_OSETLKW:
437 			newcmd = F_SETLKW;
438 			break;
439 		}
440 		arg1 = (intptr_t)&fl;
441 		break;
442 	case F_GETLK:
443 	case F_SETLK:
444 	case F_SETLKW:
445 	case F_SETLK_REMOTE:
446 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
447 		arg1 = (intptr_t)&fl;
448 		break;
449 	default:
450 		arg1 = arg;
451 		break;
452 	}
453 	if (error)
454 		return (error);
455 	error = kern_fcntl(td, fd, newcmd, arg1);
456 	if (error)
457 		return (error);
458 	if (cmd == F_OGETLK) {
459 		ofl.l_start = fl.l_start;
460 		ofl.l_len = fl.l_len;
461 		ofl.l_pid = fl.l_pid;
462 		ofl.l_type = fl.l_type;
463 		ofl.l_whence = fl.l_whence;
464 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
465 	} else if (cmd == F_GETLK) {
466 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
467 	}
468 	return (error);
469 }
470 
471 int
472 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
473 {
474 	struct filedesc *fdp;
475 	struct flock *flp;
476 	struct file *fp, *fp2;
477 	struct filedescent *fde;
478 	struct proc *p;
479 	struct vnode *vp;
480 	struct mount *mp;
481 	int error, flg, seals, tmp;
482 	uint64_t bsize;
483 	off_t foffset;
484 
485 	error = 0;
486 	flg = F_POSIX;
487 	p = td->td_proc;
488 	fdp = p->p_fd;
489 
490 	AUDIT_ARG_FD(cmd);
491 	AUDIT_ARG_CMD(cmd);
492 	switch (cmd) {
493 	case F_DUPFD:
494 		tmp = arg;
495 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
496 		break;
497 
498 	case F_DUPFD_CLOEXEC:
499 		tmp = arg;
500 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
501 		break;
502 
503 	case F_DUP2FD:
504 		tmp = arg;
505 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
506 		break;
507 
508 	case F_DUP2FD_CLOEXEC:
509 		tmp = arg;
510 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
511 		break;
512 
513 	case F_GETFD:
514 		error = EBADF;
515 		FILEDESC_SLOCK(fdp);
516 		fde = fdeget_locked(fdp, fd);
517 		if (fde != NULL) {
518 			td->td_retval[0] =
519 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
520 			error = 0;
521 		}
522 		FILEDESC_SUNLOCK(fdp);
523 		break;
524 
525 	case F_SETFD:
526 		error = EBADF;
527 		FILEDESC_XLOCK(fdp);
528 		fde = fdeget_locked(fdp, fd);
529 		if (fde != NULL) {
530 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
531 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
532 			error = 0;
533 		}
534 		FILEDESC_XUNLOCK(fdp);
535 		break;
536 
537 	case F_GETFL:
538 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
539 		if (error != 0)
540 			break;
541 		td->td_retval[0] = OFLAGS(fp->f_flag);
542 		fdrop(fp, td);
543 		break;
544 
545 	case F_SETFL:
546 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
547 		if (error != 0)
548 			break;
549 		do {
550 			tmp = flg = fp->f_flag;
551 			tmp &= ~FCNTLFLAGS;
552 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
553 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
554 		tmp = fp->f_flag & FNONBLOCK;
555 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
556 		if (error != 0) {
557 			fdrop(fp, td);
558 			break;
559 		}
560 		tmp = fp->f_flag & FASYNC;
561 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
562 		if (error == 0) {
563 			fdrop(fp, td);
564 			break;
565 		}
566 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
567 		tmp = 0;
568 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
569 		fdrop(fp, td);
570 		break;
571 
572 	case F_GETOWN:
573 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
574 		if (error != 0)
575 			break;
576 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
577 		if (error == 0)
578 			td->td_retval[0] = tmp;
579 		fdrop(fp, td);
580 		break;
581 
582 	case F_SETOWN:
583 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		tmp = arg;
587 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
588 		fdrop(fp, td);
589 		break;
590 
591 	case F_SETLK_REMOTE:
592 		error = priv_check(td, PRIV_NFS_LOCKD);
593 		if (error != 0)
594 			return (error);
595 		flg = F_REMOTE;
596 		goto do_setlk;
597 
598 	case F_SETLKW:
599 		flg |= F_WAIT;
600 		/* FALLTHROUGH F_SETLK */
601 
602 	case F_SETLK:
603 	do_setlk:
604 		flp = (struct flock *)arg;
605 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
606 			error = EINVAL;
607 			break;
608 		}
609 
610 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
611 		if (error != 0)
612 			break;
613 		if (fp->f_type != DTYPE_VNODE) {
614 			error = EBADF;
615 			fdrop(fp, td);
616 			break;
617 		}
618 
619 		if (flp->l_whence == SEEK_CUR) {
620 			foffset = foffset_get(fp);
621 			if (foffset < 0 ||
622 			    (flp->l_start > 0 &&
623 			     foffset > OFF_MAX - flp->l_start)) {
624 				error = EOVERFLOW;
625 				fdrop(fp, td);
626 				break;
627 			}
628 			flp->l_start += foffset;
629 		}
630 
631 		vp = fp->f_vnode;
632 		switch (flp->l_type) {
633 		case F_RDLCK:
634 			if ((fp->f_flag & FREAD) == 0) {
635 				error = EBADF;
636 				break;
637 			}
638 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
639 				PROC_LOCK(p->p_leader);
640 				p->p_leader->p_flag |= P_ADVLOCK;
641 				PROC_UNLOCK(p->p_leader);
642 			}
643 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
644 			    flp, flg);
645 			break;
646 		case F_WRLCK:
647 			if ((fp->f_flag & FWRITE) == 0) {
648 				error = EBADF;
649 				break;
650 			}
651 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
652 				PROC_LOCK(p->p_leader);
653 				p->p_leader->p_flag |= P_ADVLOCK;
654 				PROC_UNLOCK(p->p_leader);
655 			}
656 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
657 			    flp, flg);
658 			break;
659 		case F_UNLCK:
660 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
661 			    flp, flg);
662 			break;
663 		case F_UNLCKSYS:
664 			if (flg != F_REMOTE) {
665 				error = EINVAL;
666 				break;
667 			}
668 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
669 			    F_UNLCKSYS, flp, flg);
670 			break;
671 		default:
672 			error = EINVAL;
673 			break;
674 		}
675 		if (error != 0 || flp->l_type == F_UNLCK ||
676 		    flp->l_type == F_UNLCKSYS) {
677 			fdrop(fp, td);
678 			break;
679 		}
680 
681 		/*
682 		 * Check for a race with close.
683 		 *
684 		 * The vnode is now advisory locked (or unlocked, but this case
685 		 * is not really important) as the caller requested.
686 		 * We had to drop the filedesc lock, so we need to recheck if
687 		 * the descriptor is still valid, because if it was closed
688 		 * in the meantime we need to remove advisory lock from the
689 		 * vnode - close on any descriptor leading to an advisory
690 		 * locked vnode, removes that lock.
691 		 * We will return 0 on purpose in that case, as the result of
692 		 * successful advisory lock might have been externally visible
693 		 * already. This is fine - effectively we pretend to the caller
694 		 * that the closing thread was a bit slower and that the
695 		 * advisory lock succeeded before the close.
696 		 */
697 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
698 		if (error != 0) {
699 			fdrop(fp, td);
700 			break;
701 		}
702 		if (fp != fp2) {
703 			flp->l_whence = SEEK_SET;
704 			flp->l_start = 0;
705 			flp->l_len = 0;
706 			flp->l_type = F_UNLCK;
707 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
708 			    F_UNLCK, flp, F_POSIX);
709 		}
710 		fdrop(fp, td);
711 		fdrop(fp2, td);
712 		break;
713 
714 	case F_GETLK:
715 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
716 		if (error != 0)
717 			break;
718 		if (fp->f_type != DTYPE_VNODE) {
719 			error = EBADF;
720 			fdrop(fp, td);
721 			break;
722 		}
723 		flp = (struct flock *)arg;
724 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
725 		    flp->l_type != F_UNLCK) {
726 			error = EINVAL;
727 			fdrop(fp, td);
728 			break;
729 		}
730 		if (flp->l_whence == SEEK_CUR) {
731 			foffset = foffset_get(fp);
732 			if ((flp->l_start > 0 &&
733 			    foffset > OFF_MAX - flp->l_start) ||
734 			    (flp->l_start < 0 &&
735 			    foffset < OFF_MIN - flp->l_start)) {
736 				error = EOVERFLOW;
737 				fdrop(fp, td);
738 				break;
739 			}
740 			flp->l_start += foffset;
741 		}
742 		vp = fp->f_vnode;
743 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
744 		    F_POSIX);
745 		fdrop(fp, td);
746 		break;
747 
748 	case F_ADD_SEALS:
749 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
750 		if (error != 0)
751 			break;
752 		error = fo_add_seals(fp, arg);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_GET_SEALS:
757 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
758 		if (error != 0)
759 			break;
760 		if (fo_get_seals(fp, &seals) == 0)
761 			td->td_retval[0] = seals;
762 		else
763 			error = EINVAL;
764 		fdrop(fp, td);
765 		break;
766 
767 	case F_RDAHEAD:
768 		arg = arg ? 128 * 1024: 0;
769 		/* FALLTHROUGH */
770 	case F_READAHEAD:
771 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
772 		if (error != 0)
773 			break;
774 		if (fp->f_type != DTYPE_VNODE) {
775 			fdrop(fp, td);
776 			error = EBADF;
777 			break;
778 		}
779 		vp = fp->f_vnode;
780 		if (vp->v_type != VREG) {
781 			fdrop(fp, td);
782 			error = ENOTTY;
783 			break;
784 		}
785 
786 		/*
787 		 * Exclusive lock synchronizes against f_seqcount reads and
788 		 * writes in sequential_heuristic().
789 		 */
790 		error = vn_lock(vp, LK_EXCLUSIVE);
791 		if (error != 0) {
792 			fdrop(fp, td);
793 			break;
794 		}
795 		if (arg >= 0) {
796 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
797 			arg = MIN(arg, INT_MAX - bsize + 1);
798 			fp->f_seqcount = MIN(IO_SEQMAX,
799 			    (arg + bsize - 1) / bsize);
800 			atomic_set_int(&fp->f_flag, FRDAHEAD);
801 		} else {
802 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
803 		}
804 		VOP_UNLOCK(vp);
805 		fdrop(fp, td);
806 		break;
807 
808 	case F_ISUNIONSTACK:
809 		/*
810 		 * Check if the vnode is part of a union stack (either the
811 		 * "union" flag from mount(2) or unionfs).
812 		 *
813 		 * Prior to introduction of this op libc's readdir would call
814 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
815 		 * data just to check fs name and a mount flag.
816 		 *
817 		 * Fixing the code to handle everything in the kernel instead
818 		 * is a non-trivial endeavor and has low priority, thus this
819 		 * horrible kludge facilitates the current behavior in a much
820 		 * cheaper manner until someone(tm) sorts this out.
821 		 */
822 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
823 		if (error != 0)
824 			break;
825 		if (fp->f_type != DTYPE_VNODE) {
826 			fdrop(fp, td);
827 			error = EBADF;
828 			break;
829 		}
830 		vp = fp->f_vnode;
831 		/*
832 		 * Since we don't prevent dooming the vnode even non-null mp
833 		 * found can become immediately stale. This is tolerable since
834 		 * mount points are type-stable (providing safe memory access)
835 		 * and any vfs op on this vnode going forward will return an
836 		 * error (meaning return value in this case is meaningless).
837 		 */
838 		mp = atomic_load_ptr(&vp->v_mount);
839 		if (__predict_false(mp == NULL)) {
840 			fdrop(fp, td);
841 			error = EBADF;
842 			break;
843 		}
844 		td->td_retval[0] = 0;
845 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
846 		    mp->mnt_flag & MNT_UNION)
847 			td->td_retval[0] = 1;
848 		fdrop(fp, td);
849 		break;
850 
851 	default:
852 		error = EINVAL;
853 		break;
854 	}
855 	return (error);
856 }
857 
858 static int
859 getmaxfd(struct thread *td)
860 {
861 
862 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
863 }
864 
865 /*
866  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
867  */
868 int
869 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
870 {
871 	struct filedesc *fdp;
872 	struct filedescent *oldfde, *newfde;
873 	struct proc *p;
874 	struct file *delfp;
875 	u_long *oioctls, *nioctls;
876 	int error, maxfd;
877 
878 	p = td->td_proc;
879 	fdp = p->p_fd;
880 	oioctls = NULL;
881 
882 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
883 	MPASS(mode < FDDUP_LASTMODE);
884 
885 	AUDIT_ARG_FD(old);
886 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
887 
888 	/*
889 	 * Verify we have a valid descriptor to dup from and possibly to
890 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
891 	 * return EINVAL when the new descriptor is out of bounds.
892 	 */
893 	if (old < 0)
894 		return (EBADF);
895 	if (new < 0)
896 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
897 	maxfd = getmaxfd(td);
898 	if (new >= maxfd)
899 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
900 
901 	error = EBADF;
902 	FILEDESC_XLOCK(fdp);
903 	if (fget_locked(fdp, old) == NULL)
904 		goto unlock;
905 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
906 		td->td_retval[0] = new;
907 		if (flags & FDDUP_FLAG_CLOEXEC)
908 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
909 		error = 0;
910 		goto unlock;
911 	}
912 
913 	oldfde = &fdp->fd_ofiles[old];
914 	if (!fhold(oldfde->fde_file))
915 		goto unlock;
916 
917 	/*
918 	 * If the caller specified a file descriptor, make sure the file
919 	 * table is large enough to hold it, and grab it.  Otherwise, just
920 	 * allocate a new descriptor the usual way.
921 	 */
922 	switch (mode) {
923 	case FDDUP_NORMAL:
924 	case FDDUP_FCNTL:
925 		if ((error = fdalloc(td, new, &new)) != 0) {
926 			fdrop(oldfde->fde_file, td);
927 			goto unlock;
928 		}
929 		break;
930 	case FDDUP_MUSTREPLACE:
931 		/* Target file descriptor must exist. */
932 		if (fget_locked(fdp, new) == NULL) {
933 			fdrop(oldfde->fde_file, td);
934 			goto unlock;
935 		}
936 		break;
937 	case FDDUP_FIXED:
938 		if (new >= fdp->fd_nfiles) {
939 			/*
940 			 * The resource limits are here instead of e.g.
941 			 * fdalloc(), because the file descriptor table may be
942 			 * shared between processes, so we can't really use
943 			 * racct_add()/racct_sub().  Instead of counting the
944 			 * number of actually allocated descriptors, just put
945 			 * the limit on the size of the file descriptor table.
946 			 */
947 #ifdef RACCT
948 			if (RACCT_ENABLED()) {
949 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
950 				if (error != 0) {
951 					error = EMFILE;
952 					fdrop(oldfde->fde_file, td);
953 					goto unlock;
954 				}
955 			}
956 #endif
957 			fdgrowtable_exp(fdp, new + 1);
958 		}
959 		if (!fdisused(fdp, new))
960 			fdused(fdp, new);
961 		break;
962 	default:
963 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
964 	}
965 
966 	KASSERT(old != new, ("new fd is same as old"));
967 
968 	newfde = &fdp->fd_ofiles[new];
969 	delfp = newfde->fde_file;
970 
971 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
972 
973 	/*
974 	 * Duplicate the source descriptor.
975 	 */
976 #ifdef CAPABILITIES
977 	seqc_write_begin(&newfde->fde_seqc);
978 #endif
979 	oioctls = filecaps_free_prep(&newfde->fde_caps);
980 	memcpy(newfde, oldfde, fde_change_size);
981 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
982 	    nioctls);
983 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
984 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
985 	else
986 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
987 #ifdef CAPABILITIES
988 	seqc_write_end(&newfde->fde_seqc);
989 #endif
990 	td->td_retval[0] = new;
991 
992 	error = 0;
993 
994 	if (delfp != NULL) {
995 		(void) closefp(fdp, new, delfp, td, 1);
996 		FILEDESC_UNLOCK_ASSERT(fdp);
997 	} else {
998 unlock:
999 		FILEDESC_XUNLOCK(fdp);
1000 	}
1001 
1002 	filecaps_free_finish(oioctls);
1003 	return (error);
1004 }
1005 
1006 /*
1007  * If sigio is on the list associated with a process or process group,
1008  * disable signalling from the device, remove sigio from the list and
1009  * free sigio.
1010  */
1011 void
1012 funsetown(struct sigio **sigiop)
1013 {
1014 	struct sigio *sigio;
1015 
1016 	if (*sigiop == NULL)
1017 		return;
1018 	SIGIO_LOCK();
1019 	sigio = *sigiop;
1020 	if (sigio == NULL) {
1021 		SIGIO_UNLOCK();
1022 		return;
1023 	}
1024 	*(sigio->sio_myref) = NULL;
1025 	if ((sigio)->sio_pgid < 0) {
1026 		struct pgrp *pg = (sigio)->sio_pgrp;
1027 		PGRP_LOCK(pg);
1028 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1029 			    sigio, sio_pgsigio);
1030 		PGRP_UNLOCK(pg);
1031 	} else {
1032 		struct proc *p = (sigio)->sio_proc;
1033 		PROC_LOCK(p);
1034 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1035 			    sigio, sio_pgsigio);
1036 		PROC_UNLOCK(p);
1037 	}
1038 	SIGIO_UNLOCK();
1039 	crfree(sigio->sio_ucred);
1040 	free(sigio, M_SIGIO);
1041 }
1042 
1043 /*
1044  * Free a list of sigio structures.
1045  * We only need to lock the SIGIO_LOCK because we have made ourselves
1046  * inaccessible to callers of fsetown and therefore do not need to lock
1047  * the proc or pgrp struct for the list manipulation.
1048  */
1049 void
1050 funsetownlst(struct sigiolst *sigiolst)
1051 {
1052 	struct proc *p;
1053 	struct pgrp *pg;
1054 	struct sigio *sigio;
1055 
1056 	sigio = SLIST_FIRST(sigiolst);
1057 	if (sigio == NULL)
1058 		return;
1059 	p = NULL;
1060 	pg = NULL;
1061 
1062 	/*
1063 	 * Every entry of the list should belong
1064 	 * to a single proc or pgrp.
1065 	 */
1066 	if (sigio->sio_pgid < 0) {
1067 		pg = sigio->sio_pgrp;
1068 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1069 	} else /* if (sigio->sio_pgid > 0) */ {
1070 		p = sigio->sio_proc;
1071 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1072 	}
1073 
1074 	SIGIO_LOCK();
1075 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1076 		*(sigio->sio_myref) = NULL;
1077 		if (pg != NULL) {
1078 			KASSERT(sigio->sio_pgid < 0,
1079 			    ("Proc sigio in pgrp sigio list"));
1080 			KASSERT(sigio->sio_pgrp == pg,
1081 			    ("Bogus pgrp in sigio list"));
1082 			PGRP_LOCK(pg);
1083 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1084 			    sio_pgsigio);
1085 			PGRP_UNLOCK(pg);
1086 		} else /* if (p != NULL) */ {
1087 			KASSERT(sigio->sio_pgid > 0,
1088 			    ("Pgrp sigio in proc sigio list"));
1089 			KASSERT(sigio->sio_proc == p,
1090 			    ("Bogus proc in sigio list"));
1091 			PROC_LOCK(p);
1092 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1093 			    sio_pgsigio);
1094 			PROC_UNLOCK(p);
1095 		}
1096 		SIGIO_UNLOCK();
1097 		crfree(sigio->sio_ucred);
1098 		free(sigio, M_SIGIO);
1099 		SIGIO_LOCK();
1100 	}
1101 	SIGIO_UNLOCK();
1102 }
1103 
1104 /*
1105  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1106  *
1107  * After permission checking, add a sigio structure to the sigio list for
1108  * the process or process group.
1109  */
1110 int
1111 fsetown(pid_t pgid, struct sigio **sigiop)
1112 {
1113 	struct proc *proc;
1114 	struct pgrp *pgrp;
1115 	struct sigio *sigio;
1116 	int ret;
1117 
1118 	if (pgid == 0) {
1119 		funsetown(sigiop);
1120 		return (0);
1121 	}
1122 
1123 	ret = 0;
1124 
1125 	/* Allocate and fill in the new sigio out of locks. */
1126 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1127 	sigio->sio_pgid = pgid;
1128 	sigio->sio_ucred = crhold(curthread->td_ucred);
1129 	sigio->sio_myref = sigiop;
1130 
1131 	sx_slock(&proctree_lock);
1132 	if (pgid > 0) {
1133 		proc = pfind(pgid);
1134 		if (proc == NULL) {
1135 			ret = ESRCH;
1136 			goto fail;
1137 		}
1138 
1139 		/*
1140 		 * Policy - Don't allow a process to FSETOWN a process
1141 		 * in another session.
1142 		 *
1143 		 * Remove this test to allow maximum flexibility or
1144 		 * restrict FSETOWN to the current process or process
1145 		 * group for maximum safety.
1146 		 */
1147 		PROC_UNLOCK(proc);
1148 		if (proc->p_session != curthread->td_proc->p_session) {
1149 			ret = EPERM;
1150 			goto fail;
1151 		}
1152 
1153 		pgrp = NULL;
1154 	} else /* if (pgid < 0) */ {
1155 		pgrp = pgfind(-pgid);
1156 		if (pgrp == NULL) {
1157 			ret = ESRCH;
1158 			goto fail;
1159 		}
1160 		PGRP_UNLOCK(pgrp);
1161 
1162 		/*
1163 		 * Policy - Don't allow a process to FSETOWN a process
1164 		 * in another session.
1165 		 *
1166 		 * Remove this test to allow maximum flexibility or
1167 		 * restrict FSETOWN to the current process or process
1168 		 * group for maximum safety.
1169 		 */
1170 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1171 			ret = EPERM;
1172 			goto fail;
1173 		}
1174 
1175 		proc = NULL;
1176 	}
1177 	funsetown(sigiop);
1178 	if (pgid > 0) {
1179 		PROC_LOCK(proc);
1180 		/*
1181 		 * Since funsetownlst() is called without the proctree
1182 		 * locked, we need to check for P_WEXIT.
1183 		 * XXX: is ESRCH correct?
1184 		 */
1185 		if ((proc->p_flag & P_WEXIT) != 0) {
1186 			PROC_UNLOCK(proc);
1187 			ret = ESRCH;
1188 			goto fail;
1189 		}
1190 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1191 		sigio->sio_proc = proc;
1192 		PROC_UNLOCK(proc);
1193 	} else {
1194 		PGRP_LOCK(pgrp);
1195 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1196 		sigio->sio_pgrp = pgrp;
1197 		PGRP_UNLOCK(pgrp);
1198 	}
1199 	sx_sunlock(&proctree_lock);
1200 	SIGIO_LOCK();
1201 	*sigiop = sigio;
1202 	SIGIO_UNLOCK();
1203 	return (0);
1204 
1205 fail:
1206 	sx_sunlock(&proctree_lock);
1207 	crfree(sigio->sio_ucred);
1208 	free(sigio, M_SIGIO);
1209 	return (ret);
1210 }
1211 
1212 /*
1213  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1214  */
1215 pid_t
1216 fgetown(struct sigio **sigiop)
1217 {
1218 	pid_t pgid;
1219 
1220 	SIGIO_LOCK();
1221 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1222 	SIGIO_UNLOCK();
1223 	return (pgid);
1224 }
1225 
1226 /*
1227  * Function drops the filedesc lock on return.
1228  */
1229 static int
1230 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1231     int holdleaders)
1232 {
1233 	int error;
1234 
1235 	FILEDESC_XLOCK_ASSERT(fdp);
1236 
1237 	if (holdleaders) {
1238 		if (td->td_proc->p_fdtol != NULL) {
1239 			/*
1240 			 * Ask fdfree() to sleep to ensure that all relevant
1241 			 * process leaders can be traversed in closef().
1242 			 */
1243 			fdp->fd_holdleaderscount++;
1244 		} else {
1245 			holdleaders = 0;
1246 		}
1247 	}
1248 
1249 	/*
1250 	 * We now hold the fp reference that used to be owned by the
1251 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1252 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1253 	 * added, and deleteing a knote for the new fd.
1254 	 */
1255 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1256 		knote_fdclose(td, fd);
1257 
1258 	/*
1259 	 * We need to notify mqueue if the object is of type mqueue.
1260 	 */
1261 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1262 		mq_fdclose(td, fd, fp);
1263 	FILEDESC_XUNLOCK(fdp);
1264 
1265 	error = closef(fp, td);
1266 	if (holdleaders) {
1267 		FILEDESC_XLOCK(fdp);
1268 		fdp->fd_holdleaderscount--;
1269 		if (fdp->fd_holdleaderscount == 0 &&
1270 		    fdp->fd_holdleaderswakeup != 0) {
1271 			fdp->fd_holdleaderswakeup = 0;
1272 			wakeup(&fdp->fd_holdleaderscount);
1273 		}
1274 		FILEDESC_XUNLOCK(fdp);
1275 	}
1276 	return (error);
1277 }
1278 
1279 /*
1280  * Close a file descriptor.
1281  */
1282 #ifndef _SYS_SYSPROTO_H_
1283 struct close_args {
1284 	int     fd;
1285 };
1286 #endif
1287 /* ARGSUSED */
1288 int
1289 sys_close(struct thread *td, struct close_args *uap)
1290 {
1291 
1292 	return (kern_close(td, uap->fd));
1293 }
1294 
1295 int
1296 kern_close(struct thread *td, int fd)
1297 {
1298 	struct filedesc *fdp;
1299 	struct file *fp;
1300 
1301 	fdp = td->td_proc->p_fd;
1302 
1303 	AUDIT_SYSCLOSE(td, fd);
1304 
1305 	FILEDESC_XLOCK(fdp);
1306 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1307 		FILEDESC_XUNLOCK(fdp);
1308 		return (EBADF);
1309 	}
1310 	fdfree(fdp, fd);
1311 
1312 	/* closefp() drops the FILEDESC lock for us. */
1313 	return (closefp(fdp, fd, fp, td, 1));
1314 }
1315 
1316 /*
1317  * Close open file descriptors.
1318  */
1319 #ifndef _SYS_SYSPROTO_H_
1320 struct closefrom_args {
1321 	int	lowfd;
1322 };
1323 #endif
1324 /* ARGSUSED */
1325 int
1326 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1327 {
1328 	struct filedesc *fdp;
1329 	int fd;
1330 
1331 	fdp = td->td_proc->p_fd;
1332 	AUDIT_ARG_FD(uap->lowfd);
1333 
1334 	/*
1335 	 * Treat negative starting file descriptor values identical to
1336 	 * closefrom(0) which closes all files.
1337 	 */
1338 	if (uap->lowfd < 0)
1339 		uap->lowfd = 0;
1340 	FILEDESC_SLOCK(fdp);
1341 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1342 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1343 			FILEDESC_SUNLOCK(fdp);
1344 			(void)kern_close(td, fd);
1345 			FILEDESC_SLOCK(fdp);
1346 		}
1347 	}
1348 	FILEDESC_SUNLOCK(fdp);
1349 	return (0);
1350 }
1351 
1352 #if defined(COMPAT_43)
1353 /*
1354  * Return status information about a file descriptor.
1355  */
1356 #ifndef _SYS_SYSPROTO_H_
1357 struct ofstat_args {
1358 	int	fd;
1359 	struct	ostat *sb;
1360 };
1361 #endif
1362 /* ARGSUSED */
1363 int
1364 ofstat(struct thread *td, struct ofstat_args *uap)
1365 {
1366 	struct ostat oub;
1367 	struct stat ub;
1368 	int error;
1369 
1370 	error = kern_fstat(td, uap->fd, &ub);
1371 	if (error == 0) {
1372 		cvtstat(&ub, &oub);
1373 		error = copyout(&oub, uap->sb, sizeof(oub));
1374 	}
1375 	return (error);
1376 }
1377 #endif /* COMPAT_43 */
1378 
1379 #if defined(COMPAT_FREEBSD11)
1380 int
1381 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1382 {
1383 	struct stat sb;
1384 	struct freebsd11_stat osb;
1385 	int error;
1386 
1387 	error = kern_fstat(td, uap->fd, &sb);
1388 	if (error != 0)
1389 		return (error);
1390 	error = freebsd11_cvtstat(&sb, &osb);
1391 	if (error == 0)
1392 		error = copyout(&osb, uap->sb, sizeof(osb));
1393 	return (error);
1394 }
1395 #endif	/* COMPAT_FREEBSD11 */
1396 
1397 /*
1398  * Return status information about a file descriptor.
1399  */
1400 #ifndef _SYS_SYSPROTO_H_
1401 struct fstat_args {
1402 	int	fd;
1403 	struct	stat *sb;
1404 };
1405 #endif
1406 /* ARGSUSED */
1407 int
1408 sys_fstat(struct thread *td, struct fstat_args *uap)
1409 {
1410 	struct stat ub;
1411 	int error;
1412 
1413 	error = kern_fstat(td, uap->fd, &ub);
1414 	if (error == 0)
1415 		error = copyout(&ub, uap->sb, sizeof(ub));
1416 	return (error);
1417 }
1418 
1419 int
1420 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1421 {
1422 	struct file *fp;
1423 	int error;
1424 
1425 	AUDIT_ARG_FD(fd);
1426 
1427 	error = fget(td, fd, &cap_fstat_rights, &fp);
1428 	if (__predict_false(error != 0))
1429 		return (error);
1430 
1431 	AUDIT_ARG_FILE(td->td_proc, fp);
1432 
1433 	error = fo_stat(fp, sbp, td->td_ucred, td);
1434 	fdrop(fp, td);
1435 #ifdef __STAT_TIME_T_EXT
1436 	sbp->st_atim_ext = 0;
1437 	sbp->st_mtim_ext = 0;
1438 	sbp->st_ctim_ext = 0;
1439 	sbp->st_btim_ext = 0;
1440 #endif
1441 #ifdef KTRACE
1442 	if (KTRPOINT(td, KTR_STRUCT))
1443 		ktrstat_error(sbp, error);
1444 #endif
1445 	return (error);
1446 }
1447 
1448 #if defined(COMPAT_FREEBSD11)
1449 /*
1450  * Return status information about a file descriptor.
1451  */
1452 #ifndef _SYS_SYSPROTO_H_
1453 struct freebsd11_nfstat_args {
1454 	int	fd;
1455 	struct	nstat *sb;
1456 };
1457 #endif
1458 /* ARGSUSED */
1459 int
1460 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1461 {
1462 	struct nstat nub;
1463 	struct stat ub;
1464 	int error;
1465 
1466 	error = kern_fstat(td, uap->fd, &ub);
1467 	if (error == 0) {
1468 		freebsd11_cvtnstat(&ub, &nub);
1469 		error = copyout(&nub, uap->sb, sizeof(nub));
1470 	}
1471 	return (error);
1472 }
1473 #endif /* COMPAT_FREEBSD11 */
1474 
1475 /*
1476  * Return pathconf information about a file descriptor.
1477  */
1478 #ifndef _SYS_SYSPROTO_H_
1479 struct fpathconf_args {
1480 	int	fd;
1481 	int	name;
1482 };
1483 #endif
1484 /* ARGSUSED */
1485 int
1486 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1487 {
1488 	long value;
1489 	int error;
1490 
1491 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1492 	if (error == 0)
1493 		td->td_retval[0] = value;
1494 	return (error);
1495 }
1496 
1497 int
1498 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1499 {
1500 	struct file *fp;
1501 	struct vnode *vp;
1502 	int error;
1503 
1504 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1505 	if (error != 0)
1506 		return (error);
1507 
1508 	if (name == _PC_ASYNC_IO) {
1509 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1510 		goto out;
1511 	}
1512 	vp = fp->f_vnode;
1513 	if (vp != NULL) {
1514 		vn_lock(vp, LK_SHARED | LK_RETRY);
1515 		error = VOP_PATHCONF(vp, name, valuep);
1516 		VOP_UNLOCK(vp);
1517 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1518 		if (name != _PC_PIPE_BUF) {
1519 			error = EINVAL;
1520 		} else {
1521 			*valuep = PIPE_BUF;
1522 			error = 0;
1523 		}
1524 	} else {
1525 		error = EOPNOTSUPP;
1526 	}
1527 out:
1528 	fdrop(fp, td);
1529 	return (error);
1530 }
1531 
1532 /*
1533  * Copy filecaps structure allocating memory for ioctls array if needed.
1534  *
1535  * The last parameter indicates whether the fdtable is locked. If it is not and
1536  * ioctls are encountered, copying fails and the caller must lock the table.
1537  *
1538  * Note that if the table was not locked, the caller has to check the relevant
1539  * sequence counter to determine whether the operation was successful.
1540  */
1541 bool
1542 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1543 {
1544 	size_t size;
1545 
1546 	if (src->fc_ioctls != NULL && !locked)
1547 		return (false);
1548 	memcpy(dst, src, sizeof(*src));
1549 	if (src->fc_ioctls == NULL)
1550 		return (true);
1551 
1552 	KASSERT(src->fc_nioctls > 0,
1553 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1554 
1555 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1556 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1557 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1558 	return (true);
1559 }
1560 
1561 static u_long *
1562 filecaps_copy_prep(const struct filecaps *src)
1563 {
1564 	u_long *ioctls;
1565 	size_t size;
1566 
1567 	if (__predict_true(src->fc_ioctls == NULL))
1568 		return (NULL);
1569 
1570 	KASSERT(src->fc_nioctls > 0,
1571 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1572 
1573 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1574 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1575 	return (ioctls);
1576 }
1577 
1578 static void
1579 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1580     u_long *ioctls)
1581 {
1582 	size_t size;
1583 
1584 	*dst = *src;
1585 	if (__predict_true(src->fc_ioctls == NULL)) {
1586 		MPASS(ioctls == NULL);
1587 		return;
1588 	}
1589 
1590 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1591 	dst->fc_ioctls = ioctls;
1592 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1593 }
1594 
1595 /*
1596  * Move filecaps structure to the new place and clear the old place.
1597  */
1598 void
1599 filecaps_move(struct filecaps *src, struct filecaps *dst)
1600 {
1601 
1602 	*dst = *src;
1603 	bzero(src, sizeof(*src));
1604 }
1605 
1606 /*
1607  * Fill the given filecaps structure with full rights.
1608  */
1609 static void
1610 filecaps_fill(struct filecaps *fcaps)
1611 {
1612 
1613 	CAP_ALL(&fcaps->fc_rights);
1614 	fcaps->fc_ioctls = NULL;
1615 	fcaps->fc_nioctls = -1;
1616 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1617 }
1618 
1619 /*
1620  * Free memory allocated within filecaps structure.
1621  */
1622 void
1623 filecaps_free(struct filecaps *fcaps)
1624 {
1625 
1626 	free(fcaps->fc_ioctls, M_FILECAPS);
1627 	bzero(fcaps, sizeof(*fcaps));
1628 }
1629 
1630 static u_long *
1631 filecaps_free_prep(struct filecaps *fcaps)
1632 {
1633 	u_long *ioctls;
1634 
1635 	ioctls = fcaps->fc_ioctls;
1636 	bzero(fcaps, sizeof(*fcaps));
1637 	return (ioctls);
1638 }
1639 
1640 static void
1641 filecaps_free_finish(u_long *ioctls)
1642 {
1643 
1644 	free(ioctls, M_FILECAPS);
1645 }
1646 
1647 /*
1648  * Validate the given filecaps structure.
1649  */
1650 static void
1651 filecaps_validate(const struct filecaps *fcaps, const char *func)
1652 {
1653 
1654 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1655 	    ("%s: invalid rights", func));
1656 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1657 	    ("%s: invalid fcntls", func));
1658 	KASSERT(fcaps->fc_fcntls == 0 ||
1659 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1660 	    ("%s: fcntls without CAP_FCNTL", func));
1661 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1662 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1663 	    ("%s: invalid ioctls", func));
1664 	KASSERT(fcaps->fc_nioctls == 0 ||
1665 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1666 	    ("%s: ioctls without CAP_IOCTL", func));
1667 }
1668 
1669 static void
1670 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1671 {
1672 	int nfd1;
1673 
1674 	FILEDESC_XLOCK_ASSERT(fdp);
1675 
1676 	nfd1 = fdp->fd_nfiles * 2;
1677 	if (nfd1 < nfd)
1678 		nfd1 = nfd;
1679 	fdgrowtable(fdp, nfd1);
1680 }
1681 
1682 /*
1683  * Grow the file table to accommodate (at least) nfd descriptors.
1684  */
1685 static void
1686 fdgrowtable(struct filedesc *fdp, int nfd)
1687 {
1688 	struct filedesc0 *fdp0;
1689 	struct freetable *ft;
1690 	struct fdescenttbl *ntable;
1691 	struct fdescenttbl *otable;
1692 	int nnfiles, onfiles;
1693 	NDSLOTTYPE *nmap, *omap;
1694 
1695 	/*
1696 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1697 	 * growing it to accommodate for the one we are going to copy from. There
1698 	 * is no need to have a lock on this one as it's not visible to anyone.
1699 	 */
1700 	if (fdp->fd_lastfile != -1)
1701 		FILEDESC_XLOCK_ASSERT(fdp);
1702 
1703 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1704 
1705 	/* save old values */
1706 	onfiles = fdp->fd_nfiles;
1707 	otable = fdp->fd_files;
1708 	omap = fdp->fd_map;
1709 
1710 	/* compute the size of the new table */
1711 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1712 	if (nnfiles <= onfiles)
1713 		/* the table is already large enough */
1714 		return;
1715 
1716 	/*
1717 	 * Allocate a new table.  We need enough space for the number of
1718 	 * entries, file entries themselves and the struct freetable we will use
1719 	 * when we decommission the table and place it on the freelist.
1720 	 * We place the struct freetable in the middle so we don't have
1721 	 * to worry about padding.
1722 	 */
1723 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1724 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1725 	    sizeof(struct freetable),
1726 	    M_FILEDESC, M_ZERO | M_WAITOK);
1727 	/* copy the old data */
1728 	ntable->fdt_nfiles = nnfiles;
1729 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1730 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1731 
1732 	/*
1733 	 * Allocate a new map only if the old is not large enough.  It will
1734 	 * grow at a slower rate than the table as it can map more
1735 	 * entries than the table can hold.
1736 	 */
1737 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1738 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1739 		    M_ZERO | M_WAITOK);
1740 		/* copy over the old data and update the pointer */
1741 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1742 		fdp->fd_map = nmap;
1743 	}
1744 
1745 	/*
1746 	 * Make sure that ntable is correctly initialized before we replace
1747 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1748 	 * data.
1749 	 */
1750 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1751 
1752 	/*
1753 	 * Do not free the old file table, as some threads may still
1754 	 * reference entries within it.  Instead, place it on a freelist
1755 	 * which will be processed when the struct filedesc is released.
1756 	 *
1757 	 * Note that if onfiles == NDFILE, we're dealing with the original
1758 	 * static allocation contained within (struct filedesc0 *)fdp,
1759 	 * which must not be freed.
1760 	 */
1761 	if (onfiles > NDFILE) {
1762 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1763 		fdp0 = (struct filedesc0 *)fdp;
1764 		ft->ft_table = otable;
1765 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1766 	}
1767 	/*
1768 	 * The map does not have the same possibility of threads still
1769 	 * holding references to it.  So always free it as long as it
1770 	 * does not reference the original static allocation.
1771 	 */
1772 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1773 		free(omap, M_FILEDESC);
1774 }
1775 
1776 /*
1777  * Allocate a file descriptor for the process.
1778  */
1779 int
1780 fdalloc(struct thread *td, int minfd, int *result)
1781 {
1782 	struct proc *p = td->td_proc;
1783 	struct filedesc *fdp = p->p_fd;
1784 	int fd, maxfd, allocfd;
1785 #ifdef RACCT
1786 	int error;
1787 #endif
1788 
1789 	FILEDESC_XLOCK_ASSERT(fdp);
1790 
1791 	if (fdp->fd_freefile > minfd)
1792 		minfd = fdp->fd_freefile;
1793 
1794 	maxfd = getmaxfd(td);
1795 
1796 	/*
1797 	 * Search the bitmap for a free descriptor starting at minfd.
1798 	 * If none is found, grow the file table.
1799 	 */
1800 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1801 	if (fd >= maxfd)
1802 		return (EMFILE);
1803 	if (fd >= fdp->fd_nfiles) {
1804 		allocfd = min(fd * 2, maxfd);
1805 #ifdef RACCT
1806 		if (RACCT_ENABLED()) {
1807 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1808 			if (error != 0)
1809 				return (EMFILE);
1810 		}
1811 #endif
1812 		/*
1813 		 * fd is already equal to first free descriptor >= minfd, so
1814 		 * we only need to grow the table and we are done.
1815 		 */
1816 		fdgrowtable_exp(fdp, allocfd);
1817 	}
1818 
1819 	/*
1820 	 * Perform some sanity checks, then mark the file descriptor as
1821 	 * used and return it to the caller.
1822 	 */
1823 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1824 	    ("invalid descriptor %d", fd));
1825 	KASSERT(!fdisused(fdp, fd),
1826 	    ("fd_first_free() returned non-free descriptor"));
1827 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1828 	    ("file descriptor isn't free"));
1829 	fdused(fdp, fd);
1830 	*result = fd;
1831 	return (0);
1832 }
1833 
1834 /*
1835  * Allocate n file descriptors for the process.
1836  */
1837 int
1838 fdallocn(struct thread *td, int minfd, int *fds, int n)
1839 {
1840 	struct proc *p = td->td_proc;
1841 	struct filedesc *fdp = p->p_fd;
1842 	int i;
1843 
1844 	FILEDESC_XLOCK_ASSERT(fdp);
1845 
1846 	for (i = 0; i < n; i++)
1847 		if (fdalloc(td, 0, &fds[i]) != 0)
1848 			break;
1849 
1850 	if (i < n) {
1851 		for (i--; i >= 0; i--)
1852 			fdunused(fdp, fds[i]);
1853 		return (EMFILE);
1854 	}
1855 
1856 	return (0);
1857 }
1858 
1859 /*
1860  * Create a new open file structure and allocate a file descriptor for the
1861  * process that refers to it.  We add one reference to the file for the
1862  * descriptor table and one reference for resultfp. This is to prevent us
1863  * being preempted and the entry in the descriptor table closed after we
1864  * release the FILEDESC lock.
1865  */
1866 int
1867 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1868     struct filecaps *fcaps)
1869 {
1870 	struct file *fp;
1871 	int error, fd;
1872 
1873 	error = falloc_noinstall(td, &fp);
1874 	if (error)
1875 		return (error);		/* no reference held on error */
1876 
1877 	error = finstall(td, fp, &fd, flags, fcaps);
1878 	if (error) {
1879 		fdrop(fp, td);		/* one reference (fp only) */
1880 		return (error);
1881 	}
1882 
1883 	if (resultfp != NULL)
1884 		*resultfp = fp;		/* copy out result */
1885 	else
1886 		fdrop(fp, td);		/* release local reference */
1887 
1888 	if (resultfd != NULL)
1889 		*resultfd = fd;
1890 
1891 	return (0);
1892 }
1893 
1894 /*
1895  * Create a new open file structure without allocating a file descriptor.
1896  */
1897 int
1898 falloc_noinstall(struct thread *td, struct file **resultfp)
1899 {
1900 	struct file *fp;
1901 	int maxuserfiles = maxfiles - (maxfiles / 20);
1902 	int openfiles_new;
1903 	static struct timeval lastfail;
1904 	static int curfail;
1905 
1906 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1907 
1908 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1909 	if ((openfiles_new >= maxuserfiles &&
1910 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1911 	    openfiles_new >= maxfiles) {
1912 		atomic_subtract_int(&openfiles, 1);
1913 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1914 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1915 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1916 		}
1917 		return (ENFILE);
1918 	}
1919 	fp = uma_zalloc(file_zone, M_WAITOK);
1920 	bzero(fp, sizeof(*fp));
1921 	refcount_init(&fp->f_count, 1);
1922 	fp->f_cred = crhold(td->td_ucred);
1923 	fp->f_ops = &badfileops;
1924 	*resultfp = fp;
1925 	return (0);
1926 }
1927 
1928 /*
1929  * Install a file in a file descriptor table.
1930  */
1931 void
1932 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1933     struct filecaps *fcaps)
1934 {
1935 	struct filedescent *fde;
1936 
1937 	MPASS(fp != NULL);
1938 	if (fcaps != NULL)
1939 		filecaps_validate(fcaps, __func__);
1940 	FILEDESC_XLOCK_ASSERT(fdp);
1941 
1942 	fde = &fdp->fd_ofiles[fd];
1943 #ifdef CAPABILITIES
1944 	seqc_write_begin(&fde->fde_seqc);
1945 #endif
1946 	fde->fde_file = fp;
1947 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1948 	if (fcaps != NULL)
1949 		filecaps_move(fcaps, &fde->fde_caps);
1950 	else
1951 		filecaps_fill(&fde->fde_caps);
1952 #ifdef CAPABILITIES
1953 	seqc_write_end(&fde->fde_seqc);
1954 #endif
1955 }
1956 
1957 int
1958 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1959     struct filecaps *fcaps)
1960 {
1961 	struct filedesc *fdp = td->td_proc->p_fd;
1962 	int error;
1963 
1964 	MPASS(fd != NULL);
1965 
1966 	if (!fhold(fp))
1967 		return (EBADF);
1968 	FILEDESC_XLOCK(fdp);
1969 	error = fdalloc(td, 0, fd);
1970 	if (__predict_false(error != 0)) {
1971 		FILEDESC_XUNLOCK(fdp);
1972 		fdrop(fp, td);
1973 		return (error);
1974 	}
1975 	_finstall(fdp, fp, *fd, flags, fcaps);
1976 	FILEDESC_XUNLOCK(fdp);
1977 	return (0);
1978 }
1979 
1980 /*
1981  * Build a new filedesc structure from another.
1982  * Copy the current, root, and jail root vnode references.
1983  *
1984  * If fdp is not NULL, return with it shared locked.
1985  */
1986 struct filedesc *
1987 fdinit(struct filedesc *fdp, bool prepfiles)
1988 {
1989 	struct filedesc0 *newfdp0;
1990 	struct filedesc *newfdp;
1991 	struct pwd *newpwd;
1992 
1993 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1994 	newfdp = &newfdp0->fd_fd;
1995 
1996 	/* Create the file descriptor table. */
1997 	FILEDESC_LOCK_INIT(newfdp);
1998 	refcount_init(&newfdp->fd_refcnt, 1);
1999 	refcount_init(&newfdp->fd_holdcnt, 1);
2000 	newfdp->fd_cmask = CMASK;
2001 	newfdp->fd_map = newfdp0->fd_dmap;
2002 	newfdp->fd_lastfile = -1;
2003 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2004 	newfdp->fd_files->fdt_nfiles = NDFILE;
2005 
2006 	if (fdp == NULL) {
2007 		newpwd = pwd_alloc();
2008 		smr_serialized_store(&newfdp->fd_pwd, newpwd, true);
2009 		return (newfdp);
2010 	}
2011 
2012 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
2013 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2014 
2015 	FILEDESC_SLOCK(fdp);
2016 	newpwd = pwd_hold_filedesc(fdp);
2017 	smr_serialized_store(&newfdp->fd_pwd, newpwd, true);
2018 
2019 	if (!prepfiles) {
2020 		FILEDESC_SUNLOCK(fdp);
2021 	} else {
2022 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
2023 			FILEDESC_SUNLOCK(fdp);
2024 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2025 			FILEDESC_SLOCK(fdp);
2026 		}
2027 	}
2028 
2029 	return (newfdp);
2030 }
2031 
2032 static struct filedesc *
2033 fdhold(struct proc *p)
2034 {
2035 	struct filedesc *fdp;
2036 
2037 	PROC_LOCK_ASSERT(p, MA_OWNED);
2038 	fdp = p->p_fd;
2039 	if (fdp != NULL)
2040 		refcount_acquire(&fdp->fd_holdcnt);
2041 	return (fdp);
2042 }
2043 
2044 static void
2045 fddrop(struct filedesc *fdp)
2046 {
2047 
2048 	if (fdp->fd_holdcnt > 1) {
2049 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2050 			return;
2051 	}
2052 
2053 	FILEDESC_LOCK_DESTROY(fdp);
2054 	uma_zfree(filedesc0_zone, fdp);
2055 }
2056 
2057 /*
2058  * Share a filedesc structure.
2059  */
2060 struct filedesc *
2061 fdshare(struct filedesc *fdp)
2062 {
2063 
2064 	refcount_acquire(&fdp->fd_refcnt);
2065 	return (fdp);
2066 }
2067 
2068 /*
2069  * Unshare a filedesc structure, if necessary by making a copy
2070  */
2071 void
2072 fdunshare(struct thread *td)
2073 {
2074 	struct filedesc *tmp;
2075 	struct proc *p = td->td_proc;
2076 
2077 	if (p->p_fd->fd_refcnt == 1)
2078 		return;
2079 
2080 	tmp = fdcopy(p->p_fd);
2081 	fdescfree(td);
2082 	p->p_fd = tmp;
2083 }
2084 
2085 void
2086 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2087 {
2088 
2089 	fdescfree(td);
2090 	td->td_proc->p_fd = fdp;
2091 }
2092 
2093 /*
2094  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2095  * this is to ease callers, not catch errors.
2096  */
2097 struct filedesc *
2098 fdcopy(struct filedesc *fdp)
2099 {
2100 	struct filedesc *newfdp;
2101 	struct filedescent *nfde, *ofde;
2102 	int i;
2103 
2104 	MPASS(fdp != NULL);
2105 
2106 	newfdp = fdinit(fdp, true);
2107 	/* copy all passable descriptors (i.e. not kqueue) */
2108 	newfdp->fd_freefile = -1;
2109 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
2110 		ofde = &fdp->fd_ofiles[i];
2111 		if (ofde->fde_file == NULL ||
2112 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2113 		    !fhold(ofde->fde_file)) {
2114 			if (newfdp->fd_freefile == -1)
2115 				newfdp->fd_freefile = i;
2116 			continue;
2117 		}
2118 		nfde = &newfdp->fd_ofiles[i];
2119 		*nfde = *ofde;
2120 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2121 		fdused_init(newfdp, i);
2122 		newfdp->fd_lastfile = i;
2123 	}
2124 	if (newfdp->fd_freefile == -1)
2125 		newfdp->fd_freefile = i;
2126 	newfdp->fd_cmask = fdp->fd_cmask;
2127 	FILEDESC_SUNLOCK(fdp);
2128 	return (newfdp);
2129 }
2130 
2131 /*
2132  * Copies a filedesc structure, while remapping all file descriptors
2133  * stored inside using a translation table.
2134  *
2135  * File descriptors are copied over to the new file descriptor table,
2136  * regardless of whether the close-on-exec flag is set.
2137  */
2138 int
2139 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2140     struct filedesc **ret)
2141 {
2142 	struct filedesc *newfdp;
2143 	struct filedescent *nfde, *ofde;
2144 	int error, i;
2145 
2146 	MPASS(fdp != NULL);
2147 
2148 	newfdp = fdinit(fdp, true);
2149 	if (nfds > fdp->fd_lastfile + 1) {
2150 		/* New table cannot be larger than the old one. */
2151 		error = E2BIG;
2152 		goto bad;
2153 	}
2154 	/* Copy all passable descriptors (i.e. not kqueue). */
2155 	newfdp->fd_freefile = nfds;
2156 	for (i = 0; i < nfds; ++i) {
2157 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2158 			/* File descriptor out of bounds. */
2159 			error = EBADF;
2160 			goto bad;
2161 		}
2162 		ofde = &fdp->fd_ofiles[fds[i]];
2163 		if (ofde->fde_file == NULL) {
2164 			/* Unused file descriptor. */
2165 			error = EBADF;
2166 			goto bad;
2167 		}
2168 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2169 			/* File descriptor cannot be passed. */
2170 			error = EINVAL;
2171 			goto bad;
2172 		}
2173 		if (!fhold(nfde->fde_file)) {
2174 			error = EBADF;
2175 			goto bad;
2176 		}
2177 		nfde = &newfdp->fd_ofiles[i];
2178 		*nfde = *ofde;
2179 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2180 		fdused_init(newfdp, i);
2181 		newfdp->fd_lastfile = i;
2182 	}
2183 	newfdp->fd_cmask = fdp->fd_cmask;
2184 	FILEDESC_SUNLOCK(fdp);
2185 	*ret = newfdp;
2186 	return (0);
2187 bad:
2188 	FILEDESC_SUNLOCK(fdp);
2189 	fdescfree_remapped(newfdp);
2190 	return (error);
2191 }
2192 
2193 /*
2194  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2195  * one of processes using it exits) and the table used to be shared.
2196  */
2197 static void
2198 fdclearlocks(struct thread *td)
2199 {
2200 	struct filedesc *fdp;
2201 	struct filedesc_to_leader *fdtol;
2202 	struct flock lf;
2203 	struct file *fp;
2204 	struct proc *p;
2205 	struct vnode *vp;
2206 	int i;
2207 
2208 	p = td->td_proc;
2209 	fdp = p->p_fd;
2210 	fdtol = p->p_fdtol;
2211 	MPASS(fdtol != NULL);
2212 
2213 	FILEDESC_XLOCK(fdp);
2214 	KASSERT(fdtol->fdl_refcount > 0,
2215 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2216 	    fdtol->fdl_refcount));
2217 	if (fdtol->fdl_refcount == 1 &&
2218 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2219 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2220 			fp = fdp->fd_ofiles[i].fde_file;
2221 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2222 			    !fhold(fp))
2223 				continue;
2224 			FILEDESC_XUNLOCK(fdp);
2225 			lf.l_whence = SEEK_SET;
2226 			lf.l_start = 0;
2227 			lf.l_len = 0;
2228 			lf.l_type = F_UNLCK;
2229 			vp = fp->f_vnode;
2230 			(void) VOP_ADVLOCK(vp,
2231 			    (caddr_t)p->p_leader, F_UNLCK,
2232 			    &lf, F_POSIX);
2233 			FILEDESC_XLOCK(fdp);
2234 			fdrop(fp, td);
2235 		}
2236 	}
2237 retry:
2238 	if (fdtol->fdl_refcount == 1) {
2239 		if (fdp->fd_holdleaderscount > 0 &&
2240 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2241 			/*
2242 			 * close() or kern_dup() has cleared a reference
2243 			 * in a shared file descriptor table.
2244 			 */
2245 			fdp->fd_holdleaderswakeup = 1;
2246 			sx_sleep(&fdp->fd_holdleaderscount,
2247 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2248 			goto retry;
2249 		}
2250 		if (fdtol->fdl_holdcount > 0) {
2251 			/*
2252 			 * Ensure that fdtol->fdl_leader remains
2253 			 * valid in closef().
2254 			 */
2255 			fdtol->fdl_wakeup = 1;
2256 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2257 			    "fdlhold", 0);
2258 			goto retry;
2259 		}
2260 	}
2261 	fdtol->fdl_refcount--;
2262 	if (fdtol->fdl_refcount == 0 &&
2263 	    fdtol->fdl_holdcount == 0) {
2264 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2265 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2266 	} else
2267 		fdtol = NULL;
2268 	p->p_fdtol = NULL;
2269 	FILEDESC_XUNLOCK(fdp);
2270 	if (fdtol != NULL)
2271 		free(fdtol, M_FILEDESC_TO_LEADER);
2272 }
2273 
2274 /*
2275  * Release a filedesc structure.
2276  */
2277 static void
2278 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2279 {
2280 	struct filedesc0 *fdp0;
2281 	struct freetable *ft, *tft;
2282 	struct filedescent *fde;
2283 	struct file *fp;
2284 	int i;
2285 
2286 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2287 		fde = &fdp->fd_ofiles[i];
2288 		fp = fde->fde_file;
2289 		if (fp != NULL) {
2290 			fdefree_last(fde);
2291 			if (needclose)
2292 				(void) closef(fp, td);
2293 			else
2294 				fdrop(fp, td);
2295 		}
2296 	}
2297 
2298 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2299 		free(fdp->fd_map, M_FILEDESC);
2300 	if (fdp->fd_nfiles > NDFILE)
2301 		free(fdp->fd_files, M_FILEDESC);
2302 
2303 	fdp0 = (struct filedesc0 *)fdp;
2304 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2305 		free(ft->ft_table, M_FILEDESC);
2306 
2307 	fddrop(fdp);
2308 }
2309 
2310 void
2311 fdescfree(struct thread *td)
2312 {
2313 	struct proc *p;
2314 	struct filedesc *fdp;
2315 	struct pwd *pwd;
2316 
2317 	p = td->td_proc;
2318 	fdp = p->p_fd;
2319 	MPASS(fdp != NULL);
2320 
2321 #ifdef RACCT
2322 	if (RACCT_ENABLED())
2323 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2324 #endif
2325 
2326 	if (p->p_fdtol != NULL)
2327 		fdclearlocks(td);
2328 
2329 	PROC_LOCK(p);
2330 	p->p_fd = NULL;
2331 	PROC_UNLOCK(p);
2332 
2333 	if (refcount_release(&fdp->fd_refcnt) == 0)
2334 		return;
2335 
2336 	FILEDESC_XLOCK(fdp);
2337 	pwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
2338 	pwd_set(fdp, NULL);
2339 	FILEDESC_XUNLOCK(fdp);
2340 
2341 	pwd_drop(pwd);
2342 
2343 	fdescfree_fds(td, fdp, 1);
2344 }
2345 
2346 void
2347 fdescfree_remapped(struct filedesc *fdp)
2348 {
2349 
2350 	pwd_drop(smr_serialized_load(&fdp->fd_pwd, true));
2351 	fdescfree_fds(curthread, fdp, 0);
2352 }
2353 
2354 /*
2355  * For setugid programs, we don't want to people to use that setugidness
2356  * to generate error messages which write to a file which otherwise would
2357  * otherwise be off-limits to the process.  We check for filesystems where
2358  * the vnode can change out from under us after execve (like [lin]procfs).
2359  *
2360  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2361  * sufficient.  We also don't check for setugidness since we know we are.
2362  */
2363 static bool
2364 is_unsafe(struct file *fp)
2365 {
2366 	struct vnode *vp;
2367 
2368 	if (fp->f_type != DTYPE_VNODE)
2369 		return (false);
2370 
2371 	vp = fp->f_vnode;
2372 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2373 }
2374 
2375 /*
2376  * Make this setguid thing safe, if at all possible.
2377  */
2378 void
2379 fdsetugidsafety(struct thread *td)
2380 {
2381 	struct filedesc *fdp;
2382 	struct file *fp;
2383 	int i;
2384 
2385 	fdp = td->td_proc->p_fd;
2386 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2387 	MPASS(fdp->fd_nfiles >= 3);
2388 	for (i = 0; i <= 2; i++) {
2389 		fp = fdp->fd_ofiles[i].fde_file;
2390 		if (fp != NULL && is_unsafe(fp)) {
2391 			FILEDESC_XLOCK(fdp);
2392 			knote_fdclose(td, i);
2393 			/*
2394 			 * NULL-out descriptor prior to close to avoid
2395 			 * a race while close blocks.
2396 			 */
2397 			fdfree(fdp, i);
2398 			FILEDESC_XUNLOCK(fdp);
2399 			(void) closef(fp, td);
2400 		}
2401 	}
2402 }
2403 
2404 /*
2405  * If a specific file object occupies a specific file descriptor, close the
2406  * file descriptor entry and drop a reference on the file object.  This is a
2407  * convenience function to handle a subsequent error in a function that calls
2408  * falloc() that handles the race that another thread might have closed the
2409  * file descriptor out from under the thread creating the file object.
2410  */
2411 void
2412 fdclose(struct thread *td, struct file *fp, int idx)
2413 {
2414 	struct filedesc *fdp = td->td_proc->p_fd;
2415 
2416 	FILEDESC_XLOCK(fdp);
2417 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2418 		fdfree(fdp, idx);
2419 		FILEDESC_XUNLOCK(fdp);
2420 		fdrop(fp, td);
2421 	} else
2422 		FILEDESC_XUNLOCK(fdp);
2423 }
2424 
2425 /*
2426  * Close any files on exec?
2427  */
2428 void
2429 fdcloseexec(struct thread *td)
2430 {
2431 	struct filedesc *fdp;
2432 	struct filedescent *fde;
2433 	struct file *fp;
2434 	int i;
2435 
2436 	fdp = td->td_proc->p_fd;
2437 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2438 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2439 		fde = &fdp->fd_ofiles[i];
2440 		fp = fde->fde_file;
2441 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2442 		    (fde->fde_flags & UF_EXCLOSE))) {
2443 			FILEDESC_XLOCK(fdp);
2444 			fdfree(fdp, i);
2445 			(void) closefp(fdp, i, fp, td, 0);
2446 			FILEDESC_UNLOCK_ASSERT(fdp);
2447 		}
2448 	}
2449 }
2450 
2451 /*
2452  * It is unsafe for set[ug]id processes to be started with file
2453  * descriptors 0..2 closed, as these descriptors are given implicit
2454  * significance in the Standard C library.  fdcheckstd() will create a
2455  * descriptor referencing /dev/null for each of stdin, stdout, and
2456  * stderr that is not already open.
2457  */
2458 int
2459 fdcheckstd(struct thread *td)
2460 {
2461 	struct filedesc *fdp;
2462 	register_t save;
2463 	int i, error, devnull;
2464 
2465 	fdp = td->td_proc->p_fd;
2466 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2467 	MPASS(fdp->fd_nfiles >= 3);
2468 	devnull = -1;
2469 	for (i = 0; i <= 2; i++) {
2470 		if (fdp->fd_ofiles[i].fde_file != NULL)
2471 			continue;
2472 
2473 		save = td->td_retval[0];
2474 		if (devnull != -1) {
2475 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2476 		} else {
2477 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2478 			    UIO_SYSSPACE, O_RDWR, 0);
2479 			if (error == 0) {
2480 				devnull = td->td_retval[0];
2481 				KASSERT(devnull == i, ("we didn't get our fd"));
2482 			}
2483 		}
2484 		td->td_retval[0] = save;
2485 		if (error != 0)
2486 			return (error);
2487 	}
2488 	return (0);
2489 }
2490 
2491 /*
2492  * Internal form of close.  Decrement reference count on file structure.
2493  * Note: td may be NULL when closing a file that was being passed in a
2494  * message.
2495  */
2496 int
2497 closef(struct file *fp, struct thread *td)
2498 {
2499 	struct vnode *vp;
2500 	struct flock lf;
2501 	struct filedesc_to_leader *fdtol;
2502 	struct filedesc *fdp;
2503 
2504 	/*
2505 	 * POSIX record locking dictates that any close releases ALL
2506 	 * locks owned by this process.  This is handled by setting
2507 	 * a flag in the unlock to free ONLY locks obeying POSIX
2508 	 * semantics, and not to free BSD-style file locks.
2509 	 * If the descriptor was in a message, POSIX-style locks
2510 	 * aren't passed with the descriptor, and the thread pointer
2511 	 * will be NULL.  Callers should be careful only to pass a
2512 	 * NULL thread pointer when there really is no owning
2513 	 * context that might have locks, or the locks will be
2514 	 * leaked.
2515 	 */
2516 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2517 		vp = fp->f_vnode;
2518 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2519 			lf.l_whence = SEEK_SET;
2520 			lf.l_start = 0;
2521 			lf.l_len = 0;
2522 			lf.l_type = F_UNLCK;
2523 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2524 			    F_UNLCK, &lf, F_POSIX);
2525 		}
2526 		fdtol = td->td_proc->p_fdtol;
2527 		if (fdtol != NULL) {
2528 			/*
2529 			 * Handle special case where file descriptor table is
2530 			 * shared between multiple process leaders.
2531 			 */
2532 			fdp = td->td_proc->p_fd;
2533 			FILEDESC_XLOCK(fdp);
2534 			for (fdtol = fdtol->fdl_next;
2535 			    fdtol != td->td_proc->p_fdtol;
2536 			    fdtol = fdtol->fdl_next) {
2537 				if ((fdtol->fdl_leader->p_flag &
2538 				    P_ADVLOCK) == 0)
2539 					continue;
2540 				fdtol->fdl_holdcount++;
2541 				FILEDESC_XUNLOCK(fdp);
2542 				lf.l_whence = SEEK_SET;
2543 				lf.l_start = 0;
2544 				lf.l_len = 0;
2545 				lf.l_type = F_UNLCK;
2546 				vp = fp->f_vnode;
2547 				(void) VOP_ADVLOCK(vp,
2548 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2549 				    F_POSIX);
2550 				FILEDESC_XLOCK(fdp);
2551 				fdtol->fdl_holdcount--;
2552 				if (fdtol->fdl_holdcount == 0 &&
2553 				    fdtol->fdl_wakeup != 0) {
2554 					fdtol->fdl_wakeup = 0;
2555 					wakeup(fdtol);
2556 				}
2557 			}
2558 			FILEDESC_XUNLOCK(fdp);
2559 		}
2560 	}
2561 	return (fdrop(fp, td));
2562 }
2563 
2564 /*
2565  * Initialize the file pointer with the specified properties.
2566  *
2567  * The ops are set with release semantics to be certain that the flags, type,
2568  * and data are visible when ops is.  This is to prevent ops methods from being
2569  * called with bad data.
2570  */
2571 void
2572 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2573 {
2574 	fp->f_data = data;
2575 	fp->f_flag = flag;
2576 	fp->f_type = type;
2577 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2578 }
2579 
2580 int
2581 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2582     struct file **fpp, struct filecaps *havecapsp)
2583 {
2584 	struct filedescent *fde;
2585 	int error;
2586 
2587 	FILEDESC_LOCK_ASSERT(fdp);
2588 
2589 	fde = fdeget_locked(fdp, fd);
2590 	if (fde == NULL) {
2591 		error = EBADF;
2592 		goto out;
2593 	}
2594 
2595 #ifdef CAPABILITIES
2596 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2597 	if (error != 0)
2598 		goto out;
2599 #endif
2600 
2601 	if (havecapsp != NULL)
2602 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2603 
2604 	*fpp = fde->fde_file;
2605 
2606 	error = 0;
2607 out:
2608 	return (error);
2609 }
2610 
2611 int
2612 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2613     struct file **fpp, struct filecaps *havecapsp)
2614 {
2615 	struct filedesc *fdp = td->td_proc->p_fd;
2616 	int error;
2617 #ifndef CAPABILITIES
2618 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2619 	if (havecapsp != NULL && error == 0)
2620 		filecaps_fill(havecapsp);
2621 #else
2622 	struct file *fp;
2623 	seqc_t seq;
2624 
2625 	*fpp = NULL;
2626 	for (;;) {
2627 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2628 		if (error != 0)
2629 			return (error);
2630 
2631 		if (havecapsp != NULL) {
2632 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2633 			    havecapsp, false)) {
2634 				fdrop(fp, td);
2635 				goto get_locked;
2636 			}
2637 		}
2638 
2639 		if (!fd_modified(fdp, fd, seq))
2640 			break;
2641 		fdrop(fp, td);
2642 	}
2643 
2644 	*fpp = fp;
2645 	return (0);
2646 
2647 get_locked:
2648 	FILEDESC_SLOCK(fdp);
2649 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2650 	if (error == 0 && !fhold(*fpp))
2651 		error = EBADF;
2652 	FILEDESC_SUNLOCK(fdp);
2653 #endif
2654 	return (error);
2655 }
2656 
2657 int
2658 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2659     struct file **fpp, seqc_t *seqp)
2660 {
2661 #ifdef CAPABILITIES
2662 	const struct filedescent *fde;
2663 #endif
2664 	const struct fdescenttbl *fdt;
2665 	struct file *fp;
2666 #ifdef CAPABILITIES
2667 	seqc_t seq;
2668 	cap_rights_t haverights;
2669 	int error;
2670 #endif
2671 
2672 	fdt = fdp->fd_files;
2673 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2674 		return (EBADF);
2675 	/*
2676 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2677 	 * never raising a refcount above 0.  To accomplish this we have
2678 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2679 	 * must be re-verified once we acquire a reference to be certain
2680 	 * that the identity is still correct and we did not lose a race
2681 	 * due to preemption.
2682 	 */
2683 	for (;;) {
2684 #ifdef CAPABILITIES
2685 		seq = seqc_read(fd_seqc(fdt, fd));
2686 		fde = &fdt->fdt_ofiles[fd];
2687 		haverights = *cap_rights_fde_inline(fde);
2688 		fp = fde->fde_file;
2689 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2690 			continue;
2691 #else
2692 		fp = fdt->fdt_ofiles[fd].fde_file;
2693 #endif
2694 		if (fp == NULL)
2695 			return (EBADF);
2696 #ifdef CAPABILITIES
2697 		error = cap_check_inline(&haverights, needrightsp);
2698 		if (error != 0)
2699 			return (error);
2700 #endif
2701 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
2702 			/*
2703 			 * The count was found either saturated or zero.
2704 			 * This re-read is not any more racy than using the
2705 			 * return value from fcmpset.
2706 			 */
2707 			if (fp->f_count != 0)
2708 				return (EBADF);
2709 			/*
2710 			 * Force a reload. Other thread could reallocate the
2711 			 * table before this fd was closed, so it is possible
2712 			 * that there is a stale fp pointer in cached version.
2713 			 */
2714 			fdt = atomic_load_ptr(&fdp->fd_files);
2715 			continue;
2716 		}
2717 		/*
2718 		 * Use an acquire barrier to force re-reading of fdt so it is
2719 		 * refreshed for verification.
2720 		 */
2721 		atomic_thread_fence_acq();
2722 		fdt = fdp->fd_files;
2723 #ifdef	CAPABILITIES
2724 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2725 #else
2726 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2727 #endif
2728 			break;
2729 		fdrop(fp, curthread);
2730 	}
2731 	*fpp = fp;
2732 	if (seqp != NULL) {
2733 #ifdef CAPABILITIES
2734 		*seqp = seq;
2735 #endif
2736 	}
2737 	return (0);
2738 }
2739 
2740 /*
2741  * See the comments in fget_unlocked_seq for an explanation of how this works.
2742  *
2743  * This is a simplified variant which bails out to the aforementioned routine
2744  * if anything goes wrong. In practice this only happens when userspace is
2745  * racing with itself.
2746  */
2747 int
2748 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2749     struct file **fpp)
2750 {
2751 #ifdef CAPABILITIES
2752 	const struct filedescent *fde;
2753 #endif
2754 	const struct fdescenttbl *fdt;
2755 	struct file *fp;
2756 #ifdef CAPABILITIES
2757 	seqc_t seq;
2758 	const cap_rights_t *haverights;
2759 #endif
2760 
2761 	fdt = fdp->fd_files;
2762 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2763 		return (EBADF);
2764 #ifdef CAPABILITIES
2765 	seq = seqc_read_any(fd_seqc(fdt, fd));
2766 	if (__predict_false(seqc_in_modify(seq)))
2767 		goto out_fallback;
2768 	fde = &fdt->fdt_ofiles[fd];
2769 	haverights = cap_rights_fde_inline(fde);
2770 	fp = fde->fde_file;
2771 #else
2772 	fp = fdt->fdt_ofiles[fd].fde_file;
2773 #endif
2774 	if (__predict_false(fp == NULL))
2775 		goto out_fallback;
2776 #ifdef CAPABILITIES
2777 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
2778 		goto out_fallback;
2779 #endif
2780 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
2781 		goto out_fallback;
2782 
2783 	/*
2784 	 * Use an acquire barrier to force re-reading of fdt so it is
2785 	 * refreshed for verification.
2786 	 */
2787 	atomic_thread_fence_acq();
2788 	fdt = fdp->fd_files;
2789 #ifdef	CAPABILITIES
2790 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2791 #else
2792 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2793 #endif
2794 		goto out_fdrop;
2795 	*fpp = fp;
2796 	return (0);
2797 out_fdrop:
2798 	fdrop(fp, curthread);
2799 out_fallback:
2800 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
2801 }
2802 
2803 /*
2804  * Extract the file pointer associated with the specified descriptor for the
2805  * current user process.
2806  *
2807  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2808  * returned.
2809  *
2810  * File's rights will be checked against the capability rights mask.
2811  *
2812  * If an error occurred the non-zero error is returned and *fpp is set to
2813  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2814  * responsible for fdrop().
2815  */
2816 static __inline int
2817 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2818     cap_rights_t *needrightsp)
2819 {
2820 	struct filedesc *fdp;
2821 	struct file *fp;
2822 	int error;
2823 
2824 	*fpp = NULL;
2825 	fdp = td->td_proc->p_fd;
2826 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
2827 	if (__predict_false(error != 0))
2828 		return (error);
2829 	if (__predict_false(fp->f_ops == &badfileops)) {
2830 		fdrop(fp, td);
2831 		return (EBADF);
2832 	}
2833 
2834 	/*
2835 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2836 	 */
2837 	error = 0;
2838 	switch (flags) {
2839 	case FREAD:
2840 	case FWRITE:
2841 		if ((fp->f_flag & flags) == 0)
2842 			error = EBADF;
2843 		break;
2844 	case FEXEC:
2845 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2846 		    ((fp->f_flag & FWRITE) != 0))
2847 			error = EBADF;
2848 		break;
2849 	case 0:
2850 		break;
2851 	default:
2852 		KASSERT(0, ("wrong flags"));
2853 	}
2854 
2855 	if (error != 0) {
2856 		fdrop(fp, td);
2857 		return (error);
2858 	}
2859 
2860 	*fpp = fp;
2861 	return (0);
2862 }
2863 
2864 int
2865 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2866 {
2867 
2868 	return (_fget(td, fd, fpp, 0, rightsp));
2869 }
2870 
2871 int
2872 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
2873     struct file **fpp)
2874 {
2875 	int error;
2876 #ifndef CAPABILITIES
2877 	error = _fget(td, fd, fpp, 0, rightsp);
2878 	if (maxprotp != NULL)
2879 		*maxprotp = VM_PROT_ALL;
2880 	return (error);
2881 #else
2882 	cap_rights_t fdrights;
2883 	struct filedesc *fdp;
2884 	struct file *fp;
2885 	seqc_t seq;
2886 
2887 	*fpp = NULL;
2888 	fdp = td->td_proc->p_fd;
2889 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2890 	for (;;) {
2891 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
2892 		if (__predict_false(error != 0))
2893 			return (error);
2894 		if (__predict_false(fp->f_ops == &badfileops)) {
2895 			fdrop(fp, td);
2896 			return (EBADF);
2897 		}
2898 		if (maxprotp != NULL)
2899 			fdrights = *cap_rights(fdp, fd);
2900 		if (!fd_modified(fdp, fd, seq))
2901 			break;
2902 		fdrop(fp, td);
2903 	}
2904 
2905 	/*
2906 	 * If requested, convert capability rights to access flags.
2907 	 */
2908 	if (maxprotp != NULL)
2909 		*maxprotp = cap_rights_to_vmprot(&fdrights);
2910 	*fpp = fp;
2911 	return (0);
2912 #endif
2913 }
2914 
2915 int
2916 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2917 {
2918 
2919 	return (_fget(td, fd, fpp, FREAD, rightsp));
2920 }
2921 
2922 int
2923 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2924 {
2925 
2926 	return (_fget(td, fd, fpp, FWRITE, rightsp));
2927 }
2928 
2929 int
2930 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2931     struct file **fpp)
2932 {
2933 	struct filedesc *fdp = td->td_proc->p_fd;
2934 #ifndef CAPABILITIES
2935 	return (fget_unlocked(fdp, fd, rightsp, fpp));
2936 #else
2937 	struct file *fp;
2938 	int error;
2939 	seqc_t seq;
2940 
2941 	*fpp = NULL;
2942 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2943 	for (;;) {
2944 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
2945 		if (error != 0)
2946 			return (error);
2947 		error = cap_fcntl_check(fdp, fd, needfcntl);
2948 		if (!fd_modified(fdp, fd, seq))
2949 			break;
2950 		fdrop(fp, td);
2951 	}
2952 	if (error != 0) {
2953 		fdrop(fp, td);
2954 		return (error);
2955 	}
2956 	*fpp = fp;
2957 	return (0);
2958 #endif
2959 }
2960 
2961 /*
2962  * Like fget() but loads the underlying vnode, or returns an error if the
2963  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2964  * never have VM objects.  The returned vnode will be vref()'d.
2965  *
2966  * XXX: what about the unused flags ?
2967  */
2968 static __inline int
2969 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2970     struct vnode **vpp)
2971 {
2972 	struct file *fp;
2973 	int error;
2974 
2975 	*vpp = NULL;
2976 	error = _fget(td, fd, &fp, flags, needrightsp);
2977 	if (error != 0)
2978 		return (error);
2979 	if (fp->f_vnode == NULL) {
2980 		error = EINVAL;
2981 	} else {
2982 		*vpp = fp->f_vnode;
2983 		vrefact(*vpp);
2984 	}
2985 	fdrop(fp, td);
2986 
2987 	return (error);
2988 }
2989 
2990 int
2991 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2992 {
2993 
2994 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2995 }
2996 
2997 int
2998 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2999     struct filecaps *havecaps, struct vnode **vpp)
3000 {
3001 	struct filecaps caps;
3002 	struct file *fp;
3003 	int error;
3004 
3005 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3006 	if (error != 0)
3007 		return (error);
3008 	if (fp->f_ops == &badfileops) {
3009 		error = EBADF;
3010 		goto out;
3011 	}
3012 	if (fp->f_vnode == NULL) {
3013 		error = EINVAL;
3014 		goto out;
3015 	}
3016 
3017 	*havecaps = caps;
3018 	*vpp = fp->f_vnode;
3019 	vrefact(*vpp);
3020 	fdrop(fp, td);
3021 
3022 	return (0);
3023 out:
3024 	filecaps_free(&caps);
3025 	fdrop(fp, td);
3026 	return (error);
3027 }
3028 
3029 int
3030 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3031 {
3032 
3033 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3034 }
3035 
3036 int
3037 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3038 {
3039 
3040 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3041 }
3042 
3043 #ifdef notyet
3044 int
3045 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3046     struct vnode **vpp)
3047 {
3048 
3049 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3050 }
3051 #endif
3052 
3053 /*
3054  * Handle the last reference to a file being closed.
3055  *
3056  * Without the noinline attribute clang keeps inlining the func thorough this
3057  * file when fdrop is used.
3058  */
3059 int __noinline
3060 _fdrop(struct file *fp, struct thread *td)
3061 {
3062 	int error;
3063 
3064 	if (fp->f_count != 0)
3065 		panic("fdrop: count %d", fp->f_count);
3066 	error = fo_close(fp, td);
3067 	atomic_subtract_int(&openfiles, 1);
3068 	crfree(fp->f_cred);
3069 	free(fp->f_advice, M_FADVISE);
3070 	uma_zfree(file_zone, fp);
3071 
3072 	return (error);
3073 }
3074 
3075 /*
3076  * Apply an advisory lock on a file descriptor.
3077  *
3078  * Just attempt to get a record lock of the requested type on the entire file
3079  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3080  */
3081 #ifndef _SYS_SYSPROTO_H_
3082 struct flock_args {
3083 	int	fd;
3084 	int	how;
3085 };
3086 #endif
3087 /* ARGSUSED */
3088 int
3089 sys_flock(struct thread *td, struct flock_args *uap)
3090 {
3091 	struct file *fp;
3092 	struct vnode *vp;
3093 	struct flock lf;
3094 	int error;
3095 
3096 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3097 	if (error != 0)
3098 		return (error);
3099 	if (fp->f_type != DTYPE_VNODE) {
3100 		fdrop(fp, td);
3101 		return (EOPNOTSUPP);
3102 	}
3103 
3104 	vp = fp->f_vnode;
3105 	lf.l_whence = SEEK_SET;
3106 	lf.l_start = 0;
3107 	lf.l_len = 0;
3108 	if (uap->how & LOCK_UN) {
3109 		lf.l_type = F_UNLCK;
3110 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3111 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3112 		goto done2;
3113 	}
3114 	if (uap->how & LOCK_EX)
3115 		lf.l_type = F_WRLCK;
3116 	else if (uap->how & LOCK_SH)
3117 		lf.l_type = F_RDLCK;
3118 	else {
3119 		error = EBADF;
3120 		goto done2;
3121 	}
3122 	atomic_set_int(&fp->f_flag, FHASLOCK);
3123 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3124 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3125 done2:
3126 	fdrop(fp, td);
3127 	return (error);
3128 }
3129 /*
3130  * Duplicate the specified descriptor to a free descriptor.
3131  */
3132 int
3133 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3134     int openerror, int *indxp)
3135 {
3136 	struct filedescent *newfde, *oldfde;
3137 	struct file *fp;
3138 	u_long *ioctls;
3139 	int error, indx;
3140 
3141 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3142 	    ("unexpected error %d in %s", openerror, __func__));
3143 
3144 	/*
3145 	 * If the to-be-dup'd fd number is greater than the allowed number
3146 	 * of file descriptors, or the fd to be dup'd has already been
3147 	 * closed, then reject.
3148 	 */
3149 	FILEDESC_XLOCK(fdp);
3150 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3151 		FILEDESC_XUNLOCK(fdp);
3152 		return (EBADF);
3153 	}
3154 
3155 	error = fdalloc(td, 0, &indx);
3156 	if (error != 0) {
3157 		FILEDESC_XUNLOCK(fdp);
3158 		return (error);
3159 	}
3160 
3161 	/*
3162 	 * There are two cases of interest here.
3163 	 *
3164 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3165 	 *
3166 	 * For ENXIO steal away the file structure from (dfd) and store it in
3167 	 * (indx).  (dfd) is effectively closed by this operation.
3168 	 */
3169 	switch (openerror) {
3170 	case ENODEV:
3171 		/*
3172 		 * Check that the mode the file is being opened for is a
3173 		 * subset of the mode of the existing descriptor.
3174 		 */
3175 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3176 			fdunused(fdp, indx);
3177 			FILEDESC_XUNLOCK(fdp);
3178 			return (EACCES);
3179 		}
3180 		if (!fhold(fp)) {
3181 			fdunused(fdp, indx);
3182 			FILEDESC_XUNLOCK(fdp);
3183 			return (EBADF);
3184 		}
3185 		newfde = &fdp->fd_ofiles[indx];
3186 		oldfde = &fdp->fd_ofiles[dfd];
3187 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3188 #ifdef CAPABILITIES
3189 		seqc_write_begin(&newfde->fde_seqc);
3190 #endif
3191 		memcpy(newfde, oldfde, fde_change_size);
3192 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3193 		    ioctls);
3194 #ifdef CAPABILITIES
3195 		seqc_write_end(&newfde->fde_seqc);
3196 #endif
3197 		break;
3198 	case ENXIO:
3199 		/*
3200 		 * Steal away the file pointer from dfd and stuff it into indx.
3201 		 */
3202 		newfde = &fdp->fd_ofiles[indx];
3203 		oldfde = &fdp->fd_ofiles[dfd];
3204 #ifdef CAPABILITIES
3205 		seqc_write_begin(&newfde->fde_seqc);
3206 #endif
3207 		memcpy(newfde, oldfde, fde_change_size);
3208 		oldfde->fde_file = NULL;
3209 		fdunused(fdp, dfd);
3210 #ifdef CAPABILITIES
3211 		seqc_write_end(&newfde->fde_seqc);
3212 #endif
3213 		break;
3214 	}
3215 	FILEDESC_XUNLOCK(fdp);
3216 	*indxp = indx;
3217 	return (0);
3218 }
3219 
3220 /*
3221  * This sysctl determines if we will allow a process to chroot(2) if it
3222  * has a directory open:
3223  *	0: disallowed for all processes.
3224  *	1: allowed for processes that were not already chroot(2)'ed.
3225  *	2: allowed for all processes.
3226  */
3227 
3228 static int chroot_allow_open_directories = 1;
3229 
3230 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3231     &chroot_allow_open_directories, 0,
3232     "Allow a process to chroot(2) if it has a directory open");
3233 
3234 /*
3235  * Helper function for raised chroot(2) security function:  Refuse if
3236  * any filedescriptors are open directories.
3237  */
3238 static int
3239 chroot_refuse_vdir_fds(struct filedesc *fdp)
3240 {
3241 	struct vnode *vp;
3242 	struct file *fp;
3243 	int fd;
3244 
3245 	FILEDESC_LOCK_ASSERT(fdp);
3246 
3247 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3248 		fp = fget_locked(fdp, fd);
3249 		if (fp == NULL)
3250 			continue;
3251 		if (fp->f_type == DTYPE_VNODE) {
3252 			vp = fp->f_vnode;
3253 			if (vp->v_type == VDIR)
3254 				return (EPERM);
3255 		}
3256 	}
3257 	return (0);
3258 }
3259 
3260 static void
3261 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3262 {
3263 
3264 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3265 		vrefact(oldpwd->pwd_cdir);
3266 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3267 	}
3268 
3269 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3270 		vrefact(oldpwd->pwd_rdir);
3271 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3272 	}
3273 
3274 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3275 		vrefact(oldpwd->pwd_jdir);
3276 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3277 	}
3278 }
3279 
3280 struct pwd *
3281 pwd_hold_filedesc(struct filedesc *fdp)
3282 {
3283 	struct pwd *pwd;
3284 
3285 	FILEDESC_LOCK_ASSERT(fdp);
3286 	pwd = FILEDESC_LOCKED_LOAD_PWD(fdp);
3287 	if (pwd != NULL)
3288 		refcount_acquire(&pwd->pwd_refcount);
3289 	return (pwd);
3290 }
3291 
3292 struct pwd *
3293 pwd_hold(struct thread *td)
3294 {
3295 	struct filedesc *fdp;
3296 	struct pwd *pwd;
3297 
3298 	fdp = td->td_proc->p_fd;
3299 
3300 	smr_enter(pwd_smr);
3301 	for (;;) {
3302 		pwd = smr_entered_load(&fdp->fd_pwd, pwd_smr);
3303 		MPASS(pwd != NULL);
3304 		if (refcount_acquire_if_not_zero(&pwd->pwd_refcount))
3305 			break;
3306 	}
3307 	smr_exit(pwd_smr);
3308 	return (pwd);
3309 }
3310 
3311 static struct pwd *
3312 pwd_alloc(void)
3313 {
3314 	struct pwd *pwd;
3315 
3316 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3317 	bzero(pwd, sizeof(*pwd));
3318 	refcount_init(&pwd->pwd_refcount, 1);
3319 	return (pwd);
3320 }
3321 
3322 void
3323 pwd_drop(struct pwd *pwd)
3324 {
3325 
3326 	if (!refcount_release(&pwd->pwd_refcount))
3327 		return;
3328 
3329 	if (pwd->pwd_cdir != NULL)
3330 		vrele(pwd->pwd_cdir);
3331 	if (pwd->pwd_rdir != NULL)
3332 		vrele(pwd->pwd_rdir);
3333 	if (pwd->pwd_jdir != NULL)
3334 		vrele(pwd->pwd_jdir);
3335 	uma_zfree_smr(pwd_zone, pwd);
3336 }
3337 
3338 /*
3339 * Common routine for kern_chroot() and jail_attach().  The caller is
3340 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3341 * authorize this operation.
3342 */
3343 int
3344 pwd_chroot(struct thread *td, struct vnode *vp)
3345 {
3346 	struct filedesc *fdp;
3347 	struct pwd *newpwd, *oldpwd;
3348 	int error;
3349 
3350 	fdp = td->td_proc->p_fd;
3351 	newpwd = pwd_alloc();
3352 	FILEDESC_XLOCK(fdp);
3353 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3354 	if (chroot_allow_open_directories == 0 ||
3355 	    (chroot_allow_open_directories == 1 &&
3356 	    oldpwd->pwd_rdir != rootvnode)) {
3357 		error = chroot_refuse_vdir_fds(fdp);
3358 		if (error != 0) {
3359 			FILEDESC_XUNLOCK(fdp);
3360 			pwd_drop(newpwd);
3361 			return (error);
3362 		}
3363 	}
3364 
3365 	vrefact(vp);
3366 	newpwd->pwd_rdir = vp;
3367 	if (oldpwd->pwd_jdir == NULL) {
3368 		vrefact(vp);
3369 		newpwd->pwd_jdir = vp;
3370 	}
3371 	pwd_fill(oldpwd, newpwd);
3372 	pwd_set(fdp, newpwd);
3373 	FILEDESC_XUNLOCK(fdp);
3374 	pwd_drop(oldpwd);
3375 	return (0);
3376 }
3377 
3378 void
3379 pwd_chdir(struct thread *td, struct vnode *vp)
3380 {
3381 	struct filedesc *fdp;
3382 	struct pwd *newpwd, *oldpwd;
3383 
3384 	VNPASS(vp->v_usecount > 0, vp);
3385 
3386 	newpwd = pwd_alloc();
3387 	fdp = td->td_proc->p_fd;
3388 	FILEDESC_XLOCK(fdp);
3389 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3390 	newpwd->pwd_cdir = vp;
3391 	pwd_fill(oldpwd, newpwd);
3392 	pwd_set(fdp, newpwd);
3393 	FILEDESC_XUNLOCK(fdp);
3394 	pwd_drop(oldpwd);
3395 }
3396 
3397 void
3398 pwd_ensure_dirs(void)
3399 {
3400 	struct filedesc *fdp;
3401 	struct pwd *oldpwd, *newpwd;
3402 
3403 	fdp = curproc->p_fd;
3404 	FILEDESC_XLOCK(fdp);
3405 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3406 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3407 		FILEDESC_XUNLOCK(fdp);
3408 		return;
3409 	}
3410 	FILEDESC_XUNLOCK(fdp);
3411 
3412 	newpwd = pwd_alloc();
3413 	FILEDESC_XLOCK(fdp);
3414 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3415 	pwd_fill(oldpwd, newpwd);
3416 	if (newpwd->pwd_cdir == NULL) {
3417 		vrefact(rootvnode);
3418 		newpwd->pwd_cdir = rootvnode;
3419 	}
3420 	if (newpwd->pwd_rdir == NULL) {
3421 		vrefact(rootvnode);
3422 		newpwd->pwd_rdir = rootvnode;
3423 	}
3424 	pwd_set(fdp, newpwd);
3425 	FILEDESC_XUNLOCK(fdp);
3426 	pwd_drop(oldpwd);
3427 }
3428 
3429 /*
3430  * Scan all active processes and prisons to see if any of them have a current
3431  * or root directory of `olddp'. If so, replace them with the new mount point.
3432  */
3433 void
3434 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3435 {
3436 	struct filedesc *fdp;
3437 	struct pwd *newpwd, *oldpwd;
3438 	struct prison *pr;
3439 	struct proc *p;
3440 	int nrele;
3441 
3442 	if (vrefcnt(olddp) == 1)
3443 		return;
3444 	nrele = 0;
3445 	newpwd = pwd_alloc();
3446 	sx_slock(&allproc_lock);
3447 	FOREACH_PROC_IN_SYSTEM(p) {
3448 		PROC_LOCK(p);
3449 		fdp = fdhold(p);
3450 		PROC_UNLOCK(p);
3451 		if (fdp == NULL)
3452 			continue;
3453 		FILEDESC_XLOCK(fdp);
3454 		oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3455 		if (oldpwd == NULL ||
3456 		    (oldpwd->pwd_cdir != olddp &&
3457 		    oldpwd->pwd_rdir != olddp &&
3458 		    oldpwd->pwd_jdir != olddp)) {
3459 			FILEDESC_XUNLOCK(fdp);
3460 			fddrop(fdp);
3461 			continue;
3462 		}
3463 		if (oldpwd->pwd_cdir == olddp) {
3464 			vrefact(newdp);
3465 			newpwd->pwd_cdir = newdp;
3466 		}
3467 		if (oldpwd->pwd_rdir == olddp) {
3468 			vrefact(newdp);
3469 			newpwd->pwd_rdir = newdp;
3470 		}
3471 		if (oldpwd->pwd_jdir == olddp) {
3472 			vrefact(newdp);
3473 			newpwd->pwd_jdir = newdp;
3474 		}
3475 		pwd_fill(oldpwd, newpwd);
3476 		pwd_set(fdp, newpwd);
3477 		FILEDESC_XUNLOCK(fdp);
3478 		pwd_drop(oldpwd);
3479 		fddrop(fdp);
3480 		newpwd = pwd_alloc();
3481 	}
3482 	sx_sunlock(&allproc_lock);
3483 	pwd_drop(newpwd);
3484 	if (rootvnode == olddp) {
3485 		vrefact(newdp);
3486 		rootvnode = newdp;
3487 		nrele++;
3488 	}
3489 	mtx_lock(&prison0.pr_mtx);
3490 	if (prison0.pr_root == olddp) {
3491 		vrefact(newdp);
3492 		prison0.pr_root = newdp;
3493 		nrele++;
3494 	}
3495 	mtx_unlock(&prison0.pr_mtx);
3496 	sx_slock(&allprison_lock);
3497 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3498 		mtx_lock(&pr->pr_mtx);
3499 		if (pr->pr_root == olddp) {
3500 			vrefact(newdp);
3501 			pr->pr_root = newdp;
3502 			nrele++;
3503 		}
3504 		mtx_unlock(&pr->pr_mtx);
3505 	}
3506 	sx_sunlock(&allprison_lock);
3507 	while (nrele--)
3508 		vrele(olddp);
3509 }
3510 
3511 struct filedesc_to_leader *
3512 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3513 {
3514 	struct filedesc_to_leader *fdtol;
3515 
3516 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3517 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3518 	fdtol->fdl_refcount = 1;
3519 	fdtol->fdl_holdcount = 0;
3520 	fdtol->fdl_wakeup = 0;
3521 	fdtol->fdl_leader = leader;
3522 	if (old != NULL) {
3523 		FILEDESC_XLOCK(fdp);
3524 		fdtol->fdl_next = old->fdl_next;
3525 		fdtol->fdl_prev = old;
3526 		old->fdl_next = fdtol;
3527 		fdtol->fdl_next->fdl_prev = fdtol;
3528 		FILEDESC_XUNLOCK(fdp);
3529 	} else {
3530 		fdtol->fdl_next = fdtol;
3531 		fdtol->fdl_prev = fdtol;
3532 	}
3533 	return (fdtol);
3534 }
3535 
3536 static int
3537 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3538 {
3539 	struct filedesc *fdp;
3540 	int i, count, slots;
3541 
3542 	if (*(int *)arg1 != 0)
3543 		return (EINVAL);
3544 
3545 	fdp = curproc->p_fd;
3546 	count = 0;
3547 	FILEDESC_SLOCK(fdp);
3548 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3549 	for (i = 0; i < slots; i++)
3550 		count += bitcountl(fdp->fd_map[i]);
3551 	FILEDESC_SUNLOCK(fdp);
3552 
3553 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3554 }
3555 
3556 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3557     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3558     "Number of open file descriptors");
3559 
3560 /*
3561  * Get file structures globally.
3562  */
3563 static int
3564 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3565 {
3566 	struct xfile xf;
3567 	struct filedesc *fdp;
3568 	struct file *fp;
3569 	struct proc *p;
3570 	int error, n;
3571 
3572 	error = sysctl_wire_old_buffer(req, 0);
3573 	if (error != 0)
3574 		return (error);
3575 	if (req->oldptr == NULL) {
3576 		n = 0;
3577 		sx_slock(&allproc_lock);
3578 		FOREACH_PROC_IN_SYSTEM(p) {
3579 			PROC_LOCK(p);
3580 			if (p->p_state == PRS_NEW) {
3581 				PROC_UNLOCK(p);
3582 				continue;
3583 			}
3584 			fdp = fdhold(p);
3585 			PROC_UNLOCK(p);
3586 			if (fdp == NULL)
3587 				continue;
3588 			/* overestimates sparse tables. */
3589 			if (fdp->fd_lastfile > 0)
3590 				n += fdp->fd_lastfile;
3591 			fddrop(fdp);
3592 		}
3593 		sx_sunlock(&allproc_lock);
3594 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3595 	}
3596 	error = 0;
3597 	bzero(&xf, sizeof(xf));
3598 	xf.xf_size = sizeof(xf);
3599 	sx_slock(&allproc_lock);
3600 	FOREACH_PROC_IN_SYSTEM(p) {
3601 		PROC_LOCK(p);
3602 		if (p->p_state == PRS_NEW) {
3603 			PROC_UNLOCK(p);
3604 			continue;
3605 		}
3606 		if (p_cansee(req->td, p) != 0) {
3607 			PROC_UNLOCK(p);
3608 			continue;
3609 		}
3610 		xf.xf_pid = p->p_pid;
3611 		xf.xf_uid = p->p_ucred->cr_uid;
3612 		fdp = fdhold(p);
3613 		PROC_UNLOCK(p);
3614 		if (fdp == NULL)
3615 			continue;
3616 		FILEDESC_SLOCK(fdp);
3617 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3618 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3619 				continue;
3620 			xf.xf_fd = n;
3621 			xf.xf_file = (uintptr_t)fp;
3622 			xf.xf_data = (uintptr_t)fp->f_data;
3623 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3624 			xf.xf_type = (uintptr_t)fp->f_type;
3625 			xf.xf_count = fp->f_count;
3626 			xf.xf_msgcount = 0;
3627 			xf.xf_offset = foffset_get(fp);
3628 			xf.xf_flag = fp->f_flag;
3629 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3630 			if (error)
3631 				break;
3632 		}
3633 		FILEDESC_SUNLOCK(fdp);
3634 		fddrop(fdp);
3635 		if (error)
3636 			break;
3637 	}
3638 	sx_sunlock(&allproc_lock);
3639 	return (error);
3640 }
3641 
3642 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3643     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3644 
3645 #ifdef KINFO_FILE_SIZE
3646 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3647 #endif
3648 
3649 static int
3650 xlate_fflags(int fflags)
3651 {
3652 	static const struct {
3653 		int	fflag;
3654 		int	kf_fflag;
3655 	} fflags_table[] = {
3656 		{ FAPPEND, KF_FLAG_APPEND },
3657 		{ FASYNC, KF_FLAG_ASYNC },
3658 		{ FFSYNC, KF_FLAG_FSYNC },
3659 		{ FHASLOCK, KF_FLAG_HASLOCK },
3660 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3661 		{ FREAD, KF_FLAG_READ },
3662 		{ FWRITE, KF_FLAG_WRITE },
3663 		{ O_CREAT, KF_FLAG_CREAT },
3664 		{ O_DIRECT, KF_FLAG_DIRECT },
3665 		{ O_EXCL, KF_FLAG_EXCL },
3666 		{ O_EXEC, KF_FLAG_EXEC },
3667 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3668 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3669 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3670 		{ O_TRUNC, KF_FLAG_TRUNC }
3671 	};
3672 	unsigned int i;
3673 	int kflags;
3674 
3675 	kflags = 0;
3676 	for (i = 0; i < nitems(fflags_table); i++)
3677 		if (fflags & fflags_table[i].fflag)
3678 			kflags |=  fflags_table[i].kf_fflag;
3679 	return (kflags);
3680 }
3681 
3682 /* Trim unused data from kf_path by truncating the structure size. */
3683 void
3684 pack_kinfo(struct kinfo_file *kif)
3685 {
3686 
3687 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3688 	    strlen(kif->kf_path) + 1;
3689 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3690 }
3691 
3692 static void
3693 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3694     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3695 {
3696 	int error;
3697 
3698 	bzero(kif, sizeof(*kif));
3699 
3700 	/* Set a default type to allow for empty fill_kinfo() methods. */
3701 	kif->kf_type = KF_TYPE_UNKNOWN;
3702 	kif->kf_flags = xlate_fflags(fp->f_flag);
3703 	if (rightsp != NULL)
3704 		kif->kf_cap_rights = *rightsp;
3705 	else
3706 		cap_rights_init_zero(&kif->kf_cap_rights);
3707 	kif->kf_fd = fd;
3708 	kif->kf_ref_count = fp->f_count;
3709 	kif->kf_offset = foffset_get(fp);
3710 
3711 	/*
3712 	 * This may drop the filedesc lock, so the 'fp' cannot be
3713 	 * accessed after this call.
3714 	 */
3715 	error = fo_fill_kinfo(fp, kif, fdp);
3716 	if (error == 0)
3717 		kif->kf_status |= KF_ATTR_VALID;
3718 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3719 		pack_kinfo(kif);
3720 	else
3721 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3722 }
3723 
3724 static void
3725 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3726     struct kinfo_file *kif, int flags)
3727 {
3728 	int error;
3729 
3730 	bzero(kif, sizeof(*kif));
3731 
3732 	kif->kf_type = KF_TYPE_VNODE;
3733 	error = vn_fill_kinfo_vnode(vp, kif);
3734 	if (error == 0)
3735 		kif->kf_status |= KF_ATTR_VALID;
3736 	kif->kf_flags = xlate_fflags(fflags);
3737 	cap_rights_init_zero(&kif->kf_cap_rights);
3738 	kif->kf_fd = fd;
3739 	kif->kf_ref_count = -1;
3740 	kif->kf_offset = -1;
3741 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3742 		pack_kinfo(kif);
3743 	else
3744 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3745 	vrele(vp);
3746 }
3747 
3748 struct export_fd_buf {
3749 	struct filedesc		*fdp;
3750 	struct sbuf 		*sb;
3751 	ssize_t			remainder;
3752 	struct kinfo_file	kif;
3753 	int			flags;
3754 };
3755 
3756 static int
3757 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3758 {
3759 	struct kinfo_file *kif;
3760 
3761 	kif = &efbuf->kif;
3762 	if (efbuf->remainder != -1) {
3763 		if (efbuf->remainder < kif->kf_structsize) {
3764 			/* Terminate export. */
3765 			efbuf->remainder = 0;
3766 			return (0);
3767 		}
3768 		efbuf->remainder -= kif->kf_structsize;
3769 	}
3770 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3771 }
3772 
3773 static int
3774 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3775     struct export_fd_buf *efbuf)
3776 {
3777 	int error;
3778 
3779 	if (efbuf->remainder == 0)
3780 		return (0);
3781 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3782 	    efbuf->flags);
3783 	FILEDESC_SUNLOCK(efbuf->fdp);
3784 	error = export_kinfo_to_sb(efbuf);
3785 	FILEDESC_SLOCK(efbuf->fdp);
3786 	return (error);
3787 }
3788 
3789 static int
3790 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3791     struct export_fd_buf *efbuf)
3792 {
3793 	int error;
3794 
3795 	if (efbuf->remainder == 0)
3796 		return (0);
3797 	if (efbuf->fdp != NULL)
3798 		FILEDESC_SUNLOCK(efbuf->fdp);
3799 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3800 	error = export_kinfo_to_sb(efbuf);
3801 	if (efbuf->fdp != NULL)
3802 		FILEDESC_SLOCK(efbuf->fdp);
3803 	return (error);
3804 }
3805 
3806 /*
3807  * Store a process file descriptor information to sbuf.
3808  *
3809  * Takes a locked proc as argument, and returns with the proc unlocked.
3810  */
3811 int
3812 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3813     int flags)
3814 {
3815 	struct file *fp;
3816 	struct filedesc *fdp;
3817 	struct export_fd_buf *efbuf;
3818 	struct vnode *cttyvp, *textvp, *tracevp;
3819 	struct pwd *pwd;
3820 	int error, i;
3821 	cap_rights_t rights;
3822 
3823 	PROC_LOCK_ASSERT(p, MA_OWNED);
3824 
3825 	/* ktrace vnode */
3826 	tracevp = p->p_tracevp;
3827 	if (tracevp != NULL)
3828 		vrefact(tracevp);
3829 	/* text vnode */
3830 	textvp = p->p_textvp;
3831 	if (textvp != NULL)
3832 		vrefact(textvp);
3833 	/* Controlling tty. */
3834 	cttyvp = NULL;
3835 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3836 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3837 		if (cttyvp != NULL)
3838 			vrefact(cttyvp);
3839 	}
3840 	fdp = fdhold(p);
3841 	PROC_UNLOCK(p);
3842 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3843 	efbuf->fdp = NULL;
3844 	efbuf->sb = sb;
3845 	efbuf->remainder = maxlen;
3846 	efbuf->flags = flags;
3847 	if (tracevp != NULL)
3848 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3849 		    efbuf);
3850 	if (textvp != NULL)
3851 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3852 	if (cttyvp != NULL)
3853 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3854 		    efbuf);
3855 	error = 0;
3856 	if (fdp == NULL)
3857 		goto fail;
3858 	efbuf->fdp = fdp;
3859 	FILEDESC_SLOCK(fdp);
3860 	pwd = pwd_hold_filedesc(fdp);
3861 	if (pwd != NULL) {
3862 		/* working directory */
3863 		if (pwd->pwd_cdir != NULL) {
3864 			vrefact(pwd->pwd_cdir);
3865 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3866 		}
3867 		/* root directory */
3868 		if (pwd->pwd_rdir != NULL) {
3869 			vrefact(pwd->pwd_rdir);
3870 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3871 		}
3872 		/* jail directory */
3873 		if (pwd->pwd_jdir != NULL) {
3874 			vrefact(pwd->pwd_jdir);
3875 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3876 		}
3877 		pwd_drop(pwd);
3878 	}
3879 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3880 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3881 			continue;
3882 #ifdef CAPABILITIES
3883 		rights = *cap_rights(fdp, i);
3884 #else /* !CAPABILITIES */
3885 		rights = cap_no_rights;
3886 #endif
3887 		/*
3888 		 * Create sysctl entry.  It is OK to drop the filedesc
3889 		 * lock inside of export_file_to_sb() as we will
3890 		 * re-validate and re-evaluate its properties when the
3891 		 * loop continues.
3892 		 */
3893 		error = export_file_to_sb(fp, i, &rights, efbuf);
3894 		if (error != 0 || efbuf->remainder == 0)
3895 			break;
3896 	}
3897 	FILEDESC_SUNLOCK(fdp);
3898 	fddrop(fdp);
3899 fail:
3900 	free(efbuf, M_TEMP);
3901 	return (error);
3902 }
3903 
3904 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3905 
3906 /*
3907  * Get per-process file descriptors for use by procstat(1), et al.
3908  */
3909 static int
3910 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3911 {
3912 	struct sbuf sb;
3913 	struct proc *p;
3914 	ssize_t maxlen;
3915 	int error, error2, *name;
3916 
3917 	name = (int *)arg1;
3918 
3919 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3920 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3921 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3922 	if (error != 0) {
3923 		sbuf_delete(&sb);
3924 		return (error);
3925 	}
3926 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3927 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3928 	    KERN_FILEDESC_PACK_KINFO);
3929 	error2 = sbuf_finish(&sb);
3930 	sbuf_delete(&sb);
3931 	return (error != 0 ? error : error2);
3932 }
3933 
3934 #ifdef COMPAT_FREEBSD7
3935 #ifdef KINFO_OFILE_SIZE
3936 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3937 #endif
3938 
3939 static void
3940 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3941 {
3942 
3943 	okif->kf_structsize = sizeof(*okif);
3944 	okif->kf_type = kif->kf_type;
3945 	okif->kf_fd = kif->kf_fd;
3946 	okif->kf_ref_count = kif->kf_ref_count;
3947 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3948 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3949 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3950 	okif->kf_offset = kif->kf_offset;
3951 	if (kif->kf_type == KF_TYPE_VNODE)
3952 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3953 	else
3954 		okif->kf_vnode_type = KF_VTYPE_VNON;
3955 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3956 	if (kif->kf_type == KF_TYPE_SOCKET) {
3957 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3958 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3959 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3960 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3961 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3962 	} else {
3963 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3964 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3965 	}
3966 }
3967 
3968 static int
3969 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3970     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3971 {
3972 	int error;
3973 
3974 	vrefact(vp);
3975 	FILEDESC_SUNLOCK(fdp);
3976 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3977 	kinfo_to_okinfo(kif, okif);
3978 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3979 	FILEDESC_SLOCK(fdp);
3980 	return (error);
3981 }
3982 
3983 /*
3984  * Get per-process file descriptors for use by procstat(1), et al.
3985  */
3986 static int
3987 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3988 {
3989 	struct kinfo_ofile *okif;
3990 	struct kinfo_file *kif;
3991 	struct filedesc *fdp;
3992 	struct pwd *pwd;
3993 	int error, i, *name;
3994 	struct file *fp;
3995 	struct proc *p;
3996 
3997 	name = (int *)arg1;
3998 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3999 	if (error != 0)
4000 		return (error);
4001 	fdp = fdhold(p);
4002 	PROC_UNLOCK(p);
4003 	if (fdp == NULL)
4004 		return (ENOENT);
4005 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4006 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4007 	FILEDESC_SLOCK(fdp);
4008 	pwd = pwd_hold_filedesc(fdp);
4009 	if (pwd != NULL) {
4010 		if (pwd->pwd_cdir != NULL)
4011 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4012 			    okif, fdp, req);
4013 		if (pwd->pwd_rdir != NULL)
4014 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4015 			    okif, fdp, req);
4016 		if (pwd->pwd_jdir != NULL)
4017 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4018 			    okif, fdp, req);
4019 		pwd_drop(pwd);
4020 	}
4021 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
4022 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4023 			continue;
4024 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4025 		    KERN_FILEDESC_PACK_KINFO);
4026 		FILEDESC_SUNLOCK(fdp);
4027 		kinfo_to_okinfo(kif, okif);
4028 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4029 		FILEDESC_SLOCK(fdp);
4030 		if (error)
4031 			break;
4032 	}
4033 	FILEDESC_SUNLOCK(fdp);
4034 	fddrop(fdp);
4035 	free(kif, M_TEMP);
4036 	free(okif, M_TEMP);
4037 	return (0);
4038 }
4039 
4040 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4041     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4042     "Process ofiledesc entries");
4043 #endif	/* COMPAT_FREEBSD7 */
4044 
4045 int
4046 vntype_to_kinfo(int vtype)
4047 {
4048 	struct {
4049 		int	vtype;
4050 		int	kf_vtype;
4051 	} vtypes_table[] = {
4052 		{ VBAD, KF_VTYPE_VBAD },
4053 		{ VBLK, KF_VTYPE_VBLK },
4054 		{ VCHR, KF_VTYPE_VCHR },
4055 		{ VDIR, KF_VTYPE_VDIR },
4056 		{ VFIFO, KF_VTYPE_VFIFO },
4057 		{ VLNK, KF_VTYPE_VLNK },
4058 		{ VNON, KF_VTYPE_VNON },
4059 		{ VREG, KF_VTYPE_VREG },
4060 		{ VSOCK, KF_VTYPE_VSOCK }
4061 	};
4062 	unsigned int i;
4063 
4064 	/*
4065 	 * Perform vtype translation.
4066 	 */
4067 	for (i = 0; i < nitems(vtypes_table); i++)
4068 		if (vtypes_table[i].vtype == vtype)
4069 			return (vtypes_table[i].kf_vtype);
4070 
4071 	return (KF_VTYPE_UNKNOWN);
4072 }
4073 
4074 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4075     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4076     "Process filedesc entries");
4077 
4078 /*
4079  * Store a process current working directory information to sbuf.
4080  *
4081  * Takes a locked proc as argument, and returns with the proc unlocked.
4082  */
4083 int
4084 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4085 {
4086 	struct filedesc *fdp;
4087 	struct pwd *pwd;
4088 	struct export_fd_buf *efbuf;
4089 	struct vnode *cdir;
4090 	int error;
4091 
4092 	PROC_LOCK_ASSERT(p, MA_OWNED);
4093 
4094 	fdp = fdhold(p);
4095 	PROC_UNLOCK(p);
4096 	if (fdp == NULL)
4097 		return (EINVAL);
4098 
4099 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4100 	efbuf->fdp = fdp;
4101 	efbuf->sb = sb;
4102 	efbuf->remainder = maxlen;
4103 
4104 	FILEDESC_SLOCK(fdp);
4105 	pwd = FILEDESC_LOCKED_LOAD_PWD(fdp);
4106 	cdir = pwd->pwd_cdir;
4107 	if (cdir == NULL) {
4108 		error = EINVAL;
4109 	} else {
4110 		vrefact(cdir);
4111 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4112 	}
4113 	FILEDESC_SUNLOCK(fdp);
4114 	fddrop(fdp);
4115 	free(efbuf, M_TEMP);
4116 	return (error);
4117 }
4118 
4119 /*
4120  * Get per-process current working directory.
4121  */
4122 static int
4123 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4124 {
4125 	struct sbuf sb;
4126 	struct proc *p;
4127 	ssize_t maxlen;
4128 	int error, error2, *name;
4129 
4130 	name = (int *)arg1;
4131 
4132 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4133 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4134 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4135 	if (error != 0) {
4136 		sbuf_delete(&sb);
4137 		return (error);
4138 	}
4139 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4140 	error = kern_proc_cwd_out(p, &sb, maxlen);
4141 	error2 = sbuf_finish(&sb);
4142 	sbuf_delete(&sb);
4143 	return (error != 0 ? error : error2);
4144 }
4145 
4146 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4147     sysctl_kern_proc_cwd, "Process current working directory");
4148 
4149 #ifdef DDB
4150 /*
4151  * For the purposes of debugging, generate a human-readable string for the
4152  * file type.
4153  */
4154 static const char *
4155 file_type_to_name(short type)
4156 {
4157 
4158 	switch (type) {
4159 	case 0:
4160 		return ("zero");
4161 	case DTYPE_VNODE:
4162 		return ("vnode");
4163 	case DTYPE_SOCKET:
4164 		return ("socket");
4165 	case DTYPE_PIPE:
4166 		return ("pipe");
4167 	case DTYPE_FIFO:
4168 		return ("fifo");
4169 	case DTYPE_KQUEUE:
4170 		return ("kqueue");
4171 	case DTYPE_CRYPTO:
4172 		return ("crypto");
4173 	case DTYPE_MQUEUE:
4174 		return ("mqueue");
4175 	case DTYPE_SHM:
4176 		return ("shm");
4177 	case DTYPE_SEM:
4178 		return ("ksem");
4179 	case DTYPE_PTS:
4180 		return ("pts");
4181 	case DTYPE_DEV:
4182 		return ("dev");
4183 	case DTYPE_PROCDESC:
4184 		return ("proc");
4185 	case DTYPE_LINUXEFD:
4186 		return ("levent");
4187 	case DTYPE_LINUXTFD:
4188 		return ("ltimer");
4189 	default:
4190 		return ("unkn");
4191 	}
4192 }
4193 
4194 /*
4195  * For the purposes of debugging, identify a process (if any, perhaps one of
4196  * many) that references the passed file in its file descriptor array. Return
4197  * NULL if none.
4198  */
4199 static struct proc *
4200 file_to_first_proc(struct file *fp)
4201 {
4202 	struct filedesc *fdp;
4203 	struct proc *p;
4204 	int n;
4205 
4206 	FOREACH_PROC_IN_SYSTEM(p) {
4207 		if (p->p_state == PRS_NEW)
4208 			continue;
4209 		fdp = p->p_fd;
4210 		if (fdp == NULL)
4211 			continue;
4212 		for (n = 0; n <= fdp->fd_lastfile; n++) {
4213 			if (fp == fdp->fd_ofiles[n].fde_file)
4214 				return (p);
4215 		}
4216 	}
4217 	return (NULL);
4218 }
4219 
4220 static void
4221 db_print_file(struct file *fp, int header)
4222 {
4223 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4224 	struct proc *p;
4225 
4226 	if (header)
4227 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4228 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4229 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4230 		    "FCmd");
4231 	p = file_to_first_proc(fp);
4232 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4233 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4234 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4235 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4236 
4237 #undef XPTRWIDTH
4238 }
4239 
4240 DB_SHOW_COMMAND(file, db_show_file)
4241 {
4242 	struct file *fp;
4243 
4244 	if (!have_addr) {
4245 		db_printf("usage: show file <addr>\n");
4246 		return;
4247 	}
4248 	fp = (struct file *)addr;
4249 	db_print_file(fp, 1);
4250 }
4251 
4252 DB_SHOW_COMMAND(files, db_show_files)
4253 {
4254 	struct filedesc *fdp;
4255 	struct file *fp;
4256 	struct proc *p;
4257 	int header;
4258 	int n;
4259 
4260 	header = 1;
4261 	FOREACH_PROC_IN_SYSTEM(p) {
4262 		if (p->p_state == PRS_NEW)
4263 			continue;
4264 		if ((fdp = p->p_fd) == NULL)
4265 			continue;
4266 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
4267 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4268 				continue;
4269 			db_print_file(fp, header);
4270 			header = 0;
4271 		}
4272 	}
4273 }
4274 #endif
4275 
4276 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4277     &maxfilesperproc, 0, "Maximum files allowed open per process");
4278 
4279 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4280     &maxfiles, 0, "Maximum number of files");
4281 
4282 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4283     &openfiles, 0, "System-wide number of open files");
4284 
4285 /* ARGSUSED*/
4286 static void
4287 filelistinit(void *dummy)
4288 {
4289 
4290 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4291 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4292 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4293 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4294 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4295 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4296 	pwd_smr = uma_zone_get_smr(pwd_zone);
4297 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4298 }
4299 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4300 
4301 /*-------------------------------------------------------------------*/
4302 
4303 static int
4304 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4305     int flags, struct thread *td)
4306 {
4307 
4308 	return (EBADF);
4309 }
4310 
4311 static int
4312 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4313     struct thread *td)
4314 {
4315 
4316 	return (EINVAL);
4317 }
4318 
4319 static int
4320 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4321     struct thread *td)
4322 {
4323 
4324 	return (EBADF);
4325 }
4326 
4327 static int
4328 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4329     struct thread *td)
4330 {
4331 
4332 	return (0);
4333 }
4334 
4335 static int
4336 badfo_kqfilter(struct file *fp, struct knote *kn)
4337 {
4338 
4339 	return (EBADF);
4340 }
4341 
4342 static int
4343 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4344     struct thread *td)
4345 {
4346 
4347 	return (EBADF);
4348 }
4349 
4350 static int
4351 badfo_close(struct file *fp, struct thread *td)
4352 {
4353 
4354 	return (0);
4355 }
4356 
4357 static int
4358 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4359     struct thread *td)
4360 {
4361 
4362 	return (EBADF);
4363 }
4364 
4365 static int
4366 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4367     struct thread *td)
4368 {
4369 
4370 	return (EBADF);
4371 }
4372 
4373 static int
4374 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4375     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4376     struct thread *td)
4377 {
4378 
4379 	return (EBADF);
4380 }
4381 
4382 static int
4383 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4384 {
4385 
4386 	return (0);
4387 }
4388 
4389 struct fileops badfileops = {
4390 	.fo_read = badfo_readwrite,
4391 	.fo_write = badfo_readwrite,
4392 	.fo_truncate = badfo_truncate,
4393 	.fo_ioctl = badfo_ioctl,
4394 	.fo_poll = badfo_poll,
4395 	.fo_kqfilter = badfo_kqfilter,
4396 	.fo_stat = badfo_stat,
4397 	.fo_close = badfo_close,
4398 	.fo_chmod = badfo_chmod,
4399 	.fo_chown = badfo_chown,
4400 	.fo_sendfile = badfo_sendfile,
4401 	.fo_fill_kinfo = badfo_fill_kinfo,
4402 };
4403 
4404 int
4405 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4406     int flags, struct thread *td)
4407 {
4408 
4409 	return (EOPNOTSUPP);
4410 }
4411 
4412 int
4413 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4414     struct thread *td)
4415 {
4416 
4417 	return (EINVAL);
4418 }
4419 
4420 int
4421 invfo_ioctl(struct file *fp, u_long com, void *data,
4422     struct ucred *active_cred, struct thread *td)
4423 {
4424 
4425 	return (ENOTTY);
4426 }
4427 
4428 int
4429 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4430     struct thread *td)
4431 {
4432 
4433 	return (poll_no_poll(events));
4434 }
4435 
4436 int
4437 invfo_kqfilter(struct file *fp, struct knote *kn)
4438 {
4439 
4440 	return (EINVAL);
4441 }
4442 
4443 int
4444 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4445     struct thread *td)
4446 {
4447 
4448 	return (EINVAL);
4449 }
4450 
4451 int
4452 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4453     struct thread *td)
4454 {
4455 
4456 	return (EINVAL);
4457 }
4458 
4459 int
4460 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4461     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4462     struct thread *td)
4463 {
4464 
4465 	return (EINVAL);
4466 }
4467 
4468 /*-------------------------------------------------------------------*/
4469 
4470 /*
4471  * File Descriptor pseudo-device driver (/dev/fd/).
4472  *
4473  * Opening minor device N dup()s the file (if any) connected to file
4474  * descriptor N belonging to the calling process.  Note that this driver
4475  * consists of only the ``open()'' routine, because all subsequent
4476  * references to this file will be direct to the other driver.
4477  *
4478  * XXX: we could give this one a cloning event handler if necessary.
4479  */
4480 
4481 /* ARGSUSED */
4482 static int
4483 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4484 {
4485 
4486 	/*
4487 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4488 	 * the file descriptor being sought for duplication. The error
4489 	 * return ensures that the vnode for this device will be released
4490 	 * by vn_open. Open will detect this special error and take the
4491 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4492 	 * will simply report the error.
4493 	 */
4494 	td->td_dupfd = dev2unit(dev);
4495 	return (ENODEV);
4496 }
4497 
4498 static struct cdevsw fildesc_cdevsw = {
4499 	.d_version =	D_VERSION,
4500 	.d_open =	fdopen,
4501 	.d_name =	"FD",
4502 };
4503 
4504 static void
4505 fildesc_drvinit(void *unused)
4506 {
4507 	struct cdev *dev;
4508 
4509 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4510 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4511 	make_dev_alias(dev, "stdin");
4512 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4513 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4514 	make_dev_alias(dev, "stdout");
4515 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4516 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4517 	make_dev_alias(dev, "stderr");
4518 }
4519 
4520 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4521