xref: /freebsd/sys/kern/kern_descrip.c (revision 3823d5e198425b4f5e5a80267d195769d1063773)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ddb.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 
48 #include <sys/capsicum.h>
49 #include <sys/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/file.h>
52 #include <sys/filedesc.h>
53 #include <sys/filio.h>
54 #include <sys/jail.h>
55 #include <sys/kernel.h>
56 #include <sys/limits.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/selinfo.h>
63 #include <sys/priv.h>
64 #include <sys/proc.h>
65 #include <sys/protosw.h>
66 #include <sys/racct.h>
67 #include <sys/resourcevar.h>
68 #include <sys/sbuf.h>
69 #include <sys/signalvar.h>
70 #include <sys/socketvar.h>
71 #include <sys/stat.h>
72 #include <sys/sx.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysctl.h>
75 #include <sys/sysproto.h>
76 #include <sys/unistd.h>
77 #include <sys/user.h>
78 #include <sys/vnode.h>
79 #ifdef KTRACE
80 #include <sys/ktrace.h>
81 #endif
82 
83 #include <net/vnet.h>
84 
85 #include <security/audit/audit.h>
86 
87 #include <vm/uma.h>
88 #include <vm/vm.h>
89 
90 #include <ddb/ddb.h>
91 
92 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
93 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
94     "file desc to leader structures");
95 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
96 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
97 
98 MALLOC_DECLARE(M_FADVISE);
99 
100 static uma_zone_t file_zone;
101 
102 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
103 		    struct thread *td, int holdleaders);
104 static int	do_dup(struct thread *td, int flags, int old, int new,
105 		    register_t *retval);
106 static int	fd_first_free(struct filedesc *fdp, int low, int size);
107 static int	fd_last_used(struct filedesc *fdp, int size);
108 static void	fdgrowtable(struct filedesc *fdp, int nfd);
109 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void	fdunused(struct filedesc *fdp, int fd);
111 static void	fdused(struct filedesc *fdp, int fd);
112 static int	getmaxfd(struct proc *p);
113 
114 /* Flags for do_dup() */
115 #define	DUP_FIXED	0x1	/* Force fixed allocation. */
116 #define	DUP_FCNTL	0x2	/* fcntl()-style errors. */
117 #define	DUP_CLOEXEC	0x4	/* Atomically set FD_CLOEXEC. */
118 
119 /*
120  * Each process has:
121  *
122  * - An array of open file descriptors (fd_ofiles)
123  * - An array of file flags (fd_ofileflags)
124  * - A bitmap recording which descriptors are in use (fd_map)
125  *
126  * A process starts out with NDFILE descriptors.  The value of NDFILE has
127  * been selected based the historical limit of 20 open files, and an
128  * assumption that the majority of processes, especially short-lived
129  * processes like shells, will never need more.
130  *
131  * If this initial allocation is exhausted, a larger descriptor table and
132  * map are allocated dynamically, and the pointers in the process's struct
133  * filedesc are updated to point to those.  This is repeated every time
134  * the process runs out of file descriptors (provided it hasn't hit its
135  * resource limit).
136  *
137  * Since threads may hold references to individual descriptor table
138  * entries, the tables are never freed.  Instead, they are placed on a
139  * linked list and freed only when the struct filedesc is released.
140  */
141 #define NDFILE		20
142 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
143 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x)	((x) / NDENTRIES)
145 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
147 
148 /*
149  * SLIST entry used to keep track of ofiles which must be reclaimed when
150  * the process exits.
151  */
152 struct freetable {
153 	struct filedescent *ft_table;
154 	SLIST_ENTRY(freetable) ft_next;
155 };
156 
157 /*
158  * Initial allocation: a filedesc structure + the head of SLIST used to
159  * keep track of old ofiles + enough space for NDFILE descriptors.
160  */
161 struct filedesc0 {
162 	struct filedesc fd_fd;
163 	SLIST_HEAD(, freetable) fd_free;
164 	struct	filedescent fd_dfiles[NDFILE];
165 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
166 };
167 
168 /*
169  * Descriptor management.
170  */
171 volatile int openfiles;			/* actual number of open files */
172 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
173 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
174 
175 /* A mutex to protect the association between a proc and filedesc. */
176 static struct mtx fdesc_mtx;
177 
178 /*
179  * If low >= size, just return low. Otherwise find the first zero bit in the
180  * given bitmap, starting at low and not exceeding size - 1. Return size if
181  * not found.
182  */
183 static int
184 fd_first_free(struct filedesc *fdp, int low, int size)
185 {
186 	NDSLOTTYPE *map = fdp->fd_map;
187 	NDSLOTTYPE mask;
188 	int off, maxoff;
189 
190 	if (low >= size)
191 		return (low);
192 
193 	off = NDSLOT(low);
194 	if (low % NDENTRIES) {
195 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
196 		if ((mask &= ~map[off]) != 0UL)
197 			return (off * NDENTRIES + ffsl(mask) - 1);
198 		++off;
199 	}
200 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
201 		if (map[off] != ~0UL)
202 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
203 	return (size);
204 }
205 
206 /*
207  * Find the highest non-zero bit in the given bitmap, starting at 0 and
208  * not exceeding size - 1. Return -1 if not found.
209  */
210 static int
211 fd_last_used(struct filedesc *fdp, int size)
212 {
213 	NDSLOTTYPE *map = fdp->fd_map;
214 	NDSLOTTYPE mask;
215 	int off, minoff;
216 
217 	off = NDSLOT(size);
218 	if (size % NDENTRIES) {
219 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
220 		if ((mask &= map[off]) != 0)
221 			return (off * NDENTRIES + flsl(mask) - 1);
222 		--off;
223 	}
224 	for (minoff = NDSLOT(0); off >= minoff; --off)
225 		if (map[off] != 0)
226 			return (off * NDENTRIES + flsl(map[off]) - 1);
227 	return (-1);
228 }
229 
230 static int
231 fdisused(struct filedesc *fdp, int fd)
232 {
233 
234 	FILEDESC_LOCK_ASSERT(fdp);
235 
236 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
237 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
238 
239 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
240 }
241 
242 /*
243  * Mark a file descriptor as used.
244  */
245 static void
246 fdused(struct filedesc *fdp, int fd)
247 {
248 
249 	FILEDESC_XLOCK_ASSERT(fdp);
250 
251 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
252 
253 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
254 	if (fd > fdp->fd_lastfile)
255 		fdp->fd_lastfile = fd;
256 	if (fd == fdp->fd_freefile)
257 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
258 }
259 
260 /*
261  * Mark a file descriptor as unused.
262  */
263 static void
264 fdunused(struct filedesc *fdp, int fd)
265 {
266 
267 	FILEDESC_XLOCK_ASSERT(fdp);
268 
269 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
270 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
271 	    ("fd=%d is still in use", fd));
272 
273 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
274 	if (fd < fdp->fd_freefile)
275 		fdp->fd_freefile = fd;
276 	if (fd == fdp->fd_lastfile)
277 		fdp->fd_lastfile = fd_last_used(fdp, fd);
278 }
279 
280 /*
281  * Free a file descriptor.
282  *
283  * Avoid some work if fdp is about to be destroyed.
284  */
285 static inline void
286 _fdfree(struct filedesc *fdp, int fd, int last)
287 {
288 	struct filedescent *fde;
289 
290 	fde = &fdp->fd_ofiles[fd];
291 #ifdef CAPABILITIES
292 	if (!last)
293 		seq_write_begin(&fde->fde_seq);
294 #endif
295 	filecaps_free(&fde->fde_caps);
296 	if (last)
297 		return;
298 	bzero(fde, fde_change_size);
299 	fdunused(fdp, fd);
300 #ifdef CAPABILITIES
301 	seq_write_end(&fde->fde_seq);
302 #endif
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 
309 	_fdfree(fdp, fd, 0);
310 }
311 
312 static inline void
313 fdfree_last(struct filedesc *fdp, int fd)
314 {
315 
316 	_fdfree(fdp, fd, 1);
317 }
318 
319 /*
320  * System calls on descriptors.
321  */
322 #ifndef _SYS_SYSPROTO_H_
323 struct getdtablesize_args {
324 	int	dummy;
325 };
326 #endif
327 /* ARGSUSED */
328 int
329 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
330 {
331 	struct proc *p = td->td_proc;
332 	uint64_t lim;
333 
334 	PROC_LOCK(p);
335 	td->td_retval[0] =
336 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
337 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
338 	PROC_UNLOCK(p);
339 	if (lim < td->td_retval[0])
340 		td->td_retval[0] = lim;
341 	return (0);
342 }
343 
344 /*
345  * Duplicate a file descriptor to a particular value.
346  *
347  * Note: keep in mind that a potential race condition exists when closing
348  * descriptors from a shared descriptor table (via rfork).
349  */
350 #ifndef _SYS_SYSPROTO_H_
351 struct dup2_args {
352 	u_int	from;
353 	u_int	to;
354 };
355 #endif
356 /* ARGSUSED */
357 int
358 sys_dup2(struct thread *td, struct dup2_args *uap)
359 {
360 
361 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
362 		    td->td_retval));
363 }
364 
365 /*
366  * Duplicate a file descriptor.
367  */
368 #ifndef _SYS_SYSPROTO_H_
369 struct dup_args {
370 	u_int	fd;
371 };
372 #endif
373 /* ARGSUSED */
374 int
375 sys_dup(struct thread *td, struct dup_args *uap)
376 {
377 
378 	return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
379 }
380 
381 /*
382  * The file control system call.
383  */
384 #ifndef _SYS_SYSPROTO_H_
385 struct fcntl_args {
386 	int	fd;
387 	int	cmd;
388 	long	arg;
389 };
390 #endif
391 /* ARGSUSED */
392 int
393 sys_fcntl(struct thread *td, struct fcntl_args *uap)
394 {
395 
396 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
397 }
398 
399 int
400 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
401 {
402 	struct flock fl;
403 	struct __oflock ofl;
404 	intptr_t arg1;
405 	int error;
406 
407 	error = 0;
408 	switch (cmd) {
409 	case F_OGETLK:
410 	case F_OSETLK:
411 	case F_OSETLKW:
412 		/*
413 		 * Convert old flock structure to new.
414 		 */
415 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
416 		fl.l_start = ofl.l_start;
417 		fl.l_len = ofl.l_len;
418 		fl.l_pid = ofl.l_pid;
419 		fl.l_type = ofl.l_type;
420 		fl.l_whence = ofl.l_whence;
421 		fl.l_sysid = 0;
422 
423 		switch (cmd) {
424 		case F_OGETLK:
425 		    cmd = F_GETLK;
426 		    break;
427 		case F_OSETLK:
428 		    cmd = F_SETLK;
429 		    break;
430 		case F_OSETLKW:
431 		    cmd = F_SETLKW;
432 		    break;
433 		}
434 		arg1 = (intptr_t)&fl;
435 		break;
436         case F_GETLK:
437         case F_SETLK:
438         case F_SETLKW:
439 	case F_SETLK_REMOTE:
440                 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
441                 arg1 = (intptr_t)&fl;
442                 break;
443 	default:
444 		arg1 = arg;
445 		break;
446 	}
447 	if (error)
448 		return (error);
449 	error = kern_fcntl(td, fd, cmd, arg1);
450 	if (error)
451 		return (error);
452 	if (cmd == F_OGETLK) {
453 		ofl.l_start = fl.l_start;
454 		ofl.l_len = fl.l_len;
455 		ofl.l_pid = fl.l_pid;
456 		ofl.l_type = fl.l_type;
457 		ofl.l_whence = fl.l_whence;
458 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
459 	} else if (cmd == F_GETLK) {
460 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
461 	}
462 	return (error);
463 }
464 
465 int
466 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
467 {
468 	struct filedesc *fdp;
469 	struct flock *flp;
470 	struct file *fp, *fp2;
471 	struct filedescent *fde;
472 	struct proc *p;
473 	struct vnode *vp;
474 	cap_rights_t rights;
475 	int error, flg, tmp;
476 	uint64_t bsize;
477 	off_t foffset;
478 
479 	error = 0;
480 	flg = F_POSIX;
481 	p = td->td_proc;
482 	fdp = p->p_fd;
483 
484 	switch (cmd) {
485 	case F_DUPFD:
486 		tmp = arg;
487 		error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
488 		break;
489 
490 	case F_DUPFD_CLOEXEC:
491 		tmp = arg;
492 		error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp,
493 		    td->td_retval);
494 		break;
495 
496 	case F_DUP2FD:
497 		tmp = arg;
498 		error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
499 		break;
500 
501 	case F_DUP2FD_CLOEXEC:
502 		tmp = arg;
503 		error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp,
504 		    td->td_retval);
505 		break;
506 
507 	case F_GETFD:
508 		FILEDESC_SLOCK(fdp);
509 		if (fget_locked(fdp, fd) == NULL) {
510 			FILEDESC_SUNLOCK(fdp);
511 			error = EBADF;
512 			break;
513 		}
514 		fde = &fdp->fd_ofiles[fd];
515 		td->td_retval[0] =
516 		    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
517 		FILEDESC_SUNLOCK(fdp);
518 		break;
519 
520 	case F_SETFD:
521 		FILEDESC_XLOCK(fdp);
522 		if (fget_locked(fdp, fd) == NULL) {
523 			FILEDESC_XUNLOCK(fdp);
524 			error = EBADF;
525 			break;
526 		}
527 		fde = &fdp->fd_ofiles[fd];
528 		fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
529 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
530 		FILEDESC_XUNLOCK(fdp);
531 		break;
532 
533 	case F_GETFL:
534 		error = fget_unlocked(fdp, fd,
535 		    cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL);
536 		if (error != 0)
537 			break;
538 		td->td_retval[0] = OFLAGS(fp->f_flag);
539 		fdrop(fp, td);
540 		break;
541 
542 	case F_SETFL:
543 		error = fget_unlocked(fdp, fd,
544 		    cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL);
545 		if (error != 0)
546 			break;
547 		do {
548 			tmp = flg = fp->f_flag;
549 			tmp &= ~FCNTLFLAGS;
550 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
551 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
552 		tmp = fp->f_flag & FNONBLOCK;
553 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
554 		if (error != 0) {
555 			fdrop(fp, td);
556 			break;
557 		}
558 		tmp = fp->f_flag & FASYNC;
559 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
560 		if (error == 0) {
561 			fdrop(fp, td);
562 			break;
563 		}
564 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
565 		tmp = 0;
566 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
567 		fdrop(fp, td);
568 		break;
569 
570 	case F_GETOWN:
571 		error = fget_unlocked(fdp, fd,
572 		    cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL);
573 		if (error != 0)
574 			break;
575 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
576 		if (error == 0)
577 			td->td_retval[0] = tmp;
578 		fdrop(fp, td);
579 		break;
580 
581 	case F_SETOWN:
582 		error = fget_unlocked(fdp, fd,
583 		    cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL);
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)
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 		cap_rights_init(&rights, CAP_FLOCK);
605 		error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL);
606 		if (error != 0)
607 			break;
608 		if (fp->f_type != DTYPE_VNODE) {
609 			error = EBADF;
610 			fdrop(fp, td);
611 			break;
612 		}
613 
614 		flp = (struct flock *)arg;
615 		if (flp->l_whence == SEEK_CUR) {
616 			foffset = foffset_get(fp);
617 			if (foffset < 0 ||
618 			    (flp->l_start > 0 &&
619 			     foffset > OFF_MAX - flp->l_start)) {
620 				FILEDESC_SUNLOCK(fdp);
621 				error = EOVERFLOW;
622 				fdrop(fp, td);
623 				break;
624 			}
625 			flp->l_start += foffset;
626 		}
627 
628 		vp = fp->f_vnode;
629 		switch (flp->l_type) {
630 		case F_RDLCK:
631 			if ((fp->f_flag & FREAD) == 0) {
632 				error = EBADF;
633 				break;
634 			}
635 			PROC_LOCK(p->p_leader);
636 			p->p_leader->p_flag |= P_ADVLOCK;
637 			PROC_UNLOCK(p->p_leader);
638 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
639 			    flp, flg);
640 			break;
641 		case F_WRLCK:
642 			if ((fp->f_flag & FWRITE) == 0) {
643 				error = EBADF;
644 				break;
645 			}
646 			PROC_LOCK(p->p_leader);
647 			p->p_leader->p_flag |= P_ADVLOCK;
648 			PROC_UNLOCK(p->p_leader);
649 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
650 			    flp, flg);
651 			break;
652 		case F_UNLCK:
653 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
654 			    flp, flg);
655 			break;
656 		case F_UNLCKSYS:
657 			/*
658 			 * Temporary api for testing remote lock
659 			 * infrastructure.
660 			 */
661 			if (flg != F_REMOTE) {
662 				error = EINVAL;
663 				break;
664 			}
665 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
666 			    F_UNLCKSYS, flp, flg);
667 			break;
668 		default:
669 			error = EINVAL;
670 			break;
671 		}
672 		if (error != 0 || flp->l_type == F_UNLCK ||
673 		    flp->l_type == F_UNLCKSYS) {
674 			fdrop(fp, td);
675 			break;
676 		}
677 
678 		/*
679 		 * Check for a race with close.
680 		 *
681 		 * The vnode is now advisory locked (or unlocked, but this case
682 		 * is not really important) as the caller requested.
683 		 * We had to drop the filedesc lock, so we need to recheck if
684 		 * the descriptor is still valid, because if it was closed
685 		 * in the meantime we need to remove advisory lock from the
686 		 * vnode - close on any descriptor leading to an advisory
687 		 * locked vnode, removes that lock.
688 		 * We will return 0 on purpose in that case, as the result of
689 		 * successful advisory lock might have been externally visible
690 		 * already. This is fine - effectively we pretend to the caller
691 		 * that the closing thread was a bit slower and that the
692 		 * advisory lock succeeded before the close.
693 		 */
694 		error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL);
695 		if (error != 0) {
696 			fdrop(fp, td);
697 			break;
698 		}
699 		if (fp != fp2) {
700 			flp->l_whence = SEEK_SET;
701 			flp->l_start = 0;
702 			flp->l_len = 0;
703 			flp->l_type = F_UNLCK;
704 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
705 			    F_UNLCK, flp, F_POSIX);
706 		}
707 		fdrop(fp, td);
708 		fdrop(fp2, td);
709 		break;
710 
711 	case F_GETLK:
712 		error = fget_unlocked(fdp, fd,
713 		    cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL);
714 		if (error != 0)
715 			break;
716 		if (fp->f_type != DTYPE_VNODE) {
717 			error = EBADF;
718 			fdrop(fp, td);
719 			break;
720 		}
721 		flp = (struct flock *)arg;
722 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
723 		    flp->l_type != F_UNLCK) {
724 			error = EINVAL;
725 			fdrop(fp, td);
726 			break;
727 		}
728 		if (flp->l_whence == SEEK_CUR) {
729 			foffset = foffset_get(fp);
730 			if ((flp->l_start > 0 &&
731 			    foffset > OFF_MAX - flp->l_start) ||
732 			    (flp->l_start < 0 &&
733 			     foffset < OFF_MIN - flp->l_start)) {
734 				FILEDESC_SUNLOCK(fdp);
735 				error = EOVERFLOW;
736 				fdrop(fp, td);
737 				break;
738 			}
739 			flp->l_start += foffset;
740 		}
741 		vp = fp->f_vnode;
742 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
743 		    F_POSIX);
744 		fdrop(fp, td);
745 		break;
746 
747 	case F_RDAHEAD:
748 		arg = arg ? 128 * 1024: 0;
749 		/* FALLTHROUGH */
750 	case F_READAHEAD:
751 		error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL);
752 		if (error != 0)
753 			break;
754 		if (fp->f_type != DTYPE_VNODE) {
755 			fdrop(fp, td);
756 			error = EBADF;
757 			break;
758 		}
759 		vp = fp->f_vnode;
760 		/*
761 		 * Exclusive lock synchronizes against f_seqcount reads and
762 		 * writes in sequential_heuristic().
763 		 */
764 		error = vn_lock(vp, LK_EXCLUSIVE);
765 		if (error != 0) {
766 			fdrop(fp, td);
767 			break;
768 		}
769 		if (arg >= 0) {
770 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
771 			fp->f_seqcount = (arg + bsize - 1) / bsize;
772 			atomic_set_int(&fp->f_flag, FRDAHEAD);
773 		} else {
774 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
775 		}
776 		VOP_UNLOCK(vp, 0);
777 		fdrop(fp, td);
778 		break;
779 
780 	default:
781 		error = EINVAL;
782 		break;
783 	}
784 	return (error);
785 }
786 
787 static int
788 getmaxfd(struct proc *p)
789 {
790 	int maxfd;
791 
792 	PROC_LOCK(p);
793 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
794 	PROC_UNLOCK(p);
795 
796 	return (maxfd);
797 }
798 
799 /*
800  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
801  */
802 static int
803 do_dup(struct thread *td, int flags, int old, int new,
804     register_t *retval)
805 {
806 	struct filedesc *fdp;
807 	struct filedescent *oldfde, *newfde;
808 	struct proc *p;
809 	struct file *fp;
810 	struct file *delfp;
811 	int error, maxfd;
812 
813 	p = td->td_proc;
814 	fdp = p->p_fd;
815 
816 	/*
817 	 * Verify we have a valid descriptor to dup from and possibly to
818 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
819 	 * return EINVAL when the new descriptor is out of bounds.
820 	 */
821 	if (old < 0)
822 		return (EBADF);
823 	if (new < 0)
824 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
825 	maxfd = getmaxfd(p);
826 	if (new >= maxfd)
827 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
828 
829 	FILEDESC_XLOCK(fdp);
830 	if (fget_locked(fdp, old) == NULL) {
831 		FILEDESC_XUNLOCK(fdp);
832 		return (EBADF);
833 	}
834 	oldfde = &fdp->fd_ofiles[old];
835 	if (flags & DUP_FIXED && old == new) {
836 		*retval = new;
837 		if (flags & DUP_CLOEXEC)
838 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
839 		FILEDESC_XUNLOCK(fdp);
840 		return (0);
841 	}
842 	fp = oldfde->fde_file;
843 	fhold(fp);
844 
845 	/*
846 	 * If the caller specified a file descriptor, make sure the file
847 	 * table is large enough to hold it, and grab it.  Otherwise, just
848 	 * allocate a new descriptor the usual way.
849 	 */
850 	if (flags & DUP_FIXED) {
851 		if (new >= fdp->fd_nfiles) {
852 			/*
853 			 * The resource limits are here instead of e.g.
854 			 * fdalloc(), because the file descriptor table may be
855 			 * shared between processes, so we can't really use
856 			 * racct_add()/racct_sub().  Instead of counting the
857 			 * number of actually allocated descriptors, just put
858 			 * the limit on the size of the file descriptor table.
859 			 */
860 #ifdef RACCT
861 			PROC_LOCK(p);
862 			error = racct_set(p, RACCT_NOFILE, new + 1);
863 			PROC_UNLOCK(p);
864 			if (error != 0) {
865 				FILEDESC_XUNLOCK(fdp);
866 				fdrop(fp, td);
867 				return (EMFILE);
868 			}
869 #endif
870 			fdgrowtable_exp(fdp, new + 1);
871 			oldfde = &fdp->fd_ofiles[old];
872 		}
873 		newfde = &fdp->fd_ofiles[new];
874 		if (newfde->fde_file == NULL)
875 			fdused(fdp, new);
876 	} else {
877 		if ((error = fdalloc(td, new, &new)) != 0) {
878 			FILEDESC_XUNLOCK(fdp);
879 			fdrop(fp, td);
880 			return (error);
881 		}
882 		newfde = &fdp->fd_ofiles[new];
883 	}
884 
885 	KASSERT(fp == oldfde->fde_file, ("old fd has been modified"));
886 	KASSERT(old != new, ("new fd is same as old"));
887 
888 	delfp = newfde->fde_file;
889 
890 	/*
891 	 * Duplicate the source descriptor.
892 	 */
893 #ifdef CAPABILITIES
894 	seq_write_begin(&newfde->fde_seq);
895 #endif
896 	filecaps_free(&newfde->fde_caps);
897 	memcpy(newfde, oldfde, fde_change_size);
898 	filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
899 	if ((flags & DUP_CLOEXEC) != 0)
900 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
901 	else
902 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
903 #ifdef CAPABILITIES
904 	seq_write_end(&newfde->fde_seq);
905 #endif
906 	*retval = new;
907 
908 	if (delfp != NULL) {
909 		(void) closefp(fdp, new, delfp, td, 1);
910 		/* closefp() drops the FILEDESC lock for us. */
911 	} else {
912 		FILEDESC_XUNLOCK(fdp);
913 	}
914 
915 	return (0);
916 }
917 
918 /*
919  * If sigio is on the list associated with a process or process group,
920  * disable signalling from the device, remove sigio from the list and
921  * free sigio.
922  */
923 void
924 funsetown(struct sigio **sigiop)
925 {
926 	struct sigio *sigio;
927 
928 	SIGIO_LOCK();
929 	sigio = *sigiop;
930 	if (sigio == NULL) {
931 		SIGIO_UNLOCK();
932 		return;
933 	}
934 	*(sigio->sio_myref) = NULL;
935 	if ((sigio)->sio_pgid < 0) {
936 		struct pgrp *pg = (sigio)->sio_pgrp;
937 		PGRP_LOCK(pg);
938 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
939 			     sigio, sio_pgsigio);
940 		PGRP_UNLOCK(pg);
941 	} else {
942 		struct proc *p = (sigio)->sio_proc;
943 		PROC_LOCK(p);
944 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
945 			     sigio, sio_pgsigio);
946 		PROC_UNLOCK(p);
947 	}
948 	SIGIO_UNLOCK();
949 	crfree(sigio->sio_ucred);
950 	free(sigio, M_SIGIO);
951 }
952 
953 /*
954  * Free a list of sigio structures.
955  * We only need to lock the SIGIO_LOCK because we have made ourselves
956  * inaccessible to callers of fsetown and therefore do not need to lock
957  * the proc or pgrp struct for the list manipulation.
958  */
959 void
960 funsetownlst(struct sigiolst *sigiolst)
961 {
962 	struct proc *p;
963 	struct pgrp *pg;
964 	struct sigio *sigio;
965 
966 	sigio = SLIST_FIRST(sigiolst);
967 	if (sigio == NULL)
968 		return;
969 	p = NULL;
970 	pg = NULL;
971 
972 	/*
973 	 * Every entry of the list should belong
974 	 * to a single proc or pgrp.
975 	 */
976 	if (sigio->sio_pgid < 0) {
977 		pg = sigio->sio_pgrp;
978 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
979 	} else /* if (sigio->sio_pgid > 0) */ {
980 		p = sigio->sio_proc;
981 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
982 	}
983 
984 	SIGIO_LOCK();
985 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
986 		*(sigio->sio_myref) = NULL;
987 		if (pg != NULL) {
988 			KASSERT(sigio->sio_pgid < 0,
989 			    ("Proc sigio in pgrp sigio list"));
990 			KASSERT(sigio->sio_pgrp == pg,
991 			    ("Bogus pgrp in sigio list"));
992 			PGRP_LOCK(pg);
993 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
994 			    sio_pgsigio);
995 			PGRP_UNLOCK(pg);
996 		} else /* if (p != NULL) */ {
997 			KASSERT(sigio->sio_pgid > 0,
998 			    ("Pgrp sigio in proc sigio list"));
999 			KASSERT(sigio->sio_proc == p,
1000 			    ("Bogus proc in sigio list"));
1001 			PROC_LOCK(p);
1002 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1003 			    sio_pgsigio);
1004 			PROC_UNLOCK(p);
1005 		}
1006 		SIGIO_UNLOCK();
1007 		crfree(sigio->sio_ucred);
1008 		free(sigio, M_SIGIO);
1009 		SIGIO_LOCK();
1010 	}
1011 	SIGIO_UNLOCK();
1012 }
1013 
1014 /*
1015  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1016  *
1017  * After permission checking, add a sigio structure to the sigio list for
1018  * the process or process group.
1019  */
1020 int
1021 fsetown(pid_t pgid, struct sigio **sigiop)
1022 {
1023 	struct proc *proc;
1024 	struct pgrp *pgrp;
1025 	struct sigio *sigio;
1026 	int ret;
1027 
1028 	if (pgid == 0) {
1029 		funsetown(sigiop);
1030 		return (0);
1031 	}
1032 
1033 	ret = 0;
1034 
1035 	/* Allocate and fill in the new sigio out of locks. */
1036 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1037 	sigio->sio_pgid = pgid;
1038 	sigio->sio_ucred = crhold(curthread->td_ucred);
1039 	sigio->sio_myref = sigiop;
1040 
1041 	sx_slock(&proctree_lock);
1042 	if (pgid > 0) {
1043 		proc = pfind(pgid);
1044 		if (proc == NULL) {
1045 			ret = ESRCH;
1046 			goto fail;
1047 		}
1048 
1049 		/*
1050 		 * Policy - Don't allow a process to FSETOWN a process
1051 		 * in another session.
1052 		 *
1053 		 * Remove this test to allow maximum flexibility or
1054 		 * restrict FSETOWN to the current process or process
1055 		 * group for maximum safety.
1056 		 */
1057 		PROC_UNLOCK(proc);
1058 		if (proc->p_session != curthread->td_proc->p_session) {
1059 			ret = EPERM;
1060 			goto fail;
1061 		}
1062 
1063 		pgrp = NULL;
1064 	} else /* if (pgid < 0) */ {
1065 		pgrp = pgfind(-pgid);
1066 		if (pgrp == NULL) {
1067 			ret = ESRCH;
1068 			goto fail;
1069 		}
1070 		PGRP_UNLOCK(pgrp);
1071 
1072 		/*
1073 		 * Policy - Don't allow a process to FSETOWN a process
1074 		 * in another session.
1075 		 *
1076 		 * Remove this test to allow maximum flexibility or
1077 		 * restrict FSETOWN to the current process or process
1078 		 * group for maximum safety.
1079 		 */
1080 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1081 			ret = EPERM;
1082 			goto fail;
1083 		}
1084 
1085 		proc = NULL;
1086 	}
1087 	funsetown(sigiop);
1088 	if (pgid > 0) {
1089 		PROC_LOCK(proc);
1090 		/*
1091 		 * Since funsetownlst() is called without the proctree
1092 		 * locked, we need to check for P_WEXIT.
1093 		 * XXX: is ESRCH correct?
1094 		 */
1095 		if ((proc->p_flag & P_WEXIT) != 0) {
1096 			PROC_UNLOCK(proc);
1097 			ret = ESRCH;
1098 			goto fail;
1099 		}
1100 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1101 		sigio->sio_proc = proc;
1102 		PROC_UNLOCK(proc);
1103 	} else {
1104 		PGRP_LOCK(pgrp);
1105 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1106 		sigio->sio_pgrp = pgrp;
1107 		PGRP_UNLOCK(pgrp);
1108 	}
1109 	sx_sunlock(&proctree_lock);
1110 	SIGIO_LOCK();
1111 	*sigiop = sigio;
1112 	SIGIO_UNLOCK();
1113 	return (0);
1114 
1115 fail:
1116 	sx_sunlock(&proctree_lock);
1117 	crfree(sigio->sio_ucred);
1118 	free(sigio, M_SIGIO);
1119 	return (ret);
1120 }
1121 
1122 /*
1123  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1124  */
1125 pid_t
1126 fgetown(sigiop)
1127 	struct sigio **sigiop;
1128 {
1129 	pid_t pgid;
1130 
1131 	SIGIO_LOCK();
1132 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1133 	SIGIO_UNLOCK();
1134 	return (pgid);
1135 }
1136 
1137 /*
1138  * Function drops the filedesc lock on return.
1139  */
1140 static int
1141 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1142     int holdleaders)
1143 {
1144 	int error;
1145 
1146 	FILEDESC_XLOCK_ASSERT(fdp);
1147 
1148 	if (holdleaders) {
1149 		if (td->td_proc->p_fdtol != NULL) {
1150 			/*
1151 			 * Ask fdfree() to sleep to ensure that all relevant
1152 			 * process leaders can be traversed in closef().
1153 			 */
1154 			fdp->fd_holdleaderscount++;
1155 		} else {
1156 			holdleaders = 0;
1157 		}
1158 	}
1159 
1160 	/*
1161 	 * We now hold the fp reference that used to be owned by the
1162 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1163 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1164 	 * added, and deleteing a knote for the new fd.
1165 	 */
1166 	knote_fdclose(td, fd);
1167 
1168 	/*
1169 	 * We need to notify mqueue if the object is of type mqueue.
1170 	 */
1171 	if (fp->f_type == DTYPE_MQUEUE)
1172 		mq_fdclose(td, fd, fp);
1173 	FILEDESC_XUNLOCK(fdp);
1174 
1175 	error = closef(fp, td);
1176 	if (holdleaders) {
1177 		FILEDESC_XLOCK(fdp);
1178 		fdp->fd_holdleaderscount--;
1179 		if (fdp->fd_holdleaderscount == 0 &&
1180 		    fdp->fd_holdleaderswakeup != 0) {
1181 			fdp->fd_holdleaderswakeup = 0;
1182 			wakeup(&fdp->fd_holdleaderscount);
1183 		}
1184 		FILEDESC_XUNLOCK(fdp);
1185 	}
1186 	return (error);
1187 }
1188 
1189 /*
1190  * Close a file descriptor.
1191  */
1192 #ifndef _SYS_SYSPROTO_H_
1193 struct close_args {
1194 	int     fd;
1195 };
1196 #endif
1197 /* ARGSUSED */
1198 int
1199 sys_close(td, uap)
1200 	struct thread *td;
1201 	struct close_args *uap;
1202 {
1203 
1204 	return (kern_close(td, uap->fd));
1205 }
1206 
1207 int
1208 kern_close(td, fd)
1209 	struct thread *td;
1210 	int fd;
1211 {
1212 	struct filedesc *fdp;
1213 	struct file *fp;
1214 
1215 	fdp = td->td_proc->p_fd;
1216 
1217 	AUDIT_SYSCLOSE(td, fd);
1218 
1219 	FILEDESC_XLOCK(fdp);
1220 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1221 		FILEDESC_XUNLOCK(fdp);
1222 		return (EBADF);
1223 	}
1224 	fdfree(fdp, fd);
1225 
1226 	/* closefp() drops the FILEDESC lock for us. */
1227 	return (closefp(fdp, fd, fp, td, 1));
1228 }
1229 
1230 /*
1231  * Close open file descriptors.
1232  */
1233 #ifndef _SYS_SYSPROTO_H_
1234 struct closefrom_args {
1235 	int	lowfd;
1236 };
1237 #endif
1238 /* ARGSUSED */
1239 int
1240 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1241 {
1242 	struct filedesc *fdp;
1243 	int fd;
1244 
1245 	fdp = td->td_proc->p_fd;
1246 	AUDIT_ARG_FD(uap->lowfd);
1247 
1248 	/*
1249 	 * Treat negative starting file descriptor values identical to
1250 	 * closefrom(0) which closes all files.
1251 	 */
1252 	if (uap->lowfd < 0)
1253 		uap->lowfd = 0;
1254 	FILEDESC_SLOCK(fdp);
1255 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1256 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1257 			FILEDESC_SUNLOCK(fdp);
1258 			(void)kern_close(td, fd);
1259 			FILEDESC_SLOCK(fdp);
1260 		}
1261 	}
1262 	FILEDESC_SUNLOCK(fdp);
1263 	return (0);
1264 }
1265 
1266 #if defined(COMPAT_43)
1267 /*
1268  * Return status information about a file descriptor.
1269  */
1270 #ifndef _SYS_SYSPROTO_H_
1271 struct ofstat_args {
1272 	int	fd;
1273 	struct	ostat *sb;
1274 };
1275 #endif
1276 /* ARGSUSED */
1277 int
1278 ofstat(struct thread *td, struct ofstat_args *uap)
1279 {
1280 	struct ostat oub;
1281 	struct stat ub;
1282 	int error;
1283 
1284 	error = kern_fstat(td, uap->fd, &ub);
1285 	if (error == 0) {
1286 		cvtstat(&ub, &oub);
1287 		error = copyout(&oub, uap->sb, sizeof(oub));
1288 	}
1289 	return (error);
1290 }
1291 #endif /* COMPAT_43 */
1292 
1293 /*
1294  * Return status information about a file descriptor.
1295  */
1296 #ifndef _SYS_SYSPROTO_H_
1297 struct fstat_args {
1298 	int	fd;
1299 	struct	stat *sb;
1300 };
1301 #endif
1302 /* ARGSUSED */
1303 int
1304 sys_fstat(struct thread *td, struct fstat_args *uap)
1305 {
1306 	struct stat ub;
1307 	int error;
1308 
1309 	error = kern_fstat(td, uap->fd, &ub);
1310 	if (error == 0)
1311 		error = copyout(&ub, uap->sb, sizeof(ub));
1312 	return (error);
1313 }
1314 
1315 int
1316 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1317 {
1318 	struct file *fp;
1319 	cap_rights_t rights;
1320 	int error;
1321 
1322 	AUDIT_ARG_FD(fd);
1323 
1324 	error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1325 	if (error != 0)
1326 		return (error);
1327 
1328 	AUDIT_ARG_FILE(td->td_proc, fp);
1329 
1330 	error = fo_stat(fp, sbp, td->td_ucred, td);
1331 	fdrop(fp, td);
1332 #ifdef KTRACE
1333 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1334 		ktrstat(sbp);
1335 #endif
1336 	return (error);
1337 }
1338 
1339 /*
1340  * Return status information about a file descriptor.
1341  */
1342 #ifndef _SYS_SYSPROTO_H_
1343 struct nfstat_args {
1344 	int	fd;
1345 	struct	nstat *sb;
1346 };
1347 #endif
1348 /* ARGSUSED */
1349 int
1350 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1351 {
1352 	struct nstat nub;
1353 	struct stat ub;
1354 	int error;
1355 
1356 	error = kern_fstat(td, uap->fd, &ub);
1357 	if (error == 0) {
1358 		cvtnstat(&ub, &nub);
1359 		error = copyout(&nub, uap->sb, sizeof(nub));
1360 	}
1361 	return (error);
1362 }
1363 
1364 /*
1365  * Return pathconf information about a file descriptor.
1366  */
1367 #ifndef _SYS_SYSPROTO_H_
1368 struct fpathconf_args {
1369 	int	fd;
1370 	int	name;
1371 };
1372 #endif
1373 /* ARGSUSED */
1374 int
1375 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1376 {
1377 	struct file *fp;
1378 	struct vnode *vp;
1379 	cap_rights_t rights;
1380 	int error;
1381 
1382 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1383 	if (error != 0)
1384 		return (error);
1385 
1386 	/* If asynchronous I/O is available, it works for all descriptors. */
1387 	if (uap->name == _PC_ASYNC_IO) {
1388 		td->td_retval[0] = async_io_version;
1389 		goto out;
1390 	}
1391 	vp = fp->f_vnode;
1392 	if (vp != NULL) {
1393 		vn_lock(vp, LK_SHARED | LK_RETRY);
1394 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1395 		VOP_UNLOCK(vp, 0);
1396 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1397 		if (uap->name != _PC_PIPE_BUF) {
1398 			error = EINVAL;
1399 		} else {
1400 			td->td_retval[0] = PIPE_BUF;
1401 			error = 0;
1402 		}
1403 	} else {
1404 		error = EOPNOTSUPP;
1405 	}
1406 out:
1407 	fdrop(fp, td);
1408 	return (error);
1409 }
1410 
1411 /*
1412  * Initialize filecaps structure.
1413  */
1414 void
1415 filecaps_init(struct filecaps *fcaps)
1416 {
1417 
1418 	bzero(fcaps, sizeof(*fcaps));
1419 	fcaps->fc_nioctls = -1;
1420 }
1421 
1422 /*
1423  * Copy filecaps structure allocating memory for ioctls array if needed.
1424  */
1425 void
1426 filecaps_copy(const struct filecaps *src, struct filecaps *dst)
1427 {
1428 	size_t size;
1429 
1430 	*dst = *src;
1431 	if (src->fc_ioctls != NULL) {
1432 		KASSERT(src->fc_nioctls > 0,
1433 		    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1434 
1435 		size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1436 		dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1437 		bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1438 	}
1439 }
1440 
1441 /*
1442  * Move filecaps structure to the new place and clear the old place.
1443  */
1444 void
1445 filecaps_move(struct filecaps *src, struct filecaps *dst)
1446 {
1447 
1448 	*dst = *src;
1449 	bzero(src, sizeof(*src));
1450 }
1451 
1452 /*
1453  * Fill the given filecaps structure with full rights.
1454  */
1455 static void
1456 filecaps_fill(struct filecaps *fcaps)
1457 {
1458 
1459 	CAP_ALL(&fcaps->fc_rights);
1460 	fcaps->fc_ioctls = NULL;
1461 	fcaps->fc_nioctls = -1;
1462 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1463 }
1464 
1465 /*
1466  * Free memory allocated within filecaps structure.
1467  */
1468 void
1469 filecaps_free(struct filecaps *fcaps)
1470 {
1471 
1472 	free(fcaps->fc_ioctls, M_FILECAPS);
1473 	bzero(fcaps, sizeof(*fcaps));
1474 }
1475 
1476 /*
1477  * Validate the given filecaps structure.
1478  */
1479 static void
1480 filecaps_validate(const struct filecaps *fcaps, const char *func)
1481 {
1482 
1483 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1484 	    ("%s: invalid rights", func));
1485 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1486 	    ("%s: invalid fcntls", func));
1487 	KASSERT(fcaps->fc_fcntls == 0 ||
1488 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1489 	    ("%s: fcntls without CAP_FCNTL", func));
1490 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1491 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1492 	    ("%s: invalid ioctls", func));
1493 	KASSERT(fcaps->fc_nioctls == 0 ||
1494 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1495 	    ("%s: ioctls without CAP_IOCTL", func));
1496 }
1497 
1498 static void
1499 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1500 {
1501 	int nfd1;
1502 
1503 	FILEDESC_XLOCK_ASSERT(fdp);
1504 
1505 	nfd1 = fdp->fd_nfiles * 2;
1506 	if (nfd1 < nfd)
1507 		nfd1 = nfd;
1508 	fdgrowtable(fdp, nfd1);
1509 }
1510 
1511 /*
1512  * Grow the file table to accomodate (at least) nfd descriptors.
1513  */
1514 static void
1515 fdgrowtable(struct filedesc *fdp, int nfd)
1516 {
1517 	struct filedesc0 *fdp0;
1518 	struct freetable *ft;
1519 	struct filedescent *ntable;
1520 	struct filedescent *otable;
1521 	int nnfiles, onfiles;
1522 	NDSLOTTYPE *nmap, *omap;
1523 
1524 	FILEDESC_XLOCK_ASSERT(fdp);
1525 
1526 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1527 
1528 	/* save old values */
1529 	onfiles = fdp->fd_nfiles;
1530 	otable = fdp->fd_ofiles;
1531 	omap = fdp->fd_map;
1532 
1533 	/* compute the size of the new table */
1534 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1535 	if (nnfiles <= onfiles)
1536 		/* the table is already large enough */
1537 		return;
1538 
1539 	/*
1540 	 * Allocate a new table.  We need enough space for the
1541 	 * file entries themselves and the struct freetable we will use
1542 	 * when we decommission the table and place it on the freelist.
1543 	 * We place the struct freetable in the middle so we don't have
1544 	 * to worry about padding.
1545 	 */
1546 	ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable),
1547 	    M_FILEDESC, M_ZERO | M_WAITOK);
1548 	/* copy the old data over and point at the new tables */
1549 	memcpy(ntable, otable, onfiles * sizeof(*otable));
1550 	fdp->fd_ofiles = ntable;
1551 
1552 	/*
1553 	 * Allocate a new map only if the old is not large enough.  It will
1554 	 * grow at a slower rate than the table as it can map more
1555 	 * entries than the table can hold.
1556 	 */
1557 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1558 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1559 		    M_ZERO | M_WAITOK);
1560 		/* copy over the old data and update the pointer */
1561 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1562 		fdp->fd_map = nmap;
1563 	}
1564 
1565 	/*
1566 	 * In order to have a valid pattern for fget_unlocked()
1567 	 * fdp->fd_nfiles must be the last member to be updated, otherwise
1568 	 * fget_unlocked() consumers may reference a new, higher value for
1569 	 * fdp->fd_nfiles before to access the fdp->fd_ofiles array,
1570 	 * resulting in OOB accesses.
1571 	 */
1572 	atomic_store_rel_int(&fdp->fd_nfiles, nnfiles);
1573 
1574 	/*
1575 	 * Do not free the old file table, as some threads may still
1576 	 * reference entries within it.  Instead, place it on a freelist
1577 	 * which will be processed when the struct filedesc is released.
1578 	 *
1579 	 * Note that if onfiles == NDFILE, we're dealing with the original
1580 	 * static allocation contained within (struct filedesc0 *)fdp,
1581 	 * which must not be freed.
1582 	 */
1583 	if (onfiles > NDFILE) {
1584 		ft = (struct freetable *)&otable[onfiles];
1585 		fdp0 = (struct filedesc0 *)fdp;
1586 		ft->ft_table = otable;
1587 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1588 	}
1589 	/*
1590 	 * The map does not have the same possibility of threads still
1591 	 * holding references to it.  So always free it as long as it
1592 	 * does not reference the original static allocation.
1593 	 */
1594 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1595 		free(omap, M_FILEDESC);
1596 }
1597 
1598 /*
1599  * Allocate a file descriptor for the process.
1600  */
1601 int
1602 fdalloc(struct thread *td, int minfd, int *result)
1603 {
1604 	struct proc *p = td->td_proc;
1605 	struct filedesc *fdp = p->p_fd;
1606 	int fd = -1, maxfd, allocfd;
1607 #ifdef RACCT
1608 	int error;
1609 #endif
1610 
1611 	FILEDESC_XLOCK_ASSERT(fdp);
1612 
1613 	if (fdp->fd_freefile > minfd)
1614 		minfd = fdp->fd_freefile;
1615 
1616 	maxfd = getmaxfd(p);
1617 
1618 	/*
1619 	 * Search the bitmap for a free descriptor starting at minfd.
1620 	 * If none is found, grow the file table.
1621 	 */
1622 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1623 	if (fd >= maxfd)
1624 		return (EMFILE);
1625 	if (fd >= fdp->fd_nfiles) {
1626 		allocfd = min(fd * 2, maxfd);
1627 #ifdef RACCT
1628 		PROC_LOCK(p);
1629 		error = racct_set(p, RACCT_NOFILE, allocfd);
1630 		PROC_UNLOCK(p);
1631 		if (error != 0)
1632 			return (EMFILE);
1633 #endif
1634 		/*
1635 		 * fd is already equal to first free descriptor >= minfd, so
1636 		 * we only need to grow the table and we are done.
1637 		 */
1638 		fdgrowtable_exp(fdp, allocfd);
1639 	}
1640 
1641 	/*
1642 	 * Perform some sanity checks, then mark the file descriptor as
1643 	 * used and return it to the caller.
1644 	 */
1645 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1646 	    ("invalid descriptor %d", fd));
1647 	KASSERT(!fdisused(fdp, fd),
1648 	    ("fd_first_free() returned non-free descriptor"));
1649 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1650 	    ("file descriptor isn't free"));
1651 	KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
1652 	fdused(fdp, fd);
1653 	*result = fd;
1654 	return (0);
1655 }
1656 
1657 /*
1658  * Allocate n file descriptors for the process.
1659  */
1660 int
1661 fdallocn(struct thread *td, int minfd, int *fds, int n)
1662 {
1663 	struct proc *p = td->td_proc;
1664 	struct filedesc *fdp = p->p_fd;
1665 	int i;
1666 
1667 	FILEDESC_XLOCK_ASSERT(fdp);
1668 
1669 	for (i = 0; i < n; i++)
1670 		if (fdalloc(td, 0, &fds[i]) != 0)
1671 			break;
1672 
1673 	if (i < n) {
1674 		for (i--; i >= 0; i--)
1675 			fdunused(fdp, fds[i]);
1676 		return (EMFILE);
1677 	}
1678 
1679 	return (0);
1680 }
1681 
1682 /*
1683  * Create a new open file structure and allocate a file decriptor for the
1684  * process that refers to it.  We add one reference to the file for the
1685  * descriptor table and one reference for resultfp. This is to prevent us
1686  * being preempted and the entry in the descriptor table closed after we
1687  * release the FILEDESC lock.
1688  */
1689 int
1690 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1691 {
1692 	struct file *fp;
1693 	int error, fd;
1694 
1695 	error = falloc_noinstall(td, &fp);
1696 	if (error)
1697 		return (error);		/* no reference held on error */
1698 
1699 	error = finstall(td, fp, &fd, flags, NULL);
1700 	if (error) {
1701 		fdrop(fp, td);		/* one reference (fp only) */
1702 		return (error);
1703 	}
1704 
1705 	if (resultfp != NULL)
1706 		*resultfp = fp;		/* copy out result */
1707 	else
1708 		fdrop(fp, td);		/* release local reference */
1709 
1710 	if (resultfd != NULL)
1711 		*resultfd = fd;
1712 
1713 	return (0);
1714 }
1715 
1716 /*
1717  * Create a new open file structure without allocating a file descriptor.
1718  */
1719 int
1720 falloc_noinstall(struct thread *td, struct file **resultfp)
1721 {
1722 	struct file *fp;
1723 	int maxuserfiles = maxfiles - (maxfiles / 20);
1724 	static struct timeval lastfail;
1725 	static int curfail;
1726 
1727 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1728 
1729 	if ((openfiles >= maxuserfiles &&
1730 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1731 	    openfiles >= maxfiles) {
1732 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1733 			printf("kern.maxfiles limit exceeded by uid %i, "
1734 			    "please see tuning(7).\n", td->td_ucred->cr_ruid);
1735 		}
1736 		return (ENFILE);
1737 	}
1738 	atomic_add_int(&openfiles, 1);
1739 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1740 	refcount_init(&fp->f_count, 1);
1741 	fp->f_cred = crhold(td->td_ucred);
1742 	fp->f_ops = &badfileops;
1743 	*resultfp = fp;
1744 	return (0);
1745 }
1746 
1747 /*
1748  * Install a file in a file descriptor table.
1749  */
1750 int
1751 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1752     struct filecaps *fcaps)
1753 {
1754 	struct filedesc *fdp = td->td_proc->p_fd;
1755 	struct filedescent *fde;
1756 	int error;
1757 
1758 	KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1759 	KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1760 	if (fcaps != NULL)
1761 		filecaps_validate(fcaps, __func__);
1762 
1763 	FILEDESC_XLOCK(fdp);
1764 	if ((error = fdalloc(td, 0, fd))) {
1765 		FILEDESC_XUNLOCK(fdp);
1766 		return (error);
1767 	}
1768 	fhold(fp);
1769 	fde = &fdp->fd_ofiles[*fd];
1770 #ifdef CAPABILITIES
1771 	seq_write_begin(&fde->fde_seq);
1772 #endif
1773 	fde->fde_file = fp;
1774 	if ((flags & O_CLOEXEC) != 0)
1775 		fde->fde_flags |= UF_EXCLOSE;
1776 	if (fcaps != NULL)
1777 		filecaps_move(fcaps, &fde->fde_caps);
1778 	else
1779 		filecaps_fill(&fde->fde_caps);
1780 #ifdef CAPABILITIES
1781 	seq_write_end(&fde->fde_seq);
1782 #endif
1783 	FILEDESC_XUNLOCK(fdp);
1784 	return (0);
1785 }
1786 
1787 /*
1788  * Build a new filedesc structure from another.
1789  * Copy the current, root, and jail root vnode references.
1790  */
1791 struct filedesc *
1792 fdinit(struct filedesc *fdp)
1793 {
1794 	struct filedesc0 *newfdp;
1795 
1796 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1797 	FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1798 	if (fdp != NULL) {
1799 		FILEDESC_SLOCK(fdp);
1800 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1801 		if (newfdp->fd_fd.fd_cdir)
1802 			VREF(newfdp->fd_fd.fd_cdir);
1803 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1804 		if (newfdp->fd_fd.fd_rdir)
1805 			VREF(newfdp->fd_fd.fd_rdir);
1806 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1807 		if (newfdp->fd_fd.fd_jdir)
1808 			VREF(newfdp->fd_fd.fd_jdir);
1809 		FILEDESC_SUNLOCK(fdp);
1810 	}
1811 
1812 	/* Create the file descriptor table. */
1813 	newfdp->fd_fd.fd_refcnt = 1;
1814 	newfdp->fd_fd.fd_holdcnt = 1;
1815 	newfdp->fd_fd.fd_cmask = CMASK;
1816 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1817 	newfdp->fd_fd.fd_nfiles = NDFILE;
1818 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1819 	newfdp->fd_fd.fd_lastfile = -1;
1820 	return (&newfdp->fd_fd);
1821 }
1822 
1823 static struct filedesc *
1824 fdhold(struct proc *p)
1825 {
1826 	struct filedesc *fdp;
1827 
1828 	mtx_lock(&fdesc_mtx);
1829 	fdp = p->p_fd;
1830 	if (fdp != NULL)
1831 		fdp->fd_holdcnt++;
1832 	mtx_unlock(&fdesc_mtx);
1833 	return (fdp);
1834 }
1835 
1836 static void
1837 fddrop(struct filedesc *fdp)
1838 {
1839 	struct filedesc0 *fdp0;
1840 	struct freetable *ft;
1841 	int i;
1842 
1843 	mtx_lock(&fdesc_mtx);
1844 	i = --fdp->fd_holdcnt;
1845 	mtx_unlock(&fdesc_mtx);
1846 	if (i > 0)
1847 		return;
1848 
1849 	FILEDESC_LOCK_DESTROY(fdp);
1850 	fdp0 = (struct filedesc0 *)fdp;
1851 	while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1852 		SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1853 		free(ft->ft_table, M_FILEDESC);
1854 	}
1855 	free(fdp, M_FILEDESC);
1856 }
1857 
1858 /*
1859  * Share a filedesc structure.
1860  */
1861 struct filedesc *
1862 fdshare(struct filedesc *fdp)
1863 {
1864 
1865 	FILEDESC_XLOCK(fdp);
1866 	fdp->fd_refcnt++;
1867 	FILEDESC_XUNLOCK(fdp);
1868 	return (fdp);
1869 }
1870 
1871 /*
1872  * Unshare a filedesc structure, if necessary by making a copy
1873  */
1874 void
1875 fdunshare(struct thread *td)
1876 {
1877 	struct filedesc *tmp;
1878 	struct proc *p = td->td_proc;
1879 
1880 	if (p->p_fd->fd_refcnt == 1)
1881 		return;
1882 
1883 	tmp = fdcopy(p->p_fd);
1884 	fdescfree(td);
1885 	p->p_fd = tmp;
1886 }
1887 
1888 /*
1889  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1890  * this is to ease callers, not catch errors.
1891  */
1892 struct filedesc *
1893 fdcopy(struct filedesc *fdp)
1894 {
1895 	struct filedesc *newfdp;
1896 	struct filedescent *nfde, *ofde;
1897 	int i;
1898 
1899 	/* Certain daemons might not have file descriptors. */
1900 	if (fdp == NULL)
1901 		return (NULL);
1902 
1903 	newfdp = fdinit(fdp);
1904 	FILEDESC_SLOCK(fdp);
1905 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1906 		FILEDESC_SUNLOCK(fdp);
1907 		FILEDESC_XLOCK(newfdp);
1908 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1909 		FILEDESC_XUNLOCK(newfdp);
1910 		FILEDESC_SLOCK(fdp);
1911 	}
1912 	/* copy all passable descriptors (i.e. not kqueue) */
1913 	newfdp->fd_freefile = -1;
1914 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1915 		ofde = &fdp->fd_ofiles[i];
1916 		if (fdisused(fdp, i) &&
1917 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) &&
1918 		    ofde->fde_file->f_ops != &badfileops) {
1919 			nfde = &newfdp->fd_ofiles[i];
1920 			*nfde = *ofde;
1921 			filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
1922 			fhold(nfde->fde_file);
1923 			newfdp->fd_lastfile = i;
1924 		} else {
1925 			if (newfdp->fd_freefile == -1)
1926 				newfdp->fd_freefile = i;
1927 		}
1928 	}
1929 	newfdp->fd_cmask = fdp->fd_cmask;
1930 	FILEDESC_SUNLOCK(fdp);
1931 	FILEDESC_XLOCK(newfdp);
1932 	for (i = 0; i <= newfdp->fd_lastfile; ++i) {
1933 		if (newfdp->fd_ofiles[i].fde_file != NULL)
1934 			fdused(newfdp, i);
1935 	}
1936 	if (newfdp->fd_freefile == -1)
1937 		newfdp->fd_freefile = i;
1938 	FILEDESC_XUNLOCK(newfdp);
1939 	return (newfdp);
1940 }
1941 
1942 /*
1943  * Release a filedesc structure.
1944  */
1945 void
1946 fdescfree(struct thread *td)
1947 {
1948 	struct filedesc *fdp;
1949 	int i;
1950 	struct filedesc_to_leader *fdtol;
1951 	struct file *fp;
1952 	struct vnode *cdir, *jdir, *rdir, *vp;
1953 	struct flock lf;
1954 
1955 	/* Certain daemons might not have file descriptors. */
1956 	fdp = td->td_proc->p_fd;
1957 	if (fdp == NULL)
1958 		return;
1959 
1960 #ifdef RACCT
1961 	PROC_LOCK(td->td_proc);
1962 	racct_set(td->td_proc, RACCT_NOFILE, 0);
1963 	PROC_UNLOCK(td->td_proc);
1964 #endif
1965 
1966 	/* Check for special need to clear POSIX style locks */
1967 	fdtol = td->td_proc->p_fdtol;
1968 	if (fdtol != NULL) {
1969 		FILEDESC_XLOCK(fdp);
1970 		KASSERT(fdtol->fdl_refcount > 0,
1971 		    ("filedesc_to_refcount botch: fdl_refcount=%d",
1972 		    fdtol->fdl_refcount));
1973 		if (fdtol->fdl_refcount == 1 &&
1974 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1975 			for (i = 0; i <= fdp->fd_lastfile; i++) {
1976 				fp = fdp->fd_ofiles[i].fde_file;
1977 				if (fp == NULL || fp->f_type != DTYPE_VNODE)
1978 					continue;
1979 				fhold(fp);
1980 				FILEDESC_XUNLOCK(fdp);
1981 				lf.l_whence = SEEK_SET;
1982 				lf.l_start = 0;
1983 				lf.l_len = 0;
1984 				lf.l_type = F_UNLCK;
1985 				vp = fp->f_vnode;
1986 				(void) VOP_ADVLOCK(vp,
1987 				    (caddr_t)td->td_proc->p_leader, F_UNLCK,
1988 				    &lf, F_POSIX);
1989 				FILEDESC_XLOCK(fdp);
1990 				fdrop(fp, td);
1991 			}
1992 		}
1993 	retry:
1994 		if (fdtol->fdl_refcount == 1) {
1995 			if (fdp->fd_holdleaderscount > 0 &&
1996 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1997 				/*
1998 				 * close() or do_dup() has cleared a reference
1999 				 * in a shared file descriptor table.
2000 				 */
2001 				fdp->fd_holdleaderswakeup = 1;
2002 				sx_sleep(&fdp->fd_holdleaderscount,
2003 				    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2004 				goto retry;
2005 			}
2006 			if (fdtol->fdl_holdcount > 0) {
2007 				/*
2008 				 * Ensure that fdtol->fdl_leader remains
2009 				 * valid in closef().
2010 				 */
2011 				fdtol->fdl_wakeup = 1;
2012 				sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2013 				    "fdlhold", 0);
2014 				goto retry;
2015 			}
2016 		}
2017 		fdtol->fdl_refcount--;
2018 		if (fdtol->fdl_refcount == 0 &&
2019 		    fdtol->fdl_holdcount == 0) {
2020 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2021 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2022 		} else
2023 			fdtol = NULL;
2024 		td->td_proc->p_fdtol = NULL;
2025 		FILEDESC_XUNLOCK(fdp);
2026 		if (fdtol != NULL)
2027 			free(fdtol, M_FILEDESC_TO_LEADER);
2028 	}
2029 
2030 	mtx_lock(&fdesc_mtx);
2031 	td->td_proc->p_fd = NULL;
2032 	mtx_unlock(&fdesc_mtx);
2033 
2034 	FILEDESC_XLOCK(fdp);
2035 	i = --fdp->fd_refcnt;
2036 	if (i > 0) {
2037 		FILEDESC_XUNLOCK(fdp);
2038 		return;
2039 	}
2040 
2041 	cdir = fdp->fd_cdir;
2042 	fdp->fd_cdir = NULL;
2043 	rdir = fdp->fd_rdir;
2044 	fdp->fd_rdir = NULL;
2045 	jdir = fdp->fd_jdir;
2046 	fdp->fd_jdir = NULL;
2047 	FILEDESC_XUNLOCK(fdp);
2048 
2049 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2050 		fp = fdp->fd_ofiles[i].fde_file;
2051 		if (fp != NULL) {
2052 			fdfree_last(fdp, i);
2053 			(void) closef(fp, td);
2054 		}
2055 	}
2056 
2057 	if (fdp->fd_nfiles > NDFILE)
2058 		free(fdp->fd_ofiles, M_FILEDESC);
2059 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2060 		free(fdp->fd_map, M_FILEDESC);
2061 
2062 	if (cdir != NULL)
2063 		vrele(cdir);
2064 	if (rdir != NULL)
2065 		vrele(rdir);
2066 	if (jdir != NULL)
2067 		vrele(jdir);
2068 
2069 	fddrop(fdp);
2070 }
2071 
2072 /*
2073  * For setugid programs, we don't want to people to use that setugidness
2074  * to generate error messages which write to a file which otherwise would
2075  * otherwise be off-limits to the process.  We check for filesystems where
2076  * the vnode can change out from under us after execve (like [lin]procfs).
2077  *
2078  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2079  * sufficient.  We also don't check for setugidness since we know we are.
2080  */
2081 static bool
2082 is_unsafe(struct file *fp)
2083 {
2084 	struct vnode *vp;
2085 
2086 	if (fp->f_type != DTYPE_VNODE)
2087 		return (false);
2088 
2089 	vp = fp->f_vnode;
2090 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2091 }
2092 
2093 /*
2094  * Make this setguid thing safe, if at all possible.
2095  */
2096 void
2097 fdsetugidsafety(struct thread *td)
2098 {
2099 	struct filedesc *fdp;
2100 	struct file *fp;
2101 	int i;
2102 
2103 	fdp = td->td_proc->p_fd;
2104 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2105 	MPASS(fdp->fd_nfiles >= 3);
2106 	for (i = 0; i <= 2; i++) {
2107 		fp = fdp->fd_ofiles[i].fde_file;
2108 		if (fp != NULL && is_unsafe(fp)) {
2109 			FILEDESC_XLOCK(fdp);
2110 			knote_fdclose(td, i);
2111 			/*
2112 			 * NULL-out descriptor prior to close to avoid
2113 			 * a race while close blocks.
2114 			 */
2115 			fdfree(fdp, i);
2116 			FILEDESC_XUNLOCK(fdp);
2117 			(void) closef(fp, td);
2118 		}
2119 	}
2120 }
2121 
2122 /*
2123  * If a specific file object occupies a specific file descriptor, close the
2124  * file descriptor entry and drop a reference on the file object.  This is a
2125  * convenience function to handle a subsequent error in a function that calls
2126  * falloc() that handles the race that another thread might have closed the
2127  * file descriptor out from under the thread creating the file object.
2128  */
2129 void
2130 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2131 {
2132 
2133 	FILEDESC_XLOCK(fdp);
2134 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2135 		fdfree(fdp, idx);
2136 		FILEDESC_XUNLOCK(fdp);
2137 		fdrop(fp, td);
2138 	} else
2139 		FILEDESC_XUNLOCK(fdp);
2140 }
2141 
2142 /*
2143  * Close any files on exec?
2144  */
2145 void
2146 fdcloseexec(struct thread *td)
2147 {
2148 	struct filedesc *fdp;
2149 	struct filedescent *fde;
2150 	struct file *fp;
2151 	int i;
2152 
2153 	fdp = td->td_proc->p_fd;
2154 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2155 	FILEDESC_XLOCK(fdp);
2156 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2157 		fde = &fdp->fd_ofiles[i];
2158 		fp = fde->fde_file;
2159 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2160 		    (fde->fde_flags & UF_EXCLOSE))) {
2161 			fdfree(fdp, i);
2162 			(void) closefp(fdp, i, fp, td, 0);
2163 			/* closefp() drops the FILEDESC lock. */
2164 			FILEDESC_XLOCK(fdp);
2165 		}
2166 	}
2167 	FILEDESC_XUNLOCK(fdp);
2168 }
2169 
2170 /*
2171  * It is unsafe for set[ug]id processes to be started with file
2172  * descriptors 0..2 closed, as these descriptors are given implicit
2173  * significance in the Standard C library.  fdcheckstd() will create a
2174  * descriptor referencing /dev/null for each of stdin, stdout, and
2175  * stderr that is not already open.
2176  */
2177 int
2178 fdcheckstd(struct thread *td)
2179 {
2180 	struct filedesc *fdp;
2181 	register_t retval, save;
2182 	int i, error, devnull;
2183 
2184 	fdp = td->td_proc->p_fd;
2185 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2186 	devnull = -1;
2187 	error = 0;
2188 	for (i = 0; i < 3; i++) {
2189 		if (fdp->fd_ofiles[i].fde_file != NULL)
2190 			continue;
2191 		if (devnull < 0) {
2192 			save = td->td_retval[0];
2193 			error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2194 			    O_RDWR, 0);
2195 			devnull = td->td_retval[0];
2196 			td->td_retval[0] = save;
2197 			if (error)
2198 				break;
2199 			KASSERT(devnull == i, ("oof, we didn't get our fd"));
2200 		} else {
2201 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2202 			if (error != 0)
2203 				break;
2204 		}
2205 	}
2206 	return (error);
2207 }
2208 
2209 /*
2210  * Internal form of close.  Decrement reference count on file structure.
2211  * Note: td may be NULL when closing a file that was being passed in a
2212  * message.
2213  *
2214  * XXXRW: Giant is not required for the caller, but often will be held; this
2215  * makes it moderately likely the Giant will be recursed in the VFS case.
2216  */
2217 int
2218 closef(struct file *fp, struct thread *td)
2219 {
2220 	struct vnode *vp;
2221 	struct flock lf;
2222 	struct filedesc_to_leader *fdtol;
2223 	struct filedesc *fdp;
2224 
2225 	/*
2226 	 * POSIX record locking dictates that any close releases ALL
2227 	 * locks owned by this process.  This is handled by setting
2228 	 * a flag in the unlock to free ONLY locks obeying POSIX
2229 	 * semantics, and not to free BSD-style file locks.
2230 	 * If the descriptor was in a message, POSIX-style locks
2231 	 * aren't passed with the descriptor, and the thread pointer
2232 	 * will be NULL.  Callers should be careful only to pass a
2233 	 * NULL thread pointer when there really is no owning
2234 	 * context that might have locks, or the locks will be
2235 	 * leaked.
2236 	 */
2237 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2238 		vp = fp->f_vnode;
2239 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2240 			lf.l_whence = SEEK_SET;
2241 			lf.l_start = 0;
2242 			lf.l_len = 0;
2243 			lf.l_type = F_UNLCK;
2244 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2245 			    F_UNLCK, &lf, F_POSIX);
2246 		}
2247 		fdtol = td->td_proc->p_fdtol;
2248 		if (fdtol != NULL) {
2249 			/*
2250 			 * Handle special case where file descriptor table is
2251 			 * shared between multiple process leaders.
2252 			 */
2253 			fdp = td->td_proc->p_fd;
2254 			FILEDESC_XLOCK(fdp);
2255 			for (fdtol = fdtol->fdl_next;
2256 			     fdtol != td->td_proc->p_fdtol;
2257 			     fdtol = fdtol->fdl_next) {
2258 				if ((fdtol->fdl_leader->p_flag &
2259 				     P_ADVLOCK) == 0)
2260 					continue;
2261 				fdtol->fdl_holdcount++;
2262 				FILEDESC_XUNLOCK(fdp);
2263 				lf.l_whence = SEEK_SET;
2264 				lf.l_start = 0;
2265 				lf.l_len = 0;
2266 				lf.l_type = F_UNLCK;
2267 				vp = fp->f_vnode;
2268 				(void) VOP_ADVLOCK(vp,
2269 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2270 				    F_POSIX);
2271 				FILEDESC_XLOCK(fdp);
2272 				fdtol->fdl_holdcount--;
2273 				if (fdtol->fdl_holdcount == 0 &&
2274 				    fdtol->fdl_wakeup != 0) {
2275 					fdtol->fdl_wakeup = 0;
2276 					wakeup(fdtol);
2277 				}
2278 			}
2279 			FILEDESC_XUNLOCK(fdp);
2280 		}
2281 	}
2282 	return (fdrop(fp, td));
2283 }
2284 
2285 /*
2286  * Initialize the file pointer with the specified properties.
2287  *
2288  * The ops are set with release semantics to be certain that the flags, type,
2289  * and data are visible when ops is.  This is to prevent ops methods from being
2290  * called with bad data.
2291  */
2292 void
2293 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2294 {
2295 	fp->f_data = data;
2296 	fp->f_flag = flag;
2297 	fp->f_type = type;
2298 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2299 }
2300 
2301 int
2302 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2303     int needfcntl, struct file **fpp, cap_rights_t *haverightsp)
2304 {
2305 #ifdef CAPABILITIES
2306 	struct filedescent fde;
2307 #endif
2308 	struct file *fp;
2309 	u_int count;
2310 #ifdef CAPABILITIES
2311 	seq_t seq;
2312 	cap_rights_t haverights;
2313 	int error;
2314 #endif
2315 
2316 	/*
2317 	 * Avoid reads reordering and then a first access to the
2318 	 * fdp->fd_ofiles table which could result in OOB operation.
2319 	 */
2320 	if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles))
2321 		return (EBADF);
2322 	/*
2323 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2324 	 * never raising a refcount above 0.  To accomplish this we have
2325 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2326 	 * must be re-verified once we acquire a reference to be certain
2327 	 * that the identity is still correct and we did not lose a race
2328 	 * due to preemption.
2329 	 */
2330 	for (;;) {
2331 #ifdef CAPABILITIES
2332 		seq = seq_read(fd_seq(fdp, fd));
2333 		fde = fdp->fd_ofiles[fd];
2334 		if (!seq_consistent(fd_seq(fdp, fd), seq)) {
2335 			cpu_spinwait();
2336 			continue;
2337 		}
2338 		fp = fde.fde_file;
2339 #else
2340 		fp = fdp->fd_ofiles[fd].fde_file;
2341 #endif
2342 		if (fp == NULL)
2343 			return (EBADF);
2344 #ifdef CAPABILITIES
2345 		haverights = *cap_rights_fde(&fde);
2346 		if (needrightsp != NULL) {
2347 			error = cap_check(&haverights, needrightsp);
2348 			if (error != 0)
2349 				return (error);
2350 			if (cap_rights_is_set(needrightsp, CAP_FCNTL)) {
2351 				error = cap_fcntl_check_fde(&fde, needfcntl);
2352 				if (error != 0)
2353 					return (error);
2354 			}
2355 		}
2356 #endif
2357 		count = fp->f_count;
2358 		if (count == 0)
2359 			continue;
2360 		/*
2361 		 * Use an acquire barrier to prevent caching of fd_ofiles
2362 		 * so it is refreshed for verification.
2363 		 */
2364 		if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2365 			continue;
2366 #ifdef	CAPABILITIES
2367 		if (seq_consistent_nomb(fd_seq(fdp, fd), seq))
2368 #else
2369 		if (fp == fdp->fd_ofiles[fd].fde_file)
2370 #endif
2371 			break;
2372 		fdrop(fp, curthread);
2373 	}
2374 	*fpp = fp;
2375 	if (haverightsp != NULL) {
2376 #ifdef CAPABILITIES
2377 		*haverightsp = haverights;
2378 #else
2379 		CAP_ALL(haverightsp);
2380 #endif
2381 	}
2382 	return (0);
2383 }
2384 
2385 /*
2386  * Extract the file pointer associated with the specified descriptor for the
2387  * current user process.
2388  *
2389  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2390  * returned.
2391  *
2392  * File's rights will be checked against the capability rights mask.
2393  *
2394  * If an error occured the non-zero error is returned and *fpp is set to
2395  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2396  * responsible for fdrop().
2397  */
2398 static __inline int
2399 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2400     cap_rights_t *needrightsp, u_char *maxprotp)
2401 {
2402 	struct filedesc *fdp;
2403 	struct file *fp;
2404 	cap_rights_t haverights, needrights;
2405 	int error;
2406 
2407 	*fpp = NULL;
2408 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2409 		return (EBADF);
2410 	if (needrightsp != NULL)
2411 		needrights = *needrightsp;
2412 	else
2413 		cap_rights_init(&needrights);
2414 	if (maxprotp != NULL)
2415 		cap_rights_set(&needrights, CAP_MMAP);
2416 	error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights);
2417 	if (error != 0)
2418 		return (error);
2419 	if (fp->f_ops == &badfileops) {
2420 		fdrop(fp, td);
2421 		return (EBADF);
2422 	}
2423 
2424 #ifdef CAPABILITIES
2425 	/*
2426 	 * If requested, convert capability rights to access flags.
2427 	 */
2428 	if (maxprotp != NULL)
2429 		*maxprotp = cap_rights_to_vmprot(&haverights);
2430 #else /* !CAPABILITIES */
2431 	if (maxprotp != NULL)
2432 		*maxprotp = VM_PROT_ALL;
2433 #endif /* CAPABILITIES */
2434 
2435 	/*
2436 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2437 	 */
2438 	error = 0;
2439 	switch (flags) {
2440 	case FREAD:
2441 	case FWRITE:
2442 		if ((fp->f_flag & flags) == 0)
2443 			error = EBADF;
2444 		break;
2445 	case FEXEC:
2446 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2447 		    ((fp->f_flag & FWRITE) != 0))
2448 			error = EBADF;
2449 		break;
2450 	case 0:
2451 		break;
2452 	default:
2453 		KASSERT(0, ("wrong flags"));
2454 	}
2455 
2456 	if (error != 0) {
2457 		fdrop(fp, td);
2458 		return (error);
2459 	}
2460 
2461 	*fpp = fp;
2462 	return (0);
2463 }
2464 
2465 int
2466 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2467 {
2468 
2469 	return(_fget(td, fd, fpp, 0, rightsp, NULL));
2470 }
2471 
2472 int
2473 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2474     struct file **fpp)
2475 {
2476 
2477 	return (_fget(td, fd, fpp, 0, rightsp, maxprotp));
2478 }
2479 
2480 int
2481 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2482 {
2483 
2484 	return(_fget(td, fd, fpp, FREAD, rightsp, NULL));
2485 }
2486 
2487 int
2488 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2489 {
2490 
2491 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2492 }
2493 
2494 /*
2495  * Like fget() but loads the underlying vnode, or returns an error if the
2496  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2497  * never have VM objects.  The returned vnode will be vref()'d.
2498  *
2499  * XXX: what about the unused flags ?
2500  */
2501 static __inline int
2502 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2503     struct vnode **vpp)
2504 {
2505 	struct file *fp;
2506 	int error;
2507 
2508 	*vpp = NULL;
2509 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2510 	if (error != 0)
2511 		return (error);
2512 	if (fp->f_vnode == NULL) {
2513 		error = EINVAL;
2514 	} else {
2515 		*vpp = fp->f_vnode;
2516 		vref(*vpp);
2517 	}
2518 	fdrop(fp, td);
2519 
2520 	return (error);
2521 }
2522 
2523 int
2524 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2525 {
2526 
2527 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2528 }
2529 
2530 int
2531 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2532     struct filecaps *havecaps, struct vnode **vpp)
2533 {
2534 	struct filedesc *fdp;
2535 	struct file *fp;
2536 #ifdef CAPABILITIES
2537 	int error;
2538 #endif
2539 
2540 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2541 		return (EBADF);
2542 
2543 	fp = fget_locked(fdp, fd);
2544 	if (fp == NULL || fp->f_ops == &badfileops)
2545 		return (EBADF);
2546 
2547 #ifdef CAPABILITIES
2548 	if (needrightsp != NULL) {
2549 		error = cap_check(cap_rights(fdp, fd), needrightsp);
2550 		if (error != 0)
2551 			return (error);
2552 	}
2553 #endif
2554 
2555 	if (fp->f_vnode == NULL)
2556 		return (EINVAL);
2557 
2558 	*vpp = fp->f_vnode;
2559 	vref(*vpp);
2560 	filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
2561 
2562 	return (0);
2563 }
2564 
2565 int
2566 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2567 {
2568 
2569 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2570 }
2571 
2572 int
2573 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2574 {
2575 
2576 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2577 }
2578 
2579 #ifdef notyet
2580 int
2581 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2582     struct vnode **vpp)
2583 {
2584 
2585 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2586 }
2587 #endif
2588 
2589 /*
2590  * Like fget() but loads the underlying socket, or returns an error if the
2591  * descriptor does not represent a socket.
2592  *
2593  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
2594  * in the future.
2595  *
2596  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2597  * on their file descriptor reference to prevent the socket from being free'd
2598  * during use.
2599  */
2600 int
2601 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp,
2602     u_int *fflagp)
2603 {
2604 	struct file *fp;
2605 	int error;
2606 
2607 	*spp = NULL;
2608 	if (fflagp != NULL)
2609 		*fflagp = 0;
2610 	if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0)
2611 		return (error);
2612 	if (fp->f_type != DTYPE_SOCKET) {
2613 		error = ENOTSOCK;
2614 	} else {
2615 		*spp = fp->f_data;
2616 		if (fflagp)
2617 			*fflagp = fp->f_flag;
2618 		SOCK_LOCK(*spp);
2619 		soref(*spp);
2620 		SOCK_UNLOCK(*spp);
2621 	}
2622 	fdrop(fp, td);
2623 
2624 	return (error);
2625 }
2626 
2627 /*
2628  * Drop the reference count on the socket and XXX release the SX lock in the
2629  * future.  The last reference closes the socket.
2630  *
2631  * Note: fputsock() is deprecated, see comment for fgetsock().
2632  */
2633 void
2634 fputsock(struct socket *so)
2635 {
2636 
2637 	ACCEPT_LOCK();
2638 	SOCK_LOCK(so);
2639 	CURVNET_SET(so->so_vnet);
2640 	sorele(so);
2641 	CURVNET_RESTORE();
2642 }
2643 
2644 /*
2645  * Handle the last reference to a file being closed.
2646  */
2647 int
2648 _fdrop(struct file *fp, struct thread *td)
2649 {
2650 	int error;
2651 
2652 	error = 0;
2653 	if (fp->f_count != 0)
2654 		panic("fdrop: count %d", fp->f_count);
2655 	if (fp->f_ops != &badfileops)
2656 		error = fo_close(fp, td);
2657 	atomic_subtract_int(&openfiles, 1);
2658 	crfree(fp->f_cred);
2659 	free(fp->f_advice, M_FADVISE);
2660 	uma_zfree(file_zone, fp);
2661 
2662 	return (error);
2663 }
2664 
2665 /*
2666  * Apply an advisory lock on a file descriptor.
2667  *
2668  * Just attempt to get a record lock of the requested type on the entire file
2669  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2670  */
2671 #ifndef _SYS_SYSPROTO_H_
2672 struct flock_args {
2673 	int	fd;
2674 	int	how;
2675 };
2676 #endif
2677 /* ARGSUSED */
2678 int
2679 sys_flock(struct thread *td, struct flock_args *uap)
2680 {
2681 	struct file *fp;
2682 	struct vnode *vp;
2683 	struct flock lf;
2684 	cap_rights_t rights;
2685 	int error;
2686 
2687 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2688 	if (error != 0)
2689 		return (error);
2690 	if (fp->f_type != DTYPE_VNODE) {
2691 		fdrop(fp, td);
2692 		return (EOPNOTSUPP);
2693 	}
2694 
2695 	vp = fp->f_vnode;
2696 	lf.l_whence = SEEK_SET;
2697 	lf.l_start = 0;
2698 	lf.l_len = 0;
2699 	if (uap->how & LOCK_UN) {
2700 		lf.l_type = F_UNLCK;
2701 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2702 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2703 		goto done2;
2704 	}
2705 	if (uap->how & LOCK_EX)
2706 		lf.l_type = F_WRLCK;
2707 	else if (uap->how & LOCK_SH)
2708 		lf.l_type = F_RDLCK;
2709 	else {
2710 		error = EBADF;
2711 		goto done2;
2712 	}
2713 	atomic_set_int(&fp->f_flag, FHASLOCK);
2714 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2715 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2716 done2:
2717 	fdrop(fp, td);
2718 	return (error);
2719 }
2720 /*
2721  * Duplicate the specified descriptor to a free descriptor.
2722  */
2723 int
2724 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2725     int openerror, int *indxp)
2726 {
2727 	struct filedescent *newfde, *oldfde;
2728 	struct file *fp;
2729 	int error, indx;
2730 
2731 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2732 	    ("unexpected error %d in %s", openerror, __func__));
2733 
2734 	/*
2735 	 * If the to-be-dup'd fd number is greater than the allowed number
2736 	 * of file descriptors, or the fd to be dup'd has already been
2737 	 * closed, then reject.
2738 	 */
2739 	FILEDESC_XLOCK(fdp);
2740 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2741 		FILEDESC_XUNLOCK(fdp);
2742 		return (EBADF);
2743 	}
2744 
2745 	error = fdalloc(td, 0, &indx);
2746 	if (error != 0) {
2747 		FILEDESC_XUNLOCK(fdp);
2748 		return (error);
2749 	}
2750 
2751 	/*
2752 	 * There are two cases of interest here.
2753 	 *
2754 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2755 	 *
2756 	 * For ENXIO steal away the file structure from (dfd) and store it in
2757 	 * (indx).  (dfd) is effectively closed by this operation.
2758 	 */
2759 	switch (openerror) {
2760 	case ENODEV:
2761 		/*
2762 		 * Check that the mode the file is being opened for is a
2763 		 * subset of the mode of the existing descriptor.
2764 		 */
2765 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2766 			fdunused(fdp, indx);
2767 			FILEDESC_XUNLOCK(fdp);
2768 			return (EACCES);
2769 		}
2770 		fhold(fp);
2771 		newfde = &fdp->fd_ofiles[indx];
2772 		oldfde = &fdp->fd_ofiles[dfd];
2773 #ifdef CAPABILITIES
2774 		seq_write_begin(&newfde->fde_seq);
2775 #endif
2776 		memcpy(newfde, oldfde, fde_change_size);
2777 		filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
2778 #ifdef CAPABILITIES
2779 		seq_write_end(&newfde->fde_seq);
2780 #endif
2781 		break;
2782 	case ENXIO:
2783 		/*
2784 		 * Steal away the file pointer from dfd and stuff it into indx.
2785 		 */
2786 		newfde = &fdp->fd_ofiles[indx];
2787 		oldfde = &fdp->fd_ofiles[dfd];
2788 #ifdef CAPABILITIES
2789 		seq_write_begin(&newfde->fde_seq);
2790 #endif
2791 		memcpy(newfde, oldfde, fde_change_size);
2792 		bzero(oldfde, fde_change_size);
2793 		fdunused(fdp, dfd);
2794 #ifdef CAPABILITIES
2795 		seq_write_end(&newfde->fde_seq);
2796 #endif
2797 		break;
2798 	}
2799 	FILEDESC_XUNLOCK(fdp);
2800 	*indxp = indx;
2801 	return (0);
2802 }
2803 
2804 /*
2805  * Scan all active processes and prisons to see if any of them have a current
2806  * or root directory of `olddp'. If so, replace them with the new mount point.
2807  */
2808 void
2809 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2810 {
2811 	struct filedesc *fdp;
2812 	struct prison *pr;
2813 	struct proc *p;
2814 	int nrele;
2815 
2816 	if (vrefcnt(olddp) == 1)
2817 		return;
2818 	nrele = 0;
2819 	sx_slock(&allproc_lock);
2820 	FOREACH_PROC_IN_SYSTEM(p) {
2821 		fdp = fdhold(p);
2822 		if (fdp == NULL)
2823 			continue;
2824 		FILEDESC_XLOCK(fdp);
2825 		if (fdp->fd_cdir == olddp) {
2826 			vref(newdp);
2827 			fdp->fd_cdir = newdp;
2828 			nrele++;
2829 		}
2830 		if (fdp->fd_rdir == olddp) {
2831 			vref(newdp);
2832 			fdp->fd_rdir = newdp;
2833 			nrele++;
2834 		}
2835 		if (fdp->fd_jdir == olddp) {
2836 			vref(newdp);
2837 			fdp->fd_jdir = newdp;
2838 			nrele++;
2839 		}
2840 		FILEDESC_XUNLOCK(fdp);
2841 		fddrop(fdp);
2842 	}
2843 	sx_sunlock(&allproc_lock);
2844 	if (rootvnode == olddp) {
2845 		vref(newdp);
2846 		rootvnode = newdp;
2847 		nrele++;
2848 	}
2849 	mtx_lock(&prison0.pr_mtx);
2850 	if (prison0.pr_root == olddp) {
2851 		vref(newdp);
2852 		prison0.pr_root = newdp;
2853 		nrele++;
2854 	}
2855 	mtx_unlock(&prison0.pr_mtx);
2856 	sx_slock(&allprison_lock);
2857 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2858 		mtx_lock(&pr->pr_mtx);
2859 		if (pr->pr_root == olddp) {
2860 			vref(newdp);
2861 			pr->pr_root = newdp;
2862 			nrele++;
2863 		}
2864 		mtx_unlock(&pr->pr_mtx);
2865 	}
2866 	sx_sunlock(&allprison_lock);
2867 	while (nrele--)
2868 		vrele(olddp);
2869 }
2870 
2871 struct filedesc_to_leader *
2872 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2873 {
2874 	struct filedesc_to_leader *fdtol;
2875 
2876 	fdtol = malloc(sizeof(struct filedesc_to_leader),
2877 	       M_FILEDESC_TO_LEADER,
2878 	       M_WAITOK);
2879 	fdtol->fdl_refcount = 1;
2880 	fdtol->fdl_holdcount = 0;
2881 	fdtol->fdl_wakeup = 0;
2882 	fdtol->fdl_leader = leader;
2883 	if (old != NULL) {
2884 		FILEDESC_XLOCK(fdp);
2885 		fdtol->fdl_next = old->fdl_next;
2886 		fdtol->fdl_prev = old;
2887 		old->fdl_next = fdtol;
2888 		fdtol->fdl_next->fdl_prev = fdtol;
2889 		FILEDESC_XUNLOCK(fdp);
2890 	} else {
2891 		fdtol->fdl_next = fdtol;
2892 		fdtol->fdl_prev = fdtol;
2893 	}
2894 	return (fdtol);
2895 }
2896 
2897 /*
2898  * Get file structures globally.
2899  */
2900 static int
2901 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2902 {
2903 	struct xfile xf;
2904 	struct filedesc *fdp;
2905 	struct file *fp;
2906 	struct proc *p;
2907 	int error, n;
2908 
2909 	error = sysctl_wire_old_buffer(req, 0);
2910 	if (error != 0)
2911 		return (error);
2912 	if (req->oldptr == NULL) {
2913 		n = 0;
2914 		sx_slock(&allproc_lock);
2915 		FOREACH_PROC_IN_SYSTEM(p) {
2916 			if (p->p_state == PRS_NEW)
2917 				continue;
2918 			fdp = fdhold(p);
2919 			if (fdp == NULL)
2920 				continue;
2921 			/* overestimates sparse tables. */
2922 			if (fdp->fd_lastfile > 0)
2923 				n += fdp->fd_lastfile;
2924 			fddrop(fdp);
2925 		}
2926 		sx_sunlock(&allproc_lock);
2927 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2928 	}
2929 	error = 0;
2930 	bzero(&xf, sizeof(xf));
2931 	xf.xf_size = sizeof(xf);
2932 	sx_slock(&allproc_lock);
2933 	FOREACH_PROC_IN_SYSTEM(p) {
2934 		PROC_LOCK(p);
2935 		if (p->p_state == PRS_NEW) {
2936 			PROC_UNLOCK(p);
2937 			continue;
2938 		}
2939 		if (p_cansee(req->td, p) != 0) {
2940 			PROC_UNLOCK(p);
2941 			continue;
2942 		}
2943 		xf.xf_pid = p->p_pid;
2944 		xf.xf_uid = p->p_ucred->cr_uid;
2945 		PROC_UNLOCK(p);
2946 		fdp = fdhold(p);
2947 		if (fdp == NULL)
2948 			continue;
2949 		FILEDESC_SLOCK(fdp);
2950 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
2951 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
2952 				continue;
2953 			xf.xf_fd = n;
2954 			xf.xf_file = fp;
2955 			xf.xf_data = fp->f_data;
2956 			xf.xf_vnode = fp->f_vnode;
2957 			xf.xf_type = fp->f_type;
2958 			xf.xf_count = fp->f_count;
2959 			xf.xf_msgcount = 0;
2960 			xf.xf_offset = foffset_get(fp);
2961 			xf.xf_flag = fp->f_flag;
2962 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2963 			if (error)
2964 				break;
2965 		}
2966 		FILEDESC_SUNLOCK(fdp);
2967 		fddrop(fdp);
2968 		if (error)
2969 			break;
2970 	}
2971 	sx_sunlock(&allproc_lock);
2972 	return (error);
2973 }
2974 
2975 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
2976     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2977 
2978 #ifdef KINFO_FILE_SIZE
2979 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2980 #endif
2981 
2982 static int
2983 xlate_fflags(int fflags)
2984 {
2985 	static const struct {
2986 		int	fflag;
2987 		int	kf_fflag;
2988 	} fflags_table[] = {
2989 		{ FAPPEND, KF_FLAG_APPEND },
2990 		{ FASYNC, KF_FLAG_ASYNC },
2991 		{ FFSYNC, KF_FLAG_FSYNC },
2992 		{ FHASLOCK, KF_FLAG_HASLOCK },
2993 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
2994 		{ FREAD, KF_FLAG_READ },
2995 		{ FWRITE, KF_FLAG_WRITE },
2996 		{ O_CREAT, KF_FLAG_CREAT },
2997 		{ O_DIRECT, KF_FLAG_DIRECT },
2998 		{ O_EXCL, KF_FLAG_EXCL },
2999 		{ O_EXEC, KF_FLAG_EXEC },
3000 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3001 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3002 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3003 		{ O_TRUNC, KF_FLAG_TRUNC }
3004 	};
3005 	unsigned int i;
3006 	int kflags;
3007 
3008 	kflags = 0;
3009 	for (i = 0; i < nitems(fflags_table); i++)
3010 		if (fflags & fflags_table[i].fflag)
3011 			kflags |=  fflags_table[i].kf_fflag;
3012 	return (kflags);
3013 }
3014 
3015 /* Trim unused data from kf_path by truncating the structure size. */
3016 static void
3017 pack_kinfo(struct kinfo_file *kif)
3018 {
3019 
3020 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3021 	    strlen(kif->kf_path) + 1;
3022 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3023 }
3024 
3025 static void
3026 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3027     struct kinfo_file *kif, struct filedesc *fdp)
3028 {
3029 	int error;
3030 
3031 	bzero(kif, sizeof(*kif));
3032 
3033 	/* Set a default type to allow for empty fill_kinfo() methods. */
3034 	kif->kf_type = KF_TYPE_UNKNOWN;
3035 	kif->kf_flags = xlate_fflags(fp->f_flag);
3036 	if (rightsp != NULL)
3037 		kif->kf_cap_rights = *rightsp;
3038 	else
3039 		cap_rights_init(&kif->kf_cap_rights);
3040 	kif->kf_fd = fd;
3041 	kif->kf_ref_count = fp->f_count;
3042 	kif->kf_offset = foffset_get(fp);
3043 
3044 	/*
3045 	 * This may drop the filedesc lock, so the 'fp' cannot be
3046 	 * accessed after this call.
3047 	 */
3048 	error = fo_fill_kinfo(fp, kif, fdp);
3049 	if (error == 0)
3050 		kif->kf_status |= KF_ATTR_VALID;
3051 	pack_kinfo(kif);
3052 }
3053 
3054 static void
3055 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3056     struct kinfo_file *kif)
3057 {
3058 	int error;
3059 
3060 	bzero(kif, sizeof(*kif));
3061 
3062 	kif->kf_type = KF_TYPE_VNODE;
3063 	error = vn_fill_kinfo_vnode(vp, kif);
3064 	if (error == 0)
3065 		kif->kf_status |= KF_ATTR_VALID;
3066 	kif->kf_flags = xlate_fflags(fflags);
3067 	kif->kf_fd = fd;
3068 	kif->kf_ref_count = -1;
3069 	kif->kf_offset = -1;
3070 	pack_kinfo(kif);
3071 	vrele(vp);
3072 }
3073 
3074 struct export_fd_buf {
3075 	struct filedesc		*fdp;
3076 	struct sbuf 		*sb;
3077 	ssize_t			remainder;
3078 	struct kinfo_file	kif;
3079 };
3080 
3081 static int
3082 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3083 {
3084 	struct kinfo_file *kif;
3085 
3086 	kif = &efbuf->kif;
3087 	if (efbuf->remainder != -1) {
3088 		if (efbuf->remainder < kif->kf_structsize) {
3089 			/* Terminate export. */
3090 			efbuf->remainder = 0;
3091 			return (0);
3092 		}
3093 		efbuf->remainder -= kif->kf_structsize;
3094 	}
3095 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3096 }
3097 
3098 static int
3099 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3100     struct export_fd_buf *efbuf)
3101 {
3102 	int error;
3103 
3104 	if (efbuf->remainder == 0)
3105 		return (0);
3106 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp);
3107 	FILEDESC_SUNLOCK(efbuf->fdp);
3108 	error = export_kinfo_to_sb(efbuf);
3109 	FILEDESC_SLOCK(efbuf->fdp);
3110 	return (error);
3111 }
3112 
3113 static int
3114 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3115     struct export_fd_buf *efbuf)
3116 {
3117 	int error;
3118 
3119 	if (efbuf->remainder == 0)
3120 		return (0);
3121 	if (efbuf->fdp != NULL)
3122 		FILEDESC_SUNLOCK(efbuf->fdp);
3123 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif);
3124 	error = export_kinfo_to_sb(efbuf);
3125 	if (efbuf->fdp != NULL)
3126 		FILEDESC_SLOCK(efbuf->fdp);
3127 	return (error);
3128 }
3129 
3130 /*
3131  * Store a process file descriptor information to sbuf.
3132  *
3133  * Takes a locked proc as argument, and returns with the proc unlocked.
3134  */
3135 int
3136 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3137 {
3138 	struct file *fp;
3139 	struct filedesc *fdp;
3140 	struct export_fd_buf *efbuf;
3141 	struct vnode *cttyvp, *textvp, *tracevp;
3142 	int error, i;
3143 	cap_rights_t rights;
3144 
3145 	PROC_LOCK_ASSERT(p, MA_OWNED);
3146 
3147 	/* ktrace vnode */
3148 	tracevp = p->p_tracevp;
3149 	if (tracevp != NULL)
3150 		vref(tracevp);
3151 	/* text vnode */
3152 	textvp = p->p_textvp;
3153 	if (textvp != NULL)
3154 		vref(textvp);
3155 	/* Controlling tty. */
3156 	cttyvp = NULL;
3157 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3158 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3159 		if (cttyvp != NULL)
3160 			vref(cttyvp);
3161 	}
3162 	fdp = fdhold(p);
3163 	PROC_UNLOCK(p);
3164 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3165 	efbuf->fdp = NULL;
3166 	efbuf->sb = sb;
3167 	efbuf->remainder = maxlen;
3168 	if (tracevp != NULL)
3169 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3170 		    efbuf);
3171 	if (textvp != NULL)
3172 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3173 	if (cttyvp != NULL)
3174 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3175 		    efbuf);
3176 	error = 0;
3177 	if (fdp == NULL)
3178 		goto fail;
3179 	efbuf->fdp = fdp;
3180 	FILEDESC_SLOCK(fdp);
3181 	/* working directory */
3182 	if (fdp->fd_cdir != NULL) {
3183 		vref(fdp->fd_cdir);
3184 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3185 	}
3186 	/* root directory */
3187 	if (fdp->fd_rdir != NULL) {
3188 		vref(fdp->fd_rdir);
3189 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3190 	}
3191 	/* jail directory */
3192 	if (fdp->fd_jdir != NULL) {
3193 		vref(fdp->fd_jdir);
3194 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3195 	}
3196 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3197 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3198 			continue;
3199 #ifdef CAPABILITIES
3200 		rights = *cap_rights(fdp, i);
3201 #else /* !CAPABILITIES */
3202 		cap_rights_init(&rights);
3203 #endif
3204 		/*
3205 		 * Create sysctl entry.  It is OK to drop the filedesc
3206 		 * lock inside of export_file_to_sb() as we will
3207 		 * re-validate and re-evaluate its properties when the
3208 		 * loop continues.
3209 		 */
3210 		error = export_file_to_sb(fp, i, &rights, efbuf);
3211 		if (error != 0 || efbuf->remainder == 0)
3212 			break;
3213 	}
3214 	FILEDESC_SUNLOCK(fdp);
3215 	fddrop(fdp);
3216 fail:
3217 	free(efbuf, M_TEMP);
3218 	return (error);
3219 }
3220 
3221 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3222 
3223 /*
3224  * Get per-process file descriptors for use by procstat(1), et al.
3225  */
3226 static int
3227 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3228 {
3229 	struct sbuf sb;
3230 	struct proc *p;
3231 	ssize_t maxlen;
3232 	int error, error2, *name;
3233 
3234 	name = (int *)arg1;
3235 
3236 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3237 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3238 	if (error != 0) {
3239 		sbuf_delete(&sb);
3240 		return (error);
3241 	}
3242 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3243 	error = kern_proc_filedesc_out(p, &sb, maxlen);
3244 	error2 = sbuf_finish(&sb);
3245 	sbuf_delete(&sb);
3246 	return (error != 0 ? error : error2);
3247 }
3248 
3249 #ifdef KINFO_OFILE_SIZE
3250 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3251 #endif
3252 
3253 #ifdef COMPAT_FREEBSD7
3254 static void
3255 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3256 {
3257 
3258 	okif->kf_structsize = sizeof(*okif);
3259 	okif->kf_type = kif->kf_type;
3260 	okif->kf_fd = kif->kf_fd;
3261 	okif->kf_ref_count = kif->kf_ref_count;
3262 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3263 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3264 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3265 	okif->kf_offset = kif->kf_offset;
3266 	okif->kf_vnode_type = kif->kf_vnode_type;
3267 	okif->kf_sock_domain = kif->kf_sock_domain;
3268 	okif->kf_sock_type = kif->kf_sock_type;
3269 	okif->kf_sock_protocol = kif->kf_sock_protocol;
3270 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3271 	okif->kf_sa_local = kif->kf_sa_local;
3272 	okif->kf_sa_peer = kif->kf_sa_peer;
3273 }
3274 
3275 static int
3276 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3277     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3278 {
3279 	int error;
3280 
3281 	vref(vp);
3282 	FILEDESC_SUNLOCK(fdp);
3283 	export_vnode_to_kinfo(vp, type, 0, kif);
3284 	kinfo_to_okinfo(kif, okif);
3285 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3286 	FILEDESC_SLOCK(fdp);
3287 	return (error);
3288 }
3289 
3290 /*
3291  * Get per-process file descriptors for use by procstat(1), et al.
3292  */
3293 static int
3294 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3295 {
3296 	struct kinfo_ofile *okif;
3297 	struct kinfo_file *kif;
3298 	struct filedesc *fdp;
3299 	int error, i, *name;
3300 	struct file *fp;
3301 	struct proc *p;
3302 
3303 	name = (int *)arg1;
3304 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3305 	if (error != 0)
3306 		return (error);
3307 	fdp = fdhold(p);
3308 	PROC_UNLOCK(p);
3309 	if (fdp == NULL)
3310 		return (ENOENT);
3311 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3312 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3313 	FILEDESC_SLOCK(fdp);
3314 	if (fdp->fd_cdir != NULL)
3315 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3316 		    okif, fdp, req);
3317 	if (fdp->fd_rdir != NULL)
3318 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3319 		    okif, fdp, req);
3320 	if (fdp->fd_jdir != NULL)
3321 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3322 		    okif, fdp, req);
3323 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3324 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3325 			continue;
3326 		export_file_to_kinfo(fp, i, NULL, kif, fdp);
3327 		FILEDESC_SUNLOCK(fdp);
3328 		kinfo_to_okinfo(kif, okif);
3329 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3330 		FILEDESC_SLOCK(fdp);
3331 		if (error)
3332 			break;
3333 	}
3334 	FILEDESC_SUNLOCK(fdp);
3335 	fddrop(fdp);
3336 	free(kif, M_TEMP);
3337 	free(okif, M_TEMP);
3338 	return (0);
3339 }
3340 
3341 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3342     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3343     "Process ofiledesc entries");
3344 #endif	/* COMPAT_FREEBSD7 */
3345 
3346 int
3347 vntype_to_kinfo(int vtype)
3348 {
3349 	struct {
3350 		int	vtype;
3351 		int	kf_vtype;
3352 	} vtypes_table[] = {
3353 		{ VBAD, KF_VTYPE_VBAD },
3354 		{ VBLK, KF_VTYPE_VBLK },
3355 		{ VCHR, KF_VTYPE_VCHR },
3356 		{ VDIR, KF_VTYPE_VDIR },
3357 		{ VFIFO, KF_VTYPE_VFIFO },
3358 		{ VLNK, KF_VTYPE_VLNK },
3359 		{ VNON, KF_VTYPE_VNON },
3360 		{ VREG, KF_VTYPE_VREG },
3361 		{ VSOCK, KF_VTYPE_VSOCK }
3362 	};
3363 	unsigned int i;
3364 
3365 	/*
3366 	 * Perform vtype translation.
3367 	 */
3368 	for (i = 0; i < nitems(vtypes_table); i++)
3369 		if (vtypes_table[i].vtype == vtype)
3370 			return (vtypes_table[i].kf_vtype);
3371 
3372 	return (KF_VTYPE_UNKNOWN);
3373 }
3374 
3375 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3376     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3377     "Process filedesc entries");
3378 
3379 #ifdef DDB
3380 /*
3381  * For the purposes of debugging, generate a human-readable string for the
3382  * file type.
3383  */
3384 static const char *
3385 file_type_to_name(short type)
3386 {
3387 
3388 	switch (type) {
3389 	case 0:
3390 		return ("zero");
3391 	case DTYPE_VNODE:
3392 		return ("vnod");
3393 	case DTYPE_SOCKET:
3394 		return ("sock");
3395 	case DTYPE_PIPE:
3396 		return ("pipe");
3397 	case DTYPE_FIFO:
3398 		return ("fifo");
3399 	case DTYPE_KQUEUE:
3400 		return ("kque");
3401 	case DTYPE_CRYPTO:
3402 		return ("crpt");
3403 	case DTYPE_MQUEUE:
3404 		return ("mque");
3405 	case DTYPE_SHM:
3406 		return ("shm");
3407 	case DTYPE_SEM:
3408 		return ("ksem");
3409 	default:
3410 		return ("unkn");
3411 	}
3412 }
3413 
3414 /*
3415  * For the purposes of debugging, identify a process (if any, perhaps one of
3416  * many) that references the passed file in its file descriptor array. Return
3417  * NULL if none.
3418  */
3419 static struct proc *
3420 file_to_first_proc(struct file *fp)
3421 {
3422 	struct filedesc *fdp;
3423 	struct proc *p;
3424 	int n;
3425 
3426 	FOREACH_PROC_IN_SYSTEM(p) {
3427 		if (p->p_state == PRS_NEW)
3428 			continue;
3429 		fdp = p->p_fd;
3430 		if (fdp == NULL)
3431 			continue;
3432 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3433 			if (fp == fdp->fd_ofiles[n].fde_file)
3434 				return (p);
3435 		}
3436 	}
3437 	return (NULL);
3438 }
3439 
3440 static void
3441 db_print_file(struct file *fp, int header)
3442 {
3443 	struct proc *p;
3444 
3445 	if (header)
3446 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3447 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3448 		    "MCount", "Vnode", "FPID", "FCmd");
3449 	p = file_to_first_proc(fp);
3450 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3451 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3452 	    0, fp->f_count, 0, fp->f_vnode,
3453 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3454 }
3455 
3456 DB_SHOW_COMMAND(file, db_show_file)
3457 {
3458 	struct file *fp;
3459 
3460 	if (!have_addr) {
3461 		db_printf("usage: show file <addr>\n");
3462 		return;
3463 	}
3464 	fp = (struct file *)addr;
3465 	db_print_file(fp, 1);
3466 }
3467 
3468 DB_SHOW_COMMAND(files, db_show_files)
3469 {
3470 	struct filedesc *fdp;
3471 	struct file *fp;
3472 	struct proc *p;
3473 	int header;
3474 	int n;
3475 
3476 	header = 1;
3477 	FOREACH_PROC_IN_SYSTEM(p) {
3478 		if (p->p_state == PRS_NEW)
3479 			continue;
3480 		if ((fdp = p->p_fd) == NULL)
3481 			continue;
3482 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
3483 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3484 				continue;
3485 			db_print_file(fp, header);
3486 			header = 0;
3487 		}
3488 	}
3489 }
3490 #endif
3491 
3492 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3493     &maxfilesperproc, 0, "Maximum files allowed open per process");
3494 
3495 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3496     &maxfiles, 0, "Maximum number of files");
3497 
3498 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3499     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3500 
3501 /* ARGSUSED*/
3502 static void
3503 filelistinit(void *dummy)
3504 {
3505 
3506 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3507 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3508 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3509 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3510 }
3511 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3512 
3513 /*-------------------------------------------------------------------*/
3514 
3515 static int
3516 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3517     int flags, struct thread *td)
3518 {
3519 
3520 	return (EBADF);
3521 }
3522 
3523 static int
3524 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3525     struct thread *td)
3526 {
3527 
3528 	return (EINVAL);
3529 }
3530 
3531 static int
3532 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3533     struct thread *td)
3534 {
3535 
3536 	return (EBADF);
3537 }
3538 
3539 static int
3540 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3541     struct thread *td)
3542 {
3543 
3544 	return (0);
3545 }
3546 
3547 static int
3548 badfo_kqfilter(struct file *fp, struct knote *kn)
3549 {
3550 
3551 	return (EBADF);
3552 }
3553 
3554 static int
3555 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3556     struct thread *td)
3557 {
3558 
3559 	return (EBADF);
3560 }
3561 
3562 static int
3563 badfo_close(struct file *fp, struct thread *td)
3564 {
3565 
3566 	return (EBADF);
3567 }
3568 
3569 static int
3570 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3571     struct thread *td)
3572 {
3573 
3574 	return (EBADF);
3575 }
3576 
3577 static int
3578 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3579     struct thread *td)
3580 {
3581 
3582 	return (EBADF);
3583 }
3584 
3585 static int
3586 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3587     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3588     int kflags, struct sendfile_sync *sfs, struct thread *td)
3589 {
3590 
3591 	return (EBADF);
3592 }
3593 
3594 static int
3595 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
3596 {
3597 
3598 	return (0);
3599 }
3600 
3601 struct fileops badfileops = {
3602 	.fo_read = badfo_readwrite,
3603 	.fo_write = badfo_readwrite,
3604 	.fo_truncate = badfo_truncate,
3605 	.fo_ioctl = badfo_ioctl,
3606 	.fo_poll = badfo_poll,
3607 	.fo_kqfilter = badfo_kqfilter,
3608 	.fo_stat = badfo_stat,
3609 	.fo_close = badfo_close,
3610 	.fo_chmod = badfo_chmod,
3611 	.fo_chown = badfo_chown,
3612 	.fo_sendfile = badfo_sendfile,
3613 	.fo_fill_kinfo = badfo_fill_kinfo,
3614 };
3615 
3616 int
3617 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
3618     int flags, struct thread *td)
3619 {
3620 
3621 	return (EOPNOTSUPP);
3622 }
3623 
3624 int
3625 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3626     struct thread *td)
3627 {
3628 
3629 	return (EINVAL);
3630 }
3631 
3632 int
3633 invfo_ioctl(struct file *fp, u_long com, void *data,
3634     struct ucred *active_cred, struct thread *td)
3635 {
3636 
3637 	return (ENOTTY);
3638 }
3639 
3640 int
3641 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
3642     struct thread *td)
3643 {
3644 
3645 	return (poll_no_poll(events));
3646 }
3647 
3648 int
3649 invfo_kqfilter(struct file *fp, struct knote *kn)
3650 {
3651 
3652 	return (EINVAL);
3653 }
3654 
3655 int
3656 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3657     struct thread *td)
3658 {
3659 
3660 	return (EINVAL);
3661 }
3662 
3663 int
3664 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3665     struct thread *td)
3666 {
3667 
3668 	return (EINVAL);
3669 }
3670 
3671 int
3672 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3673     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3674     int kflags, struct sendfile_sync *sfs, struct thread *td)
3675 {
3676 
3677 	return (EINVAL);
3678 }
3679 
3680 /*-------------------------------------------------------------------*/
3681 
3682 /*
3683  * File Descriptor pseudo-device driver (/dev/fd/).
3684  *
3685  * Opening minor device N dup()s the file (if any) connected to file
3686  * descriptor N belonging to the calling process.  Note that this driver
3687  * consists of only the ``open()'' routine, because all subsequent
3688  * references to this file will be direct to the other driver.
3689  *
3690  * XXX: we could give this one a cloning event handler if necessary.
3691  */
3692 
3693 /* ARGSUSED */
3694 static int
3695 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3696 {
3697 
3698 	/*
3699 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3700 	 * the file descriptor being sought for duplication. The error
3701 	 * return ensures that the vnode for this device will be released
3702 	 * by vn_open. Open will detect this special error and take the
3703 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3704 	 * will simply report the error.
3705 	 */
3706 	td->td_dupfd = dev2unit(dev);
3707 	return (ENODEV);
3708 }
3709 
3710 static struct cdevsw fildesc_cdevsw = {
3711 	.d_version =	D_VERSION,
3712 	.d_open =	fdopen,
3713 	.d_name =	"FD",
3714 };
3715 
3716 static void
3717 fildesc_drvinit(void *unused)
3718 {
3719 	struct cdev *dev;
3720 
3721 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
3722 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
3723 	make_dev_alias(dev, "stdin");
3724 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
3725 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
3726 	make_dev_alias(dev, "stdout");
3727 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
3728 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
3729 	make_dev_alias(dev, "stderr");
3730 }
3731 
3732 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
3733