xref: /freebsd/sys/kern/kern_descrip.c (revision 38d120bc13ac1de5b739b67b87016b9122149374)
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 static uma_zone_t filedesc0_zone;
102 
103 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
104 		    struct thread *td, int holdleaders);
105 static int	do_dup(struct thread *td, int flags, int old, int new);
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 fdescenttbl *ft_table;
154 	SLIST_ENTRY(freetable) ft_next;
155 };
156 
157 /*
158  * Initial allocation: a filedesc structure + the head of SLIST used to
159  * keep track of old ofiles + enough space for NDFILE descriptors.
160  */
161 
162 struct fdescenttbl0 {
163 	int	fdt_nfiles;
164 	struct	filedescent fdt_ofiles[NDFILE];
165 };
166 
167 struct filedesc0 {
168 	struct filedesc fd_fd;
169 	SLIST_HEAD(, freetable) fd_free;
170 	struct	fdescenttbl0 fd_dfiles;
171 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
172 };
173 
174 /*
175  * Descriptor management.
176  */
177 volatile int openfiles;			/* actual number of open files */
178 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
179 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
180 
181 /* A mutex to protect the association between a proc and filedesc. */
182 static struct mtx fdesc_mtx;
183 
184 /*
185  * If low >= size, just return low. Otherwise find the first zero bit in the
186  * given bitmap, starting at low and not exceeding size - 1. Return size if
187  * not found.
188  */
189 static int
190 fd_first_free(struct filedesc *fdp, int low, int size)
191 {
192 	NDSLOTTYPE *map = fdp->fd_map;
193 	NDSLOTTYPE mask;
194 	int off, maxoff;
195 
196 	if (low >= size)
197 		return (low);
198 
199 	off = NDSLOT(low);
200 	if (low % NDENTRIES) {
201 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
202 		if ((mask &= ~map[off]) != 0UL)
203 			return (off * NDENTRIES + ffsl(mask) - 1);
204 		++off;
205 	}
206 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
207 		if (map[off] != ~0UL)
208 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
209 	return (size);
210 }
211 
212 /*
213  * Find the highest non-zero bit in the given bitmap, starting at 0 and
214  * not exceeding size - 1. Return -1 if not found.
215  */
216 static int
217 fd_last_used(struct filedesc *fdp, int size)
218 {
219 	NDSLOTTYPE *map = fdp->fd_map;
220 	NDSLOTTYPE mask;
221 	int off, minoff;
222 
223 	off = NDSLOT(size);
224 	if (size % NDENTRIES) {
225 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
226 		if ((mask &= map[off]) != 0)
227 			return (off * NDENTRIES + flsl(mask) - 1);
228 		--off;
229 	}
230 	for (minoff = NDSLOT(0); off >= minoff; --off)
231 		if (map[off] != 0)
232 			return (off * NDENTRIES + flsl(map[off]) - 1);
233 	return (-1);
234 }
235 
236 #ifdef INVARIANTS
237 static int
238 fdisused(struct filedesc *fdp, int fd)
239 {
240 
241 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
242 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
243 
244 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
245 }
246 #endif
247 
248 /*
249  * Mark a file descriptor as used.
250  */
251 static void
252 fdused_init(struct filedesc *fdp, int fd)
253 {
254 
255 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
256 
257 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
258 }
259 
260 static void
261 fdused(struct filedesc *fdp, int fd)
262 {
263 
264 	FILEDESC_XLOCK_ASSERT(fdp);
265 
266 	fdused_init(fdp, fd);
267 	if (fd > fdp->fd_lastfile)
268 		fdp->fd_lastfile = fd;
269 	if (fd == fdp->fd_freefile)
270 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
271 }
272 
273 /*
274  * Mark a file descriptor as unused.
275  */
276 static void
277 fdunused(struct filedesc *fdp, int fd)
278 {
279 
280 	FILEDESC_XLOCK_ASSERT(fdp);
281 
282 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
283 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
284 	    ("fd=%d is still in use", fd));
285 
286 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
287 	if (fd < fdp->fd_freefile)
288 		fdp->fd_freefile = fd;
289 	if (fd == fdp->fd_lastfile)
290 		fdp->fd_lastfile = fd_last_used(fdp, fd);
291 }
292 
293 /*
294  * Free a file descriptor.
295  *
296  * Avoid some work if fdp is about to be destroyed.
297  */
298 static inline void
299 fdefree_last(struct filedescent *fde)
300 {
301 
302 	filecaps_free(&fde->fde_caps);
303 }
304 
305 static inline void
306 fdfree(struct filedesc *fdp, int fd)
307 {
308 	struct filedescent *fde;
309 
310 	fde = &fdp->fd_ofiles[fd];
311 #ifdef CAPABILITIES
312 	seq_write_begin(&fde->fde_seq);
313 #endif
314 	fdefree_last(fde);
315 	bzero(fde, fde_change_size);
316 	fdunused(fdp, fd);
317 #ifdef CAPABILITIES
318 	seq_write_end(&fde->fde_seq);
319 #endif
320 }
321 
322 /*
323  * System calls on descriptors.
324  */
325 #ifndef _SYS_SYSPROTO_H_
326 struct getdtablesize_args {
327 	int	dummy;
328 };
329 #endif
330 /* ARGSUSED */
331 int
332 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
333 {
334 	struct proc *p = td->td_proc;
335 	uint64_t lim;
336 
337 	PROC_LOCK(p);
338 	td->td_retval[0] =
339 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
340 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
341 	PROC_UNLOCK(p);
342 	if (lim < td->td_retval[0])
343 		td->td_retval[0] = lim;
344 	return (0);
345 }
346 
347 /*
348  * Duplicate a file descriptor to a particular value.
349  *
350  * Note: keep in mind that a potential race condition exists when closing
351  * descriptors from a shared descriptor table (via rfork).
352  */
353 #ifndef _SYS_SYSPROTO_H_
354 struct dup2_args {
355 	u_int	from;
356 	u_int	to;
357 };
358 #endif
359 /* ARGSUSED */
360 int
361 sys_dup2(struct thread *td, struct dup2_args *uap)
362 {
363 
364 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to));
365 }
366 
367 /*
368  * Duplicate a file descriptor.
369  */
370 #ifndef _SYS_SYSPROTO_H_
371 struct dup_args {
372 	u_int	fd;
373 };
374 #endif
375 /* ARGSUSED */
376 int
377 sys_dup(struct thread *td, struct dup_args *uap)
378 {
379 
380 	return (do_dup(td, 0, (int)uap->fd, 0));
381 }
382 
383 /*
384  * The file control system call.
385  */
386 #ifndef _SYS_SYSPROTO_H_
387 struct fcntl_args {
388 	int	fd;
389 	int	cmd;
390 	long	arg;
391 };
392 #endif
393 /* ARGSUSED */
394 int
395 sys_fcntl(struct thread *td, struct fcntl_args *uap)
396 {
397 
398 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
399 }
400 
401 int
402 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
403 {
404 	struct flock fl;
405 	struct __oflock ofl;
406 	intptr_t arg1;
407 	int error;
408 
409 	error = 0;
410 	switch (cmd) {
411 	case F_OGETLK:
412 	case F_OSETLK:
413 	case F_OSETLKW:
414 		/*
415 		 * Convert old flock structure to new.
416 		 */
417 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
418 		fl.l_start = ofl.l_start;
419 		fl.l_len = ofl.l_len;
420 		fl.l_pid = ofl.l_pid;
421 		fl.l_type = ofl.l_type;
422 		fl.l_whence = ofl.l_whence;
423 		fl.l_sysid = 0;
424 
425 		switch (cmd) {
426 		case F_OGETLK:
427 		    cmd = F_GETLK;
428 		    break;
429 		case F_OSETLK:
430 		    cmd = F_SETLK;
431 		    break;
432 		case F_OSETLKW:
433 		    cmd = F_SETLKW;
434 		    break;
435 		}
436 		arg1 = (intptr_t)&fl;
437 		break;
438         case F_GETLK:
439         case F_SETLK:
440         case F_SETLKW:
441 	case F_SETLK_REMOTE:
442                 error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
443                 arg1 = (intptr_t)&fl;
444                 break;
445 	default:
446 		arg1 = arg;
447 		break;
448 	}
449 	if (error)
450 		return (error);
451 	error = kern_fcntl(td, fd, cmd, arg1);
452 	if (error)
453 		return (error);
454 	if (cmd == F_OGETLK) {
455 		ofl.l_start = fl.l_start;
456 		ofl.l_len = fl.l_len;
457 		ofl.l_pid = fl.l_pid;
458 		ofl.l_type = fl.l_type;
459 		ofl.l_whence = fl.l_whence;
460 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
461 	} else if (cmd == F_GETLK) {
462 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
463 	}
464 	return (error);
465 }
466 
467 int
468 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
469 {
470 	struct filedesc *fdp;
471 	struct flock *flp;
472 	struct file *fp, *fp2;
473 	struct filedescent *fde;
474 	struct proc *p;
475 	struct vnode *vp;
476 	cap_rights_t rights;
477 	int error, flg, tmp;
478 	uint64_t bsize;
479 	off_t foffset;
480 
481 	error = 0;
482 	flg = F_POSIX;
483 	p = td->td_proc;
484 	fdp = p->p_fd;
485 
486 	switch (cmd) {
487 	case F_DUPFD:
488 		tmp = arg;
489 		error = do_dup(td, DUP_FCNTL, fd, tmp);
490 		break;
491 
492 	case F_DUPFD_CLOEXEC:
493 		tmp = arg;
494 		error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp);
495 		break;
496 
497 	case F_DUP2FD:
498 		tmp = arg;
499 		error = do_dup(td, DUP_FIXED, fd, tmp);
500 		break;
501 
502 	case F_DUP2FD_CLOEXEC:
503 		tmp = arg;
504 		error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp);
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 {
805 	struct filedesc *fdp;
806 	struct filedescent *oldfde, *newfde;
807 	struct proc *p;
808 	struct file *fp;
809 	struct file *delfp;
810 	int error, maxfd;
811 
812 	p = td->td_proc;
813 	fdp = p->p_fd;
814 
815 	/*
816 	 * Verify we have a valid descriptor to dup from and possibly to
817 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
818 	 * return EINVAL when the new descriptor is out of bounds.
819 	 */
820 	if (old < 0)
821 		return (EBADF);
822 	if (new < 0)
823 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
824 	maxfd = getmaxfd(p);
825 	if (new >= maxfd)
826 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
827 
828 	FILEDESC_XLOCK(fdp);
829 	if (fget_locked(fdp, old) == NULL) {
830 		FILEDESC_XUNLOCK(fdp);
831 		return (EBADF);
832 	}
833 	oldfde = &fdp->fd_ofiles[old];
834 	if (flags & DUP_FIXED && old == new) {
835 		td->td_retval[0] = new;
836 		if (flags & DUP_CLOEXEC)
837 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
838 		FILEDESC_XUNLOCK(fdp);
839 		return (0);
840 	}
841 	fp = oldfde->fde_file;
842 	fhold(fp);
843 
844 	/*
845 	 * If the caller specified a file descriptor, make sure the file
846 	 * table is large enough to hold it, and grab it.  Otherwise, just
847 	 * allocate a new descriptor the usual way.
848 	 */
849 	if (flags & DUP_FIXED) {
850 		if (new >= fdp->fd_nfiles) {
851 			/*
852 			 * The resource limits are here instead of e.g.
853 			 * fdalloc(), because the file descriptor table may be
854 			 * shared between processes, so we can't really use
855 			 * racct_add()/racct_sub().  Instead of counting the
856 			 * number of actually allocated descriptors, just put
857 			 * the limit on the size of the file descriptor table.
858 			 */
859 #ifdef RACCT
860 			PROC_LOCK(p);
861 			error = racct_set(p, RACCT_NOFILE, new + 1);
862 			PROC_UNLOCK(p);
863 			if (error != 0) {
864 				FILEDESC_XUNLOCK(fdp);
865 				fdrop(fp, td);
866 				return (EMFILE);
867 			}
868 #endif
869 			fdgrowtable_exp(fdp, new + 1);
870 			oldfde = &fdp->fd_ofiles[old];
871 		}
872 		newfde = &fdp->fd_ofiles[new];
873 		if (newfde->fde_file == NULL)
874 			fdused(fdp, new);
875 	} else {
876 		if ((error = fdalloc(td, new, &new)) != 0) {
877 			FILEDESC_XUNLOCK(fdp);
878 			fdrop(fp, td);
879 			return (error);
880 		}
881 		newfde = &fdp->fd_ofiles[new];
882 	}
883 
884 	KASSERT(fp == oldfde->fde_file, ("old fd has been modified"));
885 	KASSERT(old != new, ("new fd is same as old"));
886 
887 	delfp = newfde->fde_file;
888 
889 	/*
890 	 * Duplicate the source descriptor.
891 	 */
892 #ifdef CAPABILITIES
893 	seq_write_begin(&newfde->fde_seq);
894 #endif
895 	filecaps_free(&newfde->fde_caps);
896 	memcpy(newfde, oldfde, fde_change_size);
897 	filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
898 	if ((flags & DUP_CLOEXEC) != 0)
899 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
900 	else
901 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
902 #ifdef CAPABILITIES
903 	seq_write_end(&newfde->fde_seq);
904 #endif
905 	td->td_retval[0] = new;
906 
907 	if (delfp != NULL) {
908 		(void) closefp(fdp, new, delfp, td, 1);
909 		/* closefp() drops the FILEDESC lock for us. */
910 	} else {
911 		FILEDESC_XUNLOCK(fdp);
912 	}
913 
914 	return (0);
915 }
916 
917 /*
918  * If sigio is on the list associated with a process or process group,
919  * disable signalling from the device, remove sigio from the list and
920  * free sigio.
921  */
922 void
923 funsetown(struct sigio **sigiop)
924 {
925 	struct sigio *sigio;
926 
927 	SIGIO_LOCK();
928 	sigio = *sigiop;
929 	if (sigio == NULL) {
930 		SIGIO_UNLOCK();
931 		return;
932 	}
933 	*(sigio->sio_myref) = NULL;
934 	if ((sigio)->sio_pgid < 0) {
935 		struct pgrp *pg = (sigio)->sio_pgrp;
936 		PGRP_LOCK(pg);
937 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
938 			     sigio, sio_pgsigio);
939 		PGRP_UNLOCK(pg);
940 	} else {
941 		struct proc *p = (sigio)->sio_proc;
942 		PROC_LOCK(p);
943 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
944 			     sigio, sio_pgsigio);
945 		PROC_UNLOCK(p);
946 	}
947 	SIGIO_UNLOCK();
948 	crfree(sigio->sio_ucred);
949 	free(sigio, M_SIGIO);
950 }
951 
952 /*
953  * Free a list of sigio structures.
954  * We only need to lock the SIGIO_LOCK because we have made ourselves
955  * inaccessible to callers of fsetown and therefore do not need to lock
956  * the proc or pgrp struct for the list manipulation.
957  */
958 void
959 funsetownlst(struct sigiolst *sigiolst)
960 {
961 	struct proc *p;
962 	struct pgrp *pg;
963 	struct sigio *sigio;
964 
965 	sigio = SLIST_FIRST(sigiolst);
966 	if (sigio == NULL)
967 		return;
968 	p = NULL;
969 	pg = NULL;
970 
971 	/*
972 	 * Every entry of the list should belong
973 	 * to a single proc or pgrp.
974 	 */
975 	if (sigio->sio_pgid < 0) {
976 		pg = sigio->sio_pgrp;
977 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
978 	} else /* if (sigio->sio_pgid > 0) */ {
979 		p = sigio->sio_proc;
980 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
981 	}
982 
983 	SIGIO_LOCK();
984 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
985 		*(sigio->sio_myref) = NULL;
986 		if (pg != NULL) {
987 			KASSERT(sigio->sio_pgid < 0,
988 			    ("Proc sigio in pgrp sigio list"));
989 			KASSERT(sigio->sio_pgrp == pg,
990 			    ("Bogus pgrp in sigio list"));
991 			PGRP_LOCK(pg);
992 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
993 			    sio_pgsigio);
994 			PGRP_UNLOCK(pg);
995 		} else /* if (p != NULL) */ {
996 			KASSERT(sigio->sio_pgid > 0,
997 			    ("Pgrp sigio in proc sigio list"));
998 			KASSERT(sigio->sio_proc == p,
999 			    ("Bogus proc in sigio list"));
1000 			PROC_LOCK(p);
1001 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1002 			    sio_pgsigio);
1003 			PROC_UNLOCK(p);
1004 		}
1005 		SIGIO_UNLOCK();
1006 		crfree(sigio->sio_ucred);
1007 		free(sigio, M_SIGIO);
1008 		SIGIO_LOCK();
1009 	}
1010 	SIGIO_UNLOCK();
1011 }
1012 
1013 /*
1014  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1015  *
1016  * After permission checking, add a sigio structure to the sigio list for
1017  * the process or process group.
1018  */
1019 int
1020 fsetown(pid_t pgid, struct sigio **sigiop)
1021 {
1022 	struct proc *proc;
1023 	struct pgrp *pgrp;
1024 	struct sigio *sigio;
1025 	int ret;
1026 
1027 	if (pgid == 0) {
1028 		funsetown(sigiop);
1029 		return (0);
1030 	}
1031 
1032 	ret = 0;
1033 
1034 	/* Allocate and fill in the new sigio out of locks. */
1035 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1036 	sigio->sio_pgid = pgid;
1037 	sigio->sio_ucred = crhold(curthread->td_ucred);
1038 	sigio->sio_myref = sigiop;
1039 
1040 	sx_slock(&proctree_lock);
1041 	if (pgid > 0) {
1042 		proc = pfind(pgid);
1043 		if (proc == NULL) {
1044 			ret = ESRCH;
1045 			goto fail;
1046 		}
1047 
1048 		/*
1049 		 * Policy - Don't allow a process to FSETOWN a process
1050 		 * in another session.
1051 		 *
1052 		 * Remove this test to allow maximum flexibility or
1053 		 * restrict FSETOWN to the current process or process
1054 		 * group for maximum safety.
1055 		 */
1056 		PROC_UNLOCK(proc);
1057 		if (proc->p_session != curthread->td_proc->p_session) {
1058 			ret = EPERM;
1059 			goto fail;
1060 		}
1061 
1062 		pgrp = NULL;
1063 	} else /* if (pgid < 0) */ {
1064 		pgrp = pgfind(-pgid);
1065 		if (pgrp == NULL) {
1066 			ret = ESRCH;
1067 			goto fail;
1068 		}
1069 		PGRP_UNLOCK(pgrp);
1070 
1071 		/*
1072 		 * Policy - Don't allow a process to FSETOWN a process
1073 		 * in another session.
1074 		 *
1075 		 * Remove this test to allow maximum flexibility or
1076 		 * restrict FSETOWN to the current process or process
1077 		 * group for maximum safety.
1078 		 */
1079 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1080 			ret = EPERM;
1081 			goto fail;
1082 		}
1083 
1084 		proc = NULL;
1085 	}
1086 	funsetown(sigiop);
1087 	if (pgid > 0) {
1088 		PROC_LOCK(proc);
1089 		/*
1090 		 * Since funsetownlst() is called without the proctree
1091 		 * locked, we need to check for P_WEXIT.
1092 		 * XXX: is ESRCH correct?
1093 		 */
1094 		if ((proc->p_flag & P_WEXIT) != 0) {
1095 			PROC_UNLOCK(proc);
1096 			ret = ESRCH;
1097 			goto fail;
1098 		}
1099 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1100 		sigio->sio_proc = proc;
1101 		PROC_UNLOCK(proc);
1102 	} else {
1103 		PGRP_LOCK(pgrp);
1104 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1105 		sigio->sio_pgrp = pgrp;
1106 		PGRP_UNLOCK(pgrp);
1107 	}
1108 	sx_sunlock(&proctree_lock);
1109 	SIGIO_LOCK();
1110 	*sigiop = sigio;
1111 	SIGIO_UNLOCK();
1112 	return (0);
1113 
1114 fail:
1115 	sx_sunlock(&proctree_lock);
1116 	crfree(sigio->sio_ucred);
1117 	free(sigio, M_SIGIO);
1118 	return (ret);
1119 }
1120 
1121 /*
1122  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1123  */
1124 pid_t
1125 fgetown(sigiop)
1126 	struct sigio **sigiop;
1127 {
1128 	pid_t pgid;
1129 
1130 	SIGIO_LOCK();
1131 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1132 	SIGIO_UNLOCK();
1133 	return (pgid);
1134 }
1135 
1136 /*
1137  * Function drops the filedesc lock on return.
1138  */
1139 static int
1140 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1141     int holdleaders)
1142 {
1143 	int error;
1144 
1145 	FILEDESC_XLOCK_ASSERT(fdp);
1146 
1147 	if (holdleaders) {
1148 		if (td->td_proc->p_fdtol != NULL) {
1149 			/*
1150 			 * Ask fdfree() to sleep to ensure that all relevant
1151 			 * process leaders can be traversed in closef().
1152 			 */
1153 			fdp->fd_holdleaderscount++;
1154 		} else {
1155 			holdleaders = 0;
1156 		}
1157 	}
1158 
1159 	/*
1160 	 * We now hold the fp reference that used to be owned by the
1161 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1162 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1163 	 * added, and deleteing a knote for the new fd.
1164 	 */
1165 	knote_fdclose(td, fd);
1166 
1167 	/*
1168 	 * We need to notify mqueue if the object is of type mqueue.
1169 	 */
1170 	if (fp->f_type == DTYPE_MQUEUE)
1171 		mq_fdclose(td, fd, fp);
1172 	FILEDESC_XUNLOCK(fdp);
1173 
1174 	error = closef(fp, td);
1175 	if (holdleaders) {
1176 		FILEDESC_XLOCK(fdp);
1177 		fdp->fd_holdleaderscount--;
1178 		if (fdp->fd_holdleaderscount == 0 &&
1179 		    fdp->fd_holdleaderswakeup != 0) {
1180 			fdp->fd_holdleaderswakeup = 0;
1181 			wakeup(&fdp->fd_holdleaderscount);
1182 		}
1183 		FILEDESC_XUNLOCK(fdp);
1184 	}
1185 	return (error);
1186 }
1187 
1188 /*
1189  * Close a file descriptor.
1190  */
1191 #ifndef _SYS_SYSPROTO_H_
1192 struct close_args {
1193 	int     fd;
1194 };
1195 #endif
1196 /* ARGSUSED */
1197 int
1198 sys_close(td, uap)
1199 	struct thread *td;
1200 	struct close_args *uap;
1201 {
1202 
1203 	return (kern_close(td, uap->fd));
1204 }
1205 
1206 int
1207 kern_close(td, fd)
1208 	struct thread *td;
1209 	int fd;
1210 {
1211 	struct filedesc *fdp;
1212 	struct file *fp;
1213 
1214 	fdp = td->td_proc->p_fd;
1215 
1216 	AUDIT_SYSCLOSE(td, fd);
1217 
1218 	FILEDESC_XLOCK(fdp);
1219 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1220 		FILEDESC_XUNLOCK(fdp);
1221 		return (EBADF);
1222 	}
1223 	fdfree(fdp, fd);
1224 
1225 	/* closefp() drops the FILEDESC lock for us. */
1226 	return (closefp(fdp, fd, fp, td, 1));
1227 }
1228 
1229 /*
1230  * Close open file descriptors.
1231  */
1232 #ifndef _SYS_SYSPROTO_H_
1233 struct closefrom_args {
1234 	int	lowfd;
1235 };
1236 #endif
1237 /* ARGSUSED */
1238 int
1239 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1240 {
1241 	struct filedesc *fdp;
1242 	int fd;
1243 
1244 	fdp = td->td_proc->p_fd;
1245 	AUDIT_ARG_FD(uap->lowfd);
1246 
1247 	/*
1248 	 * Treat negative starting file descriptor values identical to
1249 	 * closefrom(0) which closes all files.
1250 	 */
1251 	if (uap->lowfd < 0)
1252 		uap->lowfd = 0;
1253 	FILEDESC_SLOCK(fdp);
1254 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1255 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1256 			FILEDESC_SUNLOCK(fdp);
1257 			(void)kern_close(td, fd);
1258 			FILEDESC_SLOCK(fdp);
1259 		}
1260 	}
1261 	FILEDESC_SUNLOCK(fdp);
1262 	return (0);
1263 }
1264 
1265 #if defined(COMPAT_43)
1266 /*
1267  * Return status information about a file descriptor.
1268  */
1269 #ifndef _SYS_SYSPROTO_H_
1270 struct ofstat_args {
1271 	int	fd;
1272 	struct	ostat *sb;
1273 };
1274 #endif
1275 /* ARGSUSED */
1276 int
1277 ofstat(struct thread *td, struct ofstat_args *uap)
1278 {
1279 	struct ostat oub;
1280 	struct stat ub;
1281 	int error;
1282 
1283 	error = kern_fstat(td, uap->fd, &ub);
1284 	if (error == 0) {
1285 		cvtstat(&ub, &oub);
1286 		error = copyout(&oub, uap->sb, sizeof(oub));
1287 	}
1288 	return (error);
1289 }
1290 #endif /* COMPAT_43 */
1291 
1292 /*
1293  * Return status information about a file descriptor.
1294  */
1295 #ifndef _SYS_SYSPROTO_H_
1296 struct fstat_args {
1297 	int	fd;
1298 	struct	stat *sb;
1299 };
1300 #endif
1301 /* ARGSUSED */
1302 int
1303 sys_fstat(struct thread *td, struct fstat_args *uap)
1304 {
1305 	struct stat ub;
1306 	int error;
1307 
1308 	error = kern_fstat(td, uap->fd, &ub);
1309 	if (error == 0)
1310 		error = copyout(&ub, uap->sb, sizeof(ub));
1311 	return (error);
1312 }
1313 
1314 int
1315 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1316 {
1317 	struct file *fp;
1318 	cap_rights_t rights;
1319 	int error;
1320 
1321 	AUDIT_ARG_FD(fd);
1322 
1323 	error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1324 	if (error != 0)
1325 		return (error);
1326 
1327 	AUDIT_ARG_FILE(td->td_proc, fp);
1328 
1329 	error = fo_stat(fp, sbp, td->td_ucred, td);
1330 	fdrop(fp, td);
1331 #ifdef KTRACE
1332 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1333 		ktrstat(sbp);
1334 #endif
1335 	return (error);
1336 }
1337 
1338 /*
1339  * Return status information about a file descriptor.
1340  */
1341 #ifndef _SYS_SYSPROTO_H_
1342 struct nfstat_args {
1343 	int	fd;
1344 	struct	nstat *sb;
1345 };
1346 #endif
1347 /* ARGSUSED */
1348 int
1349 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1350 {
1351 	struct nstat nub;
1352 	struct stat ub;
1353 	int error;
1354 
1355 	error = kern_fstat(td, uap->fd, &ub);
1356 	if (error == 0) {
1357 		cvtnstat(&ub, &nub);
1358 		error = copyout(&nub, uap->sb, sizeof(nub));
1359 	}
1360 	return (error);
1361 }
1362 
1363 /*
1364  * Return pathconf information about a file descriptor.
1365  */
1366 #ifndef _SYS_SYSPROTO_H_
1367 struct fpathconf_args {
1368 	int	fd;
1369 	int	name;
1370 };
1371 #endif
1372 /* ARGSUSED */
1373 int
1374 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1375 {
1376 	struct file *fp;
1377 	struct vnode *vp;
1378 	cap_rights_t rights;
1379 	int error;
1380 
1381 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1382 	if (error != 0)
1383 		return (error);
1384 
1385 	/* If asynchronous I/O is available, it works for all descriptors. */
1386 	if (uap->name == _PC_ASYNC_IO) {
1387 		td->td_retval[0] = async_io_version;
1388 		goto out;
1389 	}
1390 	vp = fp->f_vnode;
1391 	if (vp != NULL) {
1392 		vn_lock(vp, LK_SHARED | LK_RETRY);
1393 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1394 		VOP_UNLOCK(vp, 0);
1395 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1396 		if (uap->name != _PC_PIPE_BUF) {
1397 			error = EINVAL;
1398 		} else {
1399 			td->td_retval[0] = PIPE_BUF;
1400 			error = 0;
1401 		}
1402 	} else {
1403 		error = EOPNOTSUPP;
1404 	}
1405 out:
1406 	fdrop(fp, td);
1407 	return (error);
1408 }
1409 
1410 /*
1411  * Initialize filecaps structure.
1412  */
1413 void
1414 filecaps_init(struct filecaps *fcaps)
1415 {
1416 
1417 	bzero(fcaps, sizeof(*fcaps));
1418 	fcaps->fc_nioctls = -1;
1419 }
1420 
1421 /*
1422  * Copy filecaps structure allocating memory for ioctls array if needed.
1423  */
1424 void
1425 filecaps_copy(const struct filecaps *src, struct filecaps *dst)
1426 {
1427 	size_t size;
1428 
1429 	*dst = *src;
1430 	if (src->fc_ioctls != NULL) {
1431 		KASSERT(src->fc_nioctls > 0,
1432 		    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1433 
1434 		size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1435 		dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1436 		bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1437 	}
1438 }
1439 
1440 /*
1441  * Move filecaps structure to the new place and clear the old place.
1442  */
1443 void
1444 filecaps_move(struct filecaps *src, struct filecaps *dst)
1445 {
1446 
1447 	*dst = *src;
1448 	bzero(src, sizeof(*src));
1449 }
1450 
1451 /*
1452  * Fill the given filecaps structure with full rights.
1453  */
1454 static void
1455 filecaps_fill(struct filecaps *fcaps)
1456 {
1457 
1458 	CAP_ALL(&fcaps->fc_rights);
1459 	fcaps->fc_ioctls = NULL;
1460 	fcaps->fc_nioctls = -1;
1461 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1462 }
1463 
1464 /*
1465  * Free memory allocated within filecaps structure.
1466  */
1467 void
1468 filecaps_free(struct filecaps *fcaps)
1469 {
1470 
1471 	free(fcaps->fc_ioctls, M_FILECAPS);
1472 	bzero(fcaps, sizeof(*fcaps));
1473 }
1474 
1475 /*
1476  * Validate the given filecaps structure.
1477  */
1478 static void
1479 filecaps_validate(const struct filecaps *fcaps, const char *func)
1480 {
1481 
1482 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1483 	    ("%s: invalid rights", func));
1484 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1485 	    ("%s: invalid fcntls", func));
1486 	KASSERT(fcaps->fc_fcntls == 0 ||
1487 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1488 	    ("%s: fcntls without CAP_FCNTL", func));
1489 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1490 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1491 	    ("%s: invalid ioctls", func));
1492 	KASSERT(fcaps->fc_nioctls == 0 ||
1493 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1494 	    ("%s: ioctls without CAP_IOCTL", func));
1495 }
1496 
1497 static void
1498 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1499 {
1500 	int nfd1;
1501 
1502 	FILEDESC_XLOCK_ASSERT(fdp);
1503 
1504 	nfd1 = fdp->fd_nfiles * 2;
1505 	if (nfd1 < nfd)
1506 		nfd1 = nfd;
1507 	fdgrowtable(fdp, nfd1);
1508 }
1509 
1510 /*
1511  * Grow the file table to accomodate (at least) nfd descriptors.
1512  */
1513 static void
1514 fdgrowtable(struct filedesc *fdp, int nfd)
1515 {
1516 	struct filedesc0 *fdp0;
1517 	struct freetable *ft;
1518 	struct fdescenttbl *ntable;
1519 	struct fdescenttbl *otable;
1520 	int nnfiles, onfiles;
1521 	NDSLOTTYPE *nmap, *omap;
1522 
1523 	/*
1524 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1525 	 * growing it to accomodate for the one we are going to copy from. There
1526 	 * is no need to have a lock on this one as it's not visible to anyone.
1527 	 */
1528 	if (fdp->fd_lastfile != -1)
1529 		FILEDESC_XLOCK_ASSERT(fdp);
1530 
1531 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1532 
1533 	/* save old values */
1534 	onfiles = fdp->fd_nfiles;
1535 	otable = fdp->fd_files;
1536 	omap = fdp->fd_map;
1537 
1538 	/* compute the size of the new table */
1539 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1540 	if (nnfiles <= onfiles)
1541 		/* the table is already large enough */
1542 		return;
1543 
1544 	/*
1545 	 * Allocate a new table.  We need enough space for the number of
1546 	 * entries, file entries themselves and the struct freetable we will use
1547 	 * when we decommission the table and place it on the freelist.
1548 	 * We place the struct freetable in the middle so we don't have
1549 	 * to worry about padding.
1550 	 */
1551 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1552 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1553 	    sizeof(struct freetable),
1554 	    M_FILEDESC, M_ZERO | M_WAITOK);
1555 	/* copy the old data */
1556 	ntable->fdt_nfiles = nnfiles;
1557 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1558 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1559 
1560 	/*
1561 	 * Allocate a new map only if the old is not large enough.  It will
1562 	 * grow at a slower rate than the table as it can map more
1563 	 * entries than the table can hold.
1564 	 */
1565 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1566 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1567 		    M_ZERO | M_WAITOK);
1568 		/* copy over the old data and update the pointer */
1569 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1570 		fdp->fd_map = nmap;
1571 	}
1572 
1573 	/*
1574 	 * Make sure that ntable is correctly initialized before we replace
1575 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1576 	 * data.
1577 	 */
1578 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1579 
1580 	/*
1581 	 * Do not free the old file table, as some threads may still
1582 	 * reference entries within it.  Instead, place it on a freelist
1583 	 * which will be processed when the struct filedesc is released.
1584 	 *
1585 	 * Note that if onfiles == NDFILE, we're dealing with the original
1586 	 * static allocation contained within (struct filedesc0 *)fdp,
1587 	 * which must not be freed.
1588 	 */
1589 	if (onfiles > NDFILE) {
1590 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1591 		fdp0 = (struct filedesc0 *)fdp;
1592 		ft->ft_table = otable;
1593 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1594 	}
1595 	/*
1596 	 * The map does not have the same possibility of threads still
1597 	 * holding references to it.  So always free it as long as it
1598 	 * does not reference the original static allocation.
1599 	 */
1600 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1601 		free(omap, M_FILEDESC);
1602 }
1603 
1604 /*
1605  * Allocate a file descriptor for the process.
1606  */
1607 int
1608 fdalloc(struct thread *td, int minfd, int *result)
1609 {
1610 	struct proc *p = td->td_proc;
1611 	struct filedesc *fdp = p->p_fd;
1612 	int fd = -1, maxfd, allocfd;
1613 #ifdef RACCT
1614 	int error;
1615 #endif
1616 
1617 	FILEDESC_XLOCK_ASSERT(fdp);
1618 
1619 	if (fdp->fd_freefile > minfd)
1620 		minfd = fdp->fd_freefile;
1621 
1622 	maxfd = getmaxfd(p);
1623 
1624 	/*
1625 	 * Search the bitmap for a free descriptor starting at minfd.
1626 	 * If none is found, grow the file table.
1627 	 */
1628 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1629 	if (fd >= maxfd)
1630 		return (EMFILE);
1631 	if (fd >= fdp->fd_nfiles) {
1632 		allocfd = min(fd * 2, maxfd);
1633 #ifdef RACCT
1634 		PROC_LOCK(p);
1635 		error = racct_set(p, RACCT_NOFILE, allocfd);
1636 		PROC_UNLOCK(p);
1637 		if (error != 0)
1638 			return (EMFILE);
1639 #endif
1640 		/*
1641 		 * fd is already equal to first free descriptor >= minfd, so
1642 		 * we only need to grow the table and we are done.
1643 		 */
1644 		fdgrowtable_exp(fdp, allocfd);
1645 	}
1646 
1647 	/*
1648 	 * Perform some sanity checks, then mark the file descriptor as
1649 	 * used and return it to the caller.
1650 	 */
1651 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1652 	    ("invalid descriptor %d", fd));
1653 	KASSERT(!fdisused(fdp, fd),
1654 	    ("fd_first_free() returned non-free descriptor"));
1655 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1656 	    ("file descriptor isn't free"));
1657 	KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
1658 	fdused(fdp, fd);
1659 	*result = fd;
1660 	return (0);
1661 }
1662 
1663 /*
1664  * Allocate n file descriptors for the process.
1665  */
1666 int
1667 fdallocn(struct thread *td, int minfd, int *fds, int n)
1668 {
1669 	struct proc *p = td->td_proc;
1670 	struct filedesc *fdp = p->p_fd;
1671 	int i;
1672 
1673 	FILEDESC_XLOCK_ASSERT(fdp);
1674 
1675 	for (i = 0; i < n; i++)
1676 		if (fdalloc(td, 0, &fds[i]) != 0)
1677 			break;
1678 
1679 	if (i < n) {
1680 		for (i--; i >= 0; i--)
1681 			fdunused(fdp, fds[i]);
1682 		return (EMFILE);
1683 	}
1684 
1685 	return (0);
1686 }
1687 
1688 /*
1689  * Create a new open file structure and allocate a file decriptor for the
1690  * process that refers to it.  We add one reference to the file for the
1691  * descriptor table and one reference for resultfp. This is to prevent us
1692  * being preempted and the entry in the descriptor table closed after we
1693  * release the FILEDESC lock.
1694  */
1695 int
1696 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1697 {
1698 	struct file *fp;
1699 	int error, fd;
1700 
1701 	error = falloc_noinstall(td, &fp);
1702 	if (error)
1703 		return (error);		/* no reference held on error */
1704 
1705 	error = finstall(td, fp, &fd, flags, NULL);
1706 	if (error) {
1707 		fdrop(fp, td);		/* one reference (fp only) */
1708 		return (error);
1709 	}
1710 
1711 	if (resultfp != NULL)
1712 		*resultfp = fp;		/* copy out result */
1713 	else
1714 		fdrop(fp, td);		/* release local reference */
1715 
1716 	if (resultfd != NULL)
1717 		*resultfd = fd;
1718 
1719 	return (0);
1720 }
1721 
1722 /*
1723  * Create a new open file structure without allocating a file descriptor.
1724  */
1725 int
1726 falloc_noinstall(struct thread *td, struct file **resultfp)
1727 {
1728 	struct file *fp;
1729 	int maxuserfiles = maxfiles - (maxfiles / 20);
1730 	static struct timeval lastfail;
1731 	static int curfail;
1732 
1733 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1734 
1735 	if ((openfiles >= maxuserfiles &&
1736 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1737 	    openfiles >= maxfiles) {
1738 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1739 			printf("kern.maxfiles limit exceeded by uid %i, "
1740 			    "please see tuning(7).\n", td->td_ucred->cr_ruid);
1741 		}
1742 		return (ENFILE);
1743 	}
1744 	atomic_add_int(&openfiles, 1);
1745 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1746 	refcount_init(&fp->f_count, 1);
1747 	fp->f_cred = crhold(td->td_ucred);
1748 	fp->f_ops = &badfileops;
1749 	*resultfp = fp;
1750 	return (0);
1751 }
1752 
1753 /*
1754  * Install a file in a file descriptor table.
1755  */
1756 int
1757 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1758     struct filecaps *fcaps)
1759 {
1760 	struct filedesc *fdp = td->td_proc->p_fd;
1761 	struct filedescent *fde;
1762 	int error;
1763 
1764 	KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1765 	KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1766 	if (fcaps != NULL)
1767 		filecaps_validate(fcaps, __func__);
1768 
1769 	FILEDESC_XLOCK(fdp);
1770 	if ((error = fdalloc(td, 0, fd))) {
1771 		FILEDESC_XUNLOCK(fdp);
1772 		return (error);
1773 	}
1774 	fhold(fp);
1775 	fde = &fdp->fd_ofiles[*fd];
1776 #ifdef CAPABILITIES
1777 	seq_write_begin(&fde->fde_seq);
1778 #endif
1779 	fde->fde_file = fp;
1780 	if ((flags & O_CLOEXEC) != 0)
1781 		fde->fde_flags |= UF_EXCLOSE;
1782 	if (fcaps != NULL)
1783 		filecaps_move(fcaps, &fde->fde_caps);
1784 	else
1785 		filecaps_fill(&fde->fde_caps);
1786 #ifdef CAPABILITIES
1787 	seq_write_end(&fde->fde_seq);
1788 #endif
1789 	FILEDESC_XUNLOCK(fdp);
1790 	return (0);
1791 }
1792 
1793 /*
1794  * Build a new filedesc structure from another.
1795  * Copy the current, root, and jail root vnode references.
1796  *
1797  * If fdp is not NULL, return with it shared locked.
1798  */
1799 struct filedesc *
1800 fdinit(struct filedesc *fdp, bool prepfiles)
1801 {
1802 	struct filedesc0 *newfdp0;
1803 	struct filedesc *newfdp;
1804 
1805 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1806 	newfdp = &newfdp0->fd_fd;
1807 
1808 	/* Create the file descriptor table. */
1809 	FILEDESC_LOCK_INIT(newfdp);
1810 	newfdp->fd_refcnt = 1;
1811 	newfdp->fd_holdcnt = 1;
1812 	newfdp->fd_cmask = CMASK;
1813 	newfdp->fd_map = newfdp0->fd_dmap;
1814 	newfdp->fd_lastfile = -1;
1815 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1816 	newfdp->fd_files->fdt_nfiles = NDFILE;
1817 
1818 	if (fdp == NULL)
1819 		return (newfdp);
1820 
1821 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1822 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1823 
1824 	FILEDESC_SLOCK(fdp);
1825 	newfdp->fd_cdir = fdp->fd_cdir;
1826 	if (newfdp->fd_cdir)
1827 		VREF(newfdp->fd_cdir);
1828 	newfdp->fd_rdir = fdp->fd_rdir;
1829 	if (newfdp->fd_rdir)
1830 		VREF(newfdp->fd_rdir);
1831 	newfdp->fd_jdir = fdp->fd_jdir;
1832 	if (newfdp->fd_jdir)
1833 		VREF(newfdp->fd_jdir);
1834 
1835 	if (!prepfiles) {
1836 		FILEDESC_SUNLOCK(fdp);
1837 	} else {
1838 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1839 			FILEDESC_SUNLOCK(fdp);
1840 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1841 			FILEDESC_SLOCK(fdp);
1842 		}
1843 	}
1844 
1845 	return (newfdp);
1846 }
1847 
1848 static struct filedesc *
1849 fdhold(struct proc *p)
1850 {
1851 	struct filedesc *fdp;
1852 
1853 	mtx_lock(&fdesc_mtx);
1854 	fdp = p->p_fd;
1855 	if (fdp != NULL)
1856 		fdp->fd_holdcnt++;
1857 	mtx_unlock(&fdesc_mtx);
1858 	return (fdp);
1859 }
1860 
1861 static void
1862 fddrop(struct filedesc *fdp)
1863 {
1864 	int i;
1865 
1866 	if (fdp->fd_holdcnt > 1) {
1867 		mtx_lock(&fdesc_mtx);
1868 		i = --fdp->fd_holdcnt;
1869 		mtx_unlock(&fdesc_mtx);
1870 		if (i > 0)
1871 			return;
1872 	}
1873 
1874 	FILEDESC_LOCK_DESTROY(fdp);
1875 	uma_zfree(filedesc0_zone, fdp);
1876 }
1877 
1878 /*
1879  * Share a filedesc structure.
1880  */
1881 struct filedesc *
1882 fdshare(struct filedesc *fdp)
1883 {
1884 
1885 	FILEDESC_XLOCK(fdp);
1886 	fdp->fd_refcnt++;
1887 	FILEDESC_XUNLOCK(fdp);
1888 	return (fdp);
1889 }
1890 
1891 /*
1892  * Unshare a filedesc structure, if necessary by making a copy
1893  */
1894 void
1895 fdunshare(struct thread *td)
1896 {
1897 	struct filedesc *tmp;
1898 	struct proc *p = td->td_proc;
1899 
1900 	if (p->p_fd->fd_refcnt == 1)
1901 		return;
1902 
1903 	tmp = fdcopy(p->p_fd);
1904 	fdescfree(td);
1905 	p->p_fd = tmp;
1906 }
1907 
1908 /*
1909  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1910  * this is to ease callers, not catch errors.
1911  */
1912 struct filedesc *
1913 fdcopy(struct filedesc *fdp)
1914 {
1915 	struct filedesc *newfdp;
1916 	struct filedescent *nfde, *ofde;
1917 	int i;
1918 
1919 	MPASS(fdp != NULL);
1920 
1921 	newfdp = fdinit(fdp, true);
1922 	/* copy all passable descriptors (i.e. not kqueue) */
1923 	newfdp->fd_freefile = -1;
1924 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1925 		ofde = &fdp->fd_ofiles[i];
1926 		if (ofde->fde_file == NULL ||
1927 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1928 			if (newfdp->fd_freefile == -1)
1929 				newfdp->fd_freefile = i;
1930 			continue;
1931 		}
1932 		nfde = &newfdp->fd_ofiles[i];
1933 		*nfde = *ofde;
1934 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
1935 		fhold(nfde->fde_file);
1936 		fdused_init(newfdp, i);
1937 		newfdp->fd_lastfile = i;
1938 	}
1939 	if (newfdp->fd_freefile == -1)
1940 		newfdp->fd_freefile = i;
1941 	newfdp->fd_cmask = fdp->fd_cmask;
1942 	FILEDESC_SUNLOCK(fdp);
1943 	return (newfdp);
1944 }
1945 
1946 /*
1947  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
1948  * one of processes using it exits) and the table used to be shared.
1949  */
1950 static void
1951 fdclearlocks(struct thread *td)
1952 {
1953 	struct filedesc *fdp;
1954 	struct filedesc_to_leader *fdtol;
1955 	struct flock lf;
1956 	struct file *fp;
1957 	struct proc *p;
1958 	struct vnode *vp;
1959 	int i;
1960 
1961 	p = td->td_proc;
1962 	fdp = p->p_fd;
1963 	fdtol = p->p_fdtol;
1964 	MPASS(fdtol != NULL);
1965 
1966 	FILEDESC_XLOCK(fdp);
1967 	KASSERT(fdtol->fdl_refcount > 0,
1968 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
1969 	    fdtol->fdl_refcount));
1970 	if (fdtol->fdl_refcount == 1 &&
1971 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1972 		for (i = 0; i <= fdp->fd_lastfile; i++) {
1973 			fp = fdp->fd_ofiles[i].fde_file;
1974 			if (fp == NULL || fp->f_type != DTYPE_VNODE)
1975 				continue;
1976 			fhold(fp);
1977 			FILEDESC_XUNLOCK(fdp);
1978 			lf.l_whence = SEEK_SET;
1979 			lf.l_start = 0;
1980 			lf.l_len = 0;
1981 			lf.l_type = F_UNLCK;
1982 			vp = fp->f_vnode;
1983 			(void) VOP_ADVLOCK(vp,
1984 			    (caddr_t)p->p_leader, F_UNLCK,
1985 			    &lf, F_POSIX);
1986 			FILEDESC_XLOCK(fdp);
1987 			fdrop(fp, td);
1988 		}
1989 	}
1990 retry:
1991 	if (fdtol->fdl_refcount == 1) {
1992 		if (fdp->fd_holdleaderscount > 0 &&
1993 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1994 			/*
1995 			 * close() or do_dup() has cleared a reference
1996 			 * in a shared file descriptor table.
1997 			 */
1998 			fdp->fd_holdleaderswakeup = 1;
1999 			sx_sleep(&fdp->fd_holdleaderscount,
2000 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2001 			goto retry;
2002 		}
2003 		if (fdtol->fdl_holdcount > 0) {
2004 			/*
2005 			 * Ensure that fdtol->fdl_leader remains
2006 			 * valid in closef().
2007 			 */
2008 			fdtol->fdl_wakeup = 1;
2009 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2010 			    "fdlhold", 0);
2011 			goto retry;
2012 		}
2013 	}
2014 	fdtol->fdl_refcount--;
2015 	if (fdtol->fdl_refcount == 0 &&
2016 	    fdtol->fdl_holdcount == 0) {
2017 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2018 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2019 	} else
2020 		fdtol = NULL;
2021 	p->p_fdtol = NULL;
2022 	FILEDESC_XUNLOCK(fdp);
2023 	if (fdtol != NULL)
2024 		free(fdtol, M_FILEDESC_TO_LEADER);
2025 }
2026 
2027 /*
2028  * Release a filedesc structure.
2029  */
2030 void
2031 fdescfree(struct thread *td)
2032 {
2033 	struct filedesc0 *fdp0;
2034 	struct filedesc *fdp;
2035 	struct freetable *ft, *tft;
2036 	struct filedescent *fde;
2037 	struct file *fp;
2038 	struct vnode *cdir, *jdir, *rdir;
2039 	int i;
2040 
2041 	fdp = td->td_proc->p_fd;
2042 	MPASS(fdp != NULL);
2043 
2044 #ifdef RACCT
2045 	PROC_LOCK(td->td_proc);
2046 	racct_set(td->td_proc, RACCT_NOFILE, 0);
2047 	PROC_UNLOCK(td->td_proc);
2048 #endif
2049 
2050 	if (td->td_proc->p_fdtol != NULL)
2051 		fdclearlocks(td);
2052 
2053 	mtx_lock(&fdesc_mtx);
2054 	td->td_proc->p_fd = NULL;
2055 	mtx_unlock(&fdesc_mtx);
2056 
2057 	FILEDESC_XLOCK(fdp);
2058 	i = --fdp->fd_refcnt;
2059 	if (i > 0) {
2060 		FILEDESC_XUNLOCK(fdp);
2061 		return;
2062 	}
2063 
2064 	cdir = fdp->fd_cdir;
2065 	fdp->fd_cdir = NULL;
2066 	rdir = fdp->fd_rdir;
2067 	fdp->fd_rdir = NULL;
2068 	jdir = fdp->fd_jdir;
2069 	fdp->fd_jdir = NULL;
2070 	FILEDESC_XUNLOCK(fdp);
2071 
2072 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2073 		fde = &fdp->fd_ofiles[i];
2074 		fp = fde->fde_file;
2075 		if (fp != NULL) {
2076 			fdefree_last(fde);
2077 			(void) closef(fp, td);
2078 		}
2079 	}
2080 
2081 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2082 		free(fdp->fd_map, M_FILEDESC);
2083 	if (fdp->fd_nfiles > NDFILE)
2084 		free(fdp->fd_files, M_FILEDESC);
2085 
2086 	fdp0 = (struct filedesc0 *)fdp;
2087 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2088 		free(ft->ft_table, M_FILEDESC);
2089 
2090 	if (cdir != NULL)
2091 		vrele(cdir);
2092 	if (rdir != NULL)
2093 		vrele(rdir);
2094 	if (jdir != NULL)
2095 		vrele(jdir);
2096 
2097 	fddrop(fdp);
2098 }
2099 
2100 /*
2101  * For setugid programs, we don't want to people to use that setugidness
2102  * to generate error messages which write to a file which otherwise would
2103  * otherwise be off-limits to the process.  We check for filesystems where
2104  * the vnode can change out from under us after execve (like [lin]procfs).
2105  *
2106  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2107  * sufficient.  We also don't check for setugidness since we know we are.
2108  */
2109 static bool
2110 is_unsafe(struct file *fp)
2111 {
2112 	struct vnode *vp;
2113 
2114 	if (fp->f_type != DTYPE_VNODE)
2115 		return (false);
2116 
2117 	vp = fp->f_vnode;
2118 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2119 }
2120 
2121 /*
2122  * Make this setguid thing safe, if at all possible.
2123  */
2124 void
2125 fdsetugidsafety(struct thread *td)
2126 {
2127 	struct filedesc *fdp;
2128 	struct file *fp;
2129 	int i;
2130 
2131 	fdp = td->td_proc->p_fd;
2132 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2133 	MPASS(fdp->fd_nfiles >= 3);
2134 	for (i = 0; i <= 2; i++) {
2135 		fp = fdp->fd_ofiles[i].fde_file;
2136 		if (fp != NULL && is_unsafe(fp)) {
2137 			FILEDESC_XLOCK(fdp);
2138 			knote_fdclose(td, i);
2139 			/*
2140 			 * NULL-out descriptor prior to close to avoid
2141 			 * a race while close blocks.
2142 			 */
2143 			fdfree(fdp, i);
2144 			FILEDESC_XUNLOCK(fdp);
2145 			(void) closef(fp, td);
2146 		}
2147 	}
2148 }
2149 
2150 /*
2151  * If a specific file object occupies a specific file descriptor, close the
2152  * file descriptor entry and drop a reference on the file object.  This is a
2153  * convenience function to handle a subsequent error in a function that calls
2154  * falloc() that handles the race that another thread might have closed the
2155  * file descriptor out from under the thread creating the file object.
2156  */
2157 void
2158 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2159 {
2160 
2161 	FILEDESC_XLOCK(fdp);
2162 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2163 		fdfree(fdp, idx);
2164 		FILEDESC_XUNLOCK(fdp);
2165 		fdrop(fp, td);
2166 	} else
2167 		FILEDESC_XUNLOCK(fdp);
2168 }
2169 
2170 /*
2171  * Close any files on exec?
2172  */
2173 void
2174 fdcloseexec(struct thread *td)
2175 {
2176 	struct filedesc *fdp;
2177 	struct filedescent *fde;
2178 	struct file *fp;
2179 	int i;
2180 
2181 	fdp = td->td_proc->p_fd;
2182 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2183 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2184 		fde = &fdp->fd_ofiles[i];
2185 		fp = fde->fde_file;
2186 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2187 		    (fde->fde_flags & UF_EXCLOSE))) {
2188 			FILEDESC_XLOCK(fdp);
2189 			fdfree(fdp, i);
2190 			(void) closefp(fdp, i, fp, td, 0);
2191 			/* closefp() drops the FILEDESC lock. */
2192 		}
2193 	}
2194 }
2195 
2196 /*
2197  * It is unsafe for set[ug]id processes to be started with file
2198  * descriptors 0..2 closed, as these descriptors are given implicit
2199  * significance in the Standard C library.  fdcheckstd() will create a
2200  * descriptor referencing /dev/null for each of stdin, stdout, and
2201  * stderr that is not already open.
2202  */
2203 int
2204 fdcheckstd(struct thread *td)
2205 {
2206 	struct filedesc *fdp;
2207 	register_t save;
2208 	int i, error, devnull;
2209 
2210 	fdp = td->td_proc->p_fd;
2211 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2212 	MPASS(fdp->fd_nfiles >= 3);
2213 	devnull = -1;
2214 	for (i = 0; i <= 2; i++) {
2215 		if (fdp->fd_ofiles[i].fde_file != NULL)
2216 			continue;
2217 
2218 		save = td->td_retval[0];
2219 		if (devnull != -1) {
2220 			error = do_dup(td, DUP_FIXED, devnull, i);
2221 		} else {
2222 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2223 			    UIO_SYSSPACE, O_RDWR, 0);
2224 			if (error == 0) {
2225 				devnull = td->td_retval[0];
2226 				KASSERT(devnull == i, ("we didn't get our fd"));
2227 			}
2228 		}
2229 		td->td_retval[0] = save;
2230 		if (error != 0)
2231 			return (error);
2232 	}
2233 	return (0);
2234 }
2235 
2236 /*
2237  * Internal form of close.  Decrement reference count on file structure.
2238  * Note: td may be NULL when closing a file that was being passed in a
2239  * message.
2240  *
2241  * XXXRW: Giant is not required for the caller, but often will be held; this
2242  * makes it moderately likely the Giant will be recursed in the VFS case.
2243  */
2244 int
2245 closef(struct file *fp, struct thread *td)
2246 {
2247 	struct vnode *vp;
2248 	struct flock lf;
2249 	struct filedesc_to_leader *fdtol;
2250 	struct filedesc *fdp;
2251 
2252 	/*
2253 	 * POSIX record locking dictates that any close releases ALL
2254 	 * locks owned by this process.  This is handled by setting
2255 	 * a flag in the unlock to free ONLY locks obeying POSIX
2256 	 * semantics, and not to free BSD-style file locks.
2257 	 * If the descriptor was in a message, POSIX-style locks
2258 	 * aren't passed with the descriptor, and the thread pointer
2259 	 * will be NULL.  Callers should be careful only to pass a
2260 	 * NULL thread pointer when there really is no owning
2261 	 * context that might have locks, or the locks will be
2262 	 * leaked.
2263 	 */
2264 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2265 		vp = fp->f_vnode;
2266 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2267 			lf.l_whence = SEEK_SET;
2268 			lf.l_start = 0;
2269 			lf.l_len = 0;
2270 			lf.l_type = F_UNLCK;
2271 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2272 			    F_UNLCK, &lf, F_POSIX);
2273 		}
2274 		fdtol = td->td_proc->p_fdtol;
2275 		if (fdtol != NULL) {
2276 			/*
2277 			 * Handle special case where file descriptor table is
2278 			 * shared between multiple process leaders.
2279 			 */
2280 			fdp = td->td_proc->p_fd;
2281 			FILEDESC_XLOCK(fdp);
2282 			for (fdtol = fdtol->fdl_next;
2283 			     fdtol != td->td_proc->p_fdtol;
2284 			     fdtol = fdtol->fdl_next) {
2285 				if ((fdtol->fdl_leader->p_flag &
2286 				     P_ADVLOCK) == 0)
2287 					continue;
2288 				fdtol->fdl_holdcount++;
2289 				FILEDESC_XUNLOCK(fdp);
2290 				lf.l_whence = SEEK_SET;
2291 				lf.l_start = 0;
2292 				lf.l_len = 0;
2293 				lf.l_type = F_UNLCK;
2294 				vp = fp->f_vnode;
2295 				(void) VOP_ADVLOCK(vp,
2296 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2297 				    F_POSIX);
2298 				FILEDESC_XLOCK(fdp);
2299 				fdtol->fdl_holdcount--;
2300 				if (fdtol->fdl_holdcount == 0 &&
2301 				    fdtol->fdl_wakeup != 0) {
2302 					fdtol->fdl_wakeup = 0;
2303 					wakeup(fdtol);
2304 				}
2305 			}
2306 			FILEDESC_XUNLOCK(fdp);
2307 		}
2308 	}
2309 	return (fdrop(fp, td));
2310 }
2311 
2312 /*
2313  * Initialize the file pointer with the specified properties.
2314  *
2315  * The ops are set with release semantics to be certain that the flags, type,
2316  * and data are visible when ops is.  This is to prevent ops methods from being
2317  * called with bad data.
2318  */
2319 void
2320 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2321 {
2322 	fp->f_data = data;
2323 	fp->f_flag = flag;
2324 	fp->f_type = type;
2325 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2326 }
2327 
2328 int
2329 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2330     int needfcntl, struct file **fpp, cap_rights_t *haverightsp)
2331 {
2332 #ifdef CAPABILITIES
2333 	struct filedescent fde;
2334 #endif
2335 	struct fdescenttbl *fdt;
2336 	struct file *fp;
2337 	u_int count;
2338 #ifdef CAPABILITIES
2339 	seq_t seq;
2340 	cap_rights_t haverights;
2341 	int error;
2342 #endif
2343 
2344 	fdt = fdp->fd_files;
2345 	if (fd < 0 || fd >= fdt->fdt_nfiles)
2346 		return (EBADF);
2347 	/*
2348 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2349 	 * never raising a refcount above 0.  To accomplish this we have
2350 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2351 	 * must be re-verified once we acquire a reference to be certain
2352 	 * that the identity is still correct and we did not lose a race
2353 	 * due to preemption.
2354 	 */
2355 	for (;;) {
2356 #ifdef CAPABILITIES
2357 		seq = seq_read(fd_seq(fdt, fd));
2358 		fde = fdt->fdt_ofiles[fd];
2359 		if (!seq_consistent(fd_seq(fdt, fd), seq)) {
2360 			cpu_spinwait();
2361 			continue;
2362 		}
2363 		fp = fde.fde_file;
2364 #else
2365 		fp = fdt->fdt_ofiles[fd].fde_file;
2366 #endif
2367 		if (fp == NULL)
2368 			return (EBADF);
2369 #ifdef CAPABILITIES
2370 		haverights = *cap_rights_fde(&fde);
2371 		if (needrightsp != NULL) {
2372 			error = cap_check(&haverights, needrightsp);
2373 			if (error != 0)
2374 				return (error);
2375 			if (cap_rights_is_set(needrightsp, CAP_FCNTL)) {
2376 				error = cap_fcntl_check_fde(&fde, needfcntl);
2377 				if (error != 0)
2378 					return (error);
2379 			}
2380 		}
2381 #endif
2382 	retry:
2383 		count = fp->f_count;
2384 		if (count == 0) {
2385 			/*
2386 			 * Force a reload. Other thread could reallocate the
2387 			 * table before this fd was closed, so it possible that
2388 			 * there is a stale fp pointer in cached version.
2389 			 */
2390 			fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2391 			continue;
2392 		}
2393 		/*
2394 		 * Use an acquire barrier to force re-reading of fdt so it is
2395 		 * refreshed for verification.
2396 		 */
2397 		if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) == 0)
2398 			goto retry;
2399 		fdt = fdp->fd_files;
2400 #ifdef	CAPABILITIES
2401 		if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2402 #else
2403 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2404 #endif
2405 			break;
2406 		fdrop(fp, curthread);
2407 	}
2408 	*fpp = fp;
2409 	if (haverightsp != NULL) {
2410 #ifdef CAPABILITIES
2411 		*haverightsp = haverights;
2412 #else
2413 		CAP_ALL(haverightsp);
2414 #endif
2415 	}
2416 	return (0);
2417 }
2418 
2419 /*
2420  * Extract the file pointer associated with the specified descriptor for the
2421  * current user process.
2422  *
2423  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2424  * returned.
2425  *
2426  * File's rights will be checked against the capability rights mask.
2427  *
2428  * If an error occured the non-zero error is returned and *fpp is set to
2429  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2430  * responsible for fdrop().
2431  */
2432 static __inline int
2433 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2434     cap_rights_t *needrightsp, u_char *maxprotp)
2435 {
2436 	struct filedesc *fdp;
2437 	struct file *fp;
2438 	cap_rights_t haverights, needrights;
2439 	int error;
2440 
2441 	*fpp = NULL;
2442 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2443 		return (EBADF);
2444 	if (needrightsp != NULL)
2445 		needrights = *needrightsp;
2446 	else
2447 		cap_rights_init(&needrights);
2448 	if (maxprotp != NULL)
2449 		cap_rights_set(&needrights, CAP_MMAP);
2450 	error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights);
2451 	if (error != 0)
2452 		return (error);
2453 	if (fp->f_ops == &badfileops) {
2454 		fdrop(fp, td);
2455 		return (EBADF);
2456 	}
2457 
2458 #ifdef CAPABILITIES
2459 	/*
2460 	 * If requested, convert capability rights to access flags.
2461 	 */
2462 	if (maxprotp != NULL)
2463 		*maxprotp = cap_rights_to_vmprot(&haverights);
2464 #else /* !CAPABILITIES */
2465 	if (maxprotp != NULL)
2466 		*maxprotp = VM_PROT_ALL;
2467 #endif /* CAPABILITIES */
2468 
2469 	/*
2470 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2471 	 */
2472 	error = 0;
2473 	switch (flags) {
2474 	case FREAD:
2475 	case FWRITE:
2476 		if ((fp->f_flag & flags) == 0)
2477 			error = EBADF;
2478 		break;
2479 	case FEXEC:
2480 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2481 		    ((fp->f_flag & FWRITE) != 0))
2482 			error = EBADF;
2483 		break;
2484 	case 0:
2485 		break;
2486 	default:
2487 		KASSERT(0, ("wrong flags"));
2488 	}
2489 
2490 	if (error != 0) {
2491 		fdrop(fp, td);
2492 		return (error);
2493 	}
2494 
2495 	*fpp = fp;
2496 	return (0);
2497 }
2498 
2499 int
2500 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2501 {
2502 
2503 	return(_fget(td, fd, fpp, 0, rightsp, NULL));
2504 }
2505 
2506 int
2507 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2508     struct file **fpp)
2509 {
2510 
2511 	return (_fget(td, fd, fpp, 0, rightsp, maxprotp));
2512 }
2513 
2514 int
2515 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2516 {
2517 
2518 	return(_fget(td, fd, fpp, FREAD, rightsp, NULL));
2519 }
2520 
2521 int
2522 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2523 {
2524 
2525 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2526 }
2527 
2528 /*
2529  * Like fget() but loads the underlying vnode, or returns an error if the
2530  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2531  * never have VM objects.  The returned vnode will be vref()'d.
2532  *
2533  * XXX: what about the unused flags ?
2534  */
2535 static __inline int
2536 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2537     struct vnode **vpp)
2538 {
2539 	struct file *fp;
2540 	int error;
2541 
2542 	*vpp = NULL;
2543 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2544 	if (error != 0)
2545 		return (error);
2546 	if (fp->f_vnode == NULL) {
2547 		error = EINVAL;
2548 	} else {
2549 		*vpp = fp->f_vnode;
2550 		vref(*vpp);
2551 	}
2552 	fdrop(fp, td);
2553 
2554 	return (error);
2555 }
2556 
2557 int
2558 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2559 {
2560 
2561 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2562 }
2563 
2564 int
2565 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2566     struct filecaps *havecaps, struct vnode **vpp)
2567 {
2568 	struct filedesc *fdp;
2569 	struct file *fp;
2570 #ifdef CAPABILITIES
2571 	int error;
2572 #endif
2573 
2574 	fdp = td->td_proc->p_fd;
2575 	fp = fget_locked(fdp, fd);
2576 	if (fp == NULL || fp->f_ops == &badfileops)
2577 		return (EBADF);
2578 
2579 #ifdef CAPABILITIES
2580 	if (needrightsp != NULL) {
2581 		error = cap_check(cap_rights(fdp, fd), needrightsp);
2582 		if (error != 0)
2583 			return (error);
2584 	}
2585 #endif
2586 
2587 	if (fp->f_vnode == NULL)
2588 		return (EINVAL);
2589 
2590 	*vpp = fp->f_vnode;
2591 	vref(*vpp);
2592 	filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
2593 
2594 	return (0);
2595 }
2596 
2597 int
2598 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2599 {
2600 
2601 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2602 }
2603 
2604 int
2605 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2606 {
2607 
2608 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2609 }
2610 
2611 #ifdef notyet
2612 int
2613 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2614     struct vnode **vpp)
2615 {
2616 
2617 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2618 }
2619 #endif
2620 
2621 /*
2622  * Like fget() but loads the underlying socket, or returns an error if the
2623  * descriptor does not represent a socket.
2624  *
2625  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
2626  * in the future.
2627  *
2628  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2629  * on their file descriptor reference to prevent the socket from being free'd
2630  * during use.
2631  */
2632 int
2633 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp,
2634     u_int *fflagp)
2635 {
2636 	struct file *fp;
2637 	int error;
2638 
2639 	*spp = NULL;
2640 	if (fflagp != NULL)
2641 		*fflagp = 0;
2642 	if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0)
2643 		return (error);
2644 	if (fp->f_type != DTYPE_SOCKET) {
2645 		error = ENOTSOCK;
2646 	} else {
2647 		*spp = fp->f_data;
2648 		if (fflagp)
2649 			*fflagp = fp->f_flag;
2650 		SOCK_LOCK(*spp);
2651 		soref(*spp);
2652 		SOCK_UNLOCK(*spp);
2653 	}
2654 	fdrop(fp, td);
2655 
2656 	return (error);
2657 }
2658 
2659 /*
2660  * Drop the reference count on the socket and XXX release the SX lock in the
2661  * future.  The last reference closes the socket.
2662  *
2663  * Note: fputsock() is deprecated, see comment for fgetsock().
2664  */
2665 void
2666 fputsock(struct socket *so)
2667 {
2668 
2669 	ACCEPT_LOCK();
2670 	SOCK_LOCK(so);
2671 	CURVNET_SET(so->so_vnet);
2672 	sorele(so);
2673 	CURVNET_RESTORE();
2674 }
2675 
2676 /*
2677  * Handle the last reference to a file being closed.
2678  */
2679 int
2680 _fdrop(struct file *fp, struct thread *td)
2681 {
2682 	int error;
2683 
2684 	error = 0;
2685 	if (fp->f_count != 0)
2686 		panic("fdrop: count %d", fp->f_count);
2687 	if (fp->f_ops != &badfileops)
2688 		error = fo_close(fp, td);
2689 	atomic_subtract_int(&openfiles, 1);
2690 	crfree(fp->f_cred);
2691 	free(fp->f_advice, M_FADVISE);
2692 	uma_zfree(file_zone, fp);
2693 
2694 	return (error);
2695 }
2696 
2697 /*
2698  * Apply an advisory lock on a file descriptor.
2699  *
2700  * Just attempt to get a record lock of the requested type on the entire file
2701  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2702  */
2703 #ifndef _SYS_SYSPROTO_H_
2704 struct flock_args {
2705 	int	fd;
2706 	int	how;
2707 };
2708 #endif
2709 /* ARGSUSED */
2710 int
2711 sys_flock(struct thread *td, struct flock_args *uap)
2712 {
2713 	struct file *fp;
2714 	struct vnode *vp;
2715 	struct flock lf;
2716 	cap_rights_t rights;
2717 	int error;
2718 
2719 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2720 	if (error != 0)
2721 		return (error);
2722 	if (fp->f_type != DTYPE_VNODE) {
2723 		fdrop(fp, td);
2724 		return (EOPNOTSUPP);
2725 	}
2726 
2727 	vp = fp->f_vnode;
2728 	lf.l_whence = SEEK_SET;
2729 	lf.l_start = 0;
2730 	lf.l_len = 0;
2731 	if (uap->how & LOCK_UN) {
2732 		lf.l_type = F_UNLCK;
2733 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2734 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2735 		goto done2;
2736 	}
2737 	if (uap->how & LOCK_EX)
2738 		lf.l_type = F_WRLCK;
2739 	else if (uap->how & LOCK_SH)
2740 		lf.l_type = F_RDLCK;
2741 	else {
2742 		error = EBADF;
2743 		goto done2;
2744 	}
2745 	atomic_set_int(&fp->f_flag, FHASLOCK);
2746 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2747 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2748 done2:
2749 	fdrop(fp, td);
2750 	return (error);
2751 }
2752 /*
2753  * Duplicate the specified descriptor to a free descriptor.
2754  */
2755 int
2756 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2757     int openerror, int *indxp)
2758 {
2759 	struct filedescent *newfde, *oldfde;
2760 	struct file *fp;
2761 	int error, indx;
2762 
2763 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2764 	    ("unexpected error %d in %s", openerror, __func__));
2765 
2766 	/*
2767 	 * If the to-be-dup'd fd number is greater than the allowed number
2768 	 * of file descriptors, or the fd to be dup'd has already been
2769 	 * closed, then reject.
2770 	 */
2771 	FILEDESC_XLOCK(fdp);
2772 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2773 		FILEDESC_XUNLOCK(fdp);
2774 		return (EBADF);
2775 	}
2776 
2777 	error = fdalloc(td, 0, &indx);
2778 	if (error != 0) {
2779 		FILEDESC_XUNLOCK(fdp);
2780 		return (error);
2781 	}
2782 
2783 	/*
2784 	 * There are two cases of interest here.
2785 	 *
2786 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2787 	 *
2788 	 * For ENXIO steal away the file structure from (dfd) and store it in
2789 	 * (indx).  (dfd) is effectively closed by this operation.
2790 	 */
2791 	switch (openerror) {
2792 	case ENODEV:
2793 		/*
2794 		 * Check that the mode the file is being opened for is a
2795 		 * subset of the mode of the existing descriptor.
2796 		 */
2797 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2798 			fdunused(fdp, indx);
2799 			FILEDESC_XUNLOCK(fdp);
2800 			return (EACCES);
2801 		}
2802 		fhold(fp);
2803 		newfde = &fdp->fd_ofiles[indx];
2804 		oldfde = &fdp->fd_ofiles[dfd];
2805 #ifdef CAPABILITIES
2806 		seq_write_begin(&newfde->fde_seq);
2807 #endif
2808 		memcpy(newfde, oldfde, fde_change_size);
2809 		filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
2810 #ifdef CAPABILITIES
2811 		seq_write_end(&newfde->fde_seq);
2812 #endif
2813 		break;
2814 	case ENXIO:
2815 		/*
2816 		 * Steal away the file pointer from dfd and stuff it into indx.
2817 		 */
2818 		newfde = &fdp->fd_ofiles[indx];
2819 		oldfde = &fdp->fd_ofiles[dfd];
2820 #ifdef CAPABILITIES
2821 		seq_write_begin(&newfde->fde_seq);
2822 #endif
2823 		memcpy(newfde, oldfde, fde_change_size);
2824 		bzero(oldfde, fde_change_size);
2825 		fdunused(fdp, dfd);
2826 #ifdef CAPABILITIES
2827 		seq_write_end(&newfde->fde_seq);
2828 #endif
2829 		break;
2830 	}
2831 	FILEDESC_XUNLOCK(fdp);
2832 	*indxp = indx;
2833 	return (0);
2834 }
2835 
2836 /*
2837  * Scan all active processes and prisons to see if any of them have a current
2838  * or root directory of `olddp'. If so, replace them with the new mount point.
2839  */
2840 void
2841 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2842 {
2843 	struct filedesc *fdp;
2844 	struct prison *pr;
2845 	struct proc *p;
2846 	int nrele;
2847 
2848 	if (vrefcnt(olddp) == 1)
2849 		return;
2850 	nrele = 0;
2851 	sx_slock(&allproc_lock);
2852 	FOREACH_PROC_IN_SYSTEM(p) {
2853 		fdp = fdhold(p);
2854 		if (fdp == NULL)
2855 			continue;
2856 		FILEDESC_XLOCK(fdp);
2857 		if (fdp->fd_cdir == olddp) {
2858 			vref(newdp);
2859 			fdp->fd_cdir = newdp;
2860 			nrele++;
2861 		}
2862 		if (fdp->fd_rdir == olddp) {
2863 			vref(newdp);
2864 			fdp->fd_rdir = newdp;
2865 			nrele++;
2866 		}
2867 		if (fdp->fd_jdir == olddp) {
2868 			vref(newdp);
2869 			fdp->fd_jdir = newdp;
2870 			nrele++;
2871 		}
2872 		FILEDESC_XUNLOCK(fdp);
2873 		fddrop(fdp);
2874 	}
2875 	sx_sunlock(&allproc_lock);
2876 	if (rootvnode == olddp) {
2877 		vref(newdp);
2878 		rootvnode = newdp;
2879 		nrele++;
2880 	}
2881 	mtx_lock(&prison0.pr_mtx);
2882 	if (prison0.pr_root == olddp) {
2883 		vref(newdp);
2884 		prison0.pr_root = newdp;
2885 		nrele++;
2886 	}
2887 	mtx_unlock(&prison0.pr_mtx);
2888 	sx_slock(&allprison_lock);
2889 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2890 		mtx_lock(&pr->pr_mtx);
2891 		if (pr->pr_root == olddp) {
2892 			vref(newdp);
2893 			pr->pr_root = newdp;
2894 			nrele++;
2895 		}
2896 		mtx_unlock(&pr->pr_mtx);
2897 	}
2898 	sx_sunlock(&allprison_lock);
2899 	while (nrele--)
2900 		vrele(olddp);
2901 }
2902 
2903 struct filedesc_to_leader *
2904 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2905 {
2906 	struct filedesc_to_leader *fdtol;
2907 
2908 	fdtol = malloc(sizeof(struct filedesc_to_leader),
2909 	       M_FILEDESC_TO_LEADER,
2910 	       M_WAITOK);
2911 	fdtol->fdl_refcount = 1;
2912 	fdtol->fdl_holdcount = 0;
2913 	fdtol->fdl_wakeup = 0;
2914 	fdtol->fdl_leader = leader;
2915 	if (old != NULL) {
2916 		FILEDESC_XLOCK(fdp);
2917 		fdtol->fdl_next = old->fdl_next;
2918 		fdtol->fdl_prev = old;
2919 		old->fdl_next = fdtol;
2920 		fdtol->fdl_next->fdl_prev = fdtol;
2921 		FILEDESC_XUNLOCK(fdp);
2922 	} else {
2923 		fdtol->fdl_next = fdtol;
2924 		fdtol->fdl_prev = fdtol;
2925 	}
2926 	return (fdtol);
2927 }
2928 
2929 /*
2930  * Get file structures globally.
2931  */
2932 static int
2933 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2934 {
2935 	struct xfile xf;
2936 	struct filedesc *fdp;
2937 	struct file *fp;
2938 	struct proc *p;
2939 	int error, n;
2940 
2941 	error = sysctl_wire_old_buffer(req, 0);
2942 	if (error != 0)
2943 		return (error);
2944 	if (req->oldptr == NULL) {
2945 		n = 0;
2946 		sx_slock(&allproc_lock);
2947 		FOREACH_PROC_IN_SYSTEM(p) {
2948 			if (p->p_state == PRS_NEW)
2949 				continue;
2950 			fdp = fdhold(p);
2951 			if (fdp == NULL)
2952 				continue;
2953 			/* overestimates sparse tables. */
2954 			if (fdp->fd_lastfile > 0)
2955 				n += fdp->fd_lastfile;
2956 			fddrop(fdp);
2957 		}
2958 		sx_sunlock(&allproc_lock);
2959 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2960 	}
2961 	error = 0;
2962 	bzero(&xf, sizeof(xf));
2963 	xf.xf_size = sizeof(xf);
2964 	sx_slock(&allproc_lock);
2965 	FOREACH_PROC_IN_SYSTEM(p) {
2966 		PROC_LOCK(p);
2967 		if (p->p_state == PRS_NEW) {
2968 			PROC_UNLOCK(p);
2969 			continue;
2970 		}
2971 		if (p_cansee(req->td, p) != 0) {
2972 			PROC_UNLOCK(p);
2973 			continue;
2974 		}
2975 		xf.xf_pid = p->p_pid;
2976 		xf.xf_uid = p->p_ucred->cr_uid;
2977 		PROC_UNLOCK(p);
2978 		fdp = fdhold(p);
2979 		if (fdp == NULL)
2980 			continue;
2981 		FILEDESC_SLOCK(fdp);
2982 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
2983 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
2984 				continue;
2985 			xf.xf_fd = n;
2986 			xf.xf_file = fp;
2987 			xf.xf_data = fp->f_data;
2988 			xf.xf_vnode = fp->f_vnode;
2989 			xf.xf_type = fp->f_type;
2990 			xf.xf_count = fp->f_count;
2991 			xf.xf_msgcount = 0;
2992 			xf.xf_offset = foffset_get(fp);
2993 			xf.xf_flag = fp->f_flag;
2994 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2995 			if (error)
2996 				break;
2997 		}
2998 		FILEDESC_SUNLOCK(fdp);
2999 		fddrop(fdp);
3000 		if (error)
3001 			break;
3002 	}
3003 	sx_sunlock(&allproc_lock);
3004 	return (error);
3005 }
3006 
3007 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3008     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3009 
3010 #ifdef KINFO_FILE_SIZE
3011 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3012 #endif
3013 
3014 static int
3015 xlate_fflags(int fflags)
3016 {
3017 	static const struct {
3018 		int	fflag;
3019 		int	kf_fflag;
3020 	} fflags_table[] = {
3021 		{ FAPPEND, KF_FLAG_APPEND },
3022 		{ FASYNC, KF_FLAG_ASYNC },
3023 		{ FFSYNC, KF_FLAG_FSYNC },
3024 		{ FHASLOCK, KF_FLAG_HASLOCK },
3025 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3026 		{ FREAD, KF_FLAG_READ },
3027 		{ FWRITE, KF_FLAG_WRITE },
3028 		{ O_CREAT, KF_FLAG_CREAT },
3029 		{ O_DIRECT, KF_FLAG_DIRECT },
3030 		{ O_EXCL, KF_FLAG_EXCL },
3031 		{ O_EXEC, KF_FLAG_EXEC },
3032 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3033 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3034 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3035 		{ O_TRUNC, KF_FLAG_TRUNC }
3036 	};
3037 	unsigned int i;
3038 	int kflags;
3039 
3040 	kflags = 0;
3041 	for (i = 0; i < nitems(fflags_table); i++)
3042 		if (fflags & fflags_table[i].fflag)
3043 			kflags |=  fflags_table[i].kf_fflag;
3044 	return (kflags);
3045 }
3046 
3047 /* Trim unused data from kf_path by truncating the structure size. */
3048 static void
3049 pack_kinfo(struct kinfo_file *kif)
3050 {
3051 
3052 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3053 	    strlen(kif->kf_path) + 1;
3054 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3055 }
3056 
3057 static void
3058 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3059     struct kinfo_file *kif, struct filedesc *fdp)
3060 {
3061 	int error;
3062 
3063 	bzero(kif, sizeof(*kif));
3064 
3065 	/* Set a default type to allow for empty fill_kinfo() methods. */
3066 	kif->kf_type = KF_TYPE_UNKNOWN;
3067 	kif->kf_flags = xlate_fflags(fp->f_flag);
3068 	if (rightsp != NULL)
3069 		kif->kf_cap_rights = *rightsp;
3070 	else
3071 		cap_rights_init(&kif->kf_cap_rights);
3072 	kif->kf_fd = fd;
3073 	kif->kf_ref_count = fp->f_count;
3074 	kif->kf_offset = foffset_get(fp);
3075 
3076 	/*
3077 	 * This may drop the filedesc lock, so the 'fp' cannot be
3078 	 * accessed after this call.
3079 	 */
3080 	error = fo_fill_kinfo(fp, kif, fdp);
3081 	if (error == 0)
3082 		kif->kf_status |= KF_ATTR_VALID;
3083 	pack_kinfo(kif);
3084 }
3085 
3086 static void
3087 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3088     struct kinfo_file *kif)
3089 {
3090 	int error;
3091 
3092 	bzero(kif, sizeof(*kif));
3093 
3094 	kif->kf_type = KF_TYPE_VNODE;
3095 	error = vn_fill_kinfo_vnode(vp, kif);
3096 	if (error == 0)
3097 		kif->kf_status |= KF_ATTR_VALID;
3098 	kif->kf_flags = xlate_fflags(fflags);
3099 	cap_rights_init(&kif->kf_cap_rights);
3100 	kif->kf_fd = fd;
3101 	kif->kf_ref_count = -1;
3102 	kif->kf_offset = -1;
3103 	pack_kinfo(kif);
3104 	vrele(vp);
3105 }
3106 
3107 struct export_fd_buf {
3108 	struct filedesc		*fdp;
3109 	struct sbuf 		*sb;
3110 	ssize_t			remainder;
3111 	struct kinfo_file	kif;
3112 };
3113 
3114 static int
3115 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3116 {
3117 	struct kinfo_file *kif;
3118 
3119 	kif = &efbuf->kif;
3120 	if (efbuf->remainder != -1) {
3121 		if (efbuf->remainder < kif->kf_structsize) {
3122 			/* Terminate export. */
3123 			efbuf->remainder = 0;
3124 			return (0);
3125 		}
3126 		efbuf->remainder -= kif->kf_structsize;
3127 	}
3128 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3129 }
3130 
3131 static int
3132 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3133     struct export_fd_buf *efbuf)
3134 {
3135 	int error;
3136 
3137 	if (efbuf->remainder == 0)
3138 		return (0);
3139 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp);
3140 	FILEDESC_SUNLOCK(efbuf->fdp);
3141 	error = export_kinfo_to_sb(efbuf);
3142 	FILEDESC_SLOCK(efbuf->fdp);
3143 	return (error);
3144 }
3145 
3146 static int
3147 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3148     struct export_fd_buf *efbuf)
3149 {
3150 	int error;
3151 
3152 	if (efbuf->remainder == 0)
3153 		return (0);
3154 	if (efbuf->fdp != NULL)
3155 		FILEDESC_SUNLOCK(efbuf->fdp);
3156 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif);
3157 	error = export_kinfo_to_sb(efbuf);
3158 	if (efbuf->fdp != NULL)
3159 		FILEDESC_SLOCK(efbuf->fdp);
3160 	return (error);
3161 }
3162 
3163 /*
3164  * Store a process file descriptor information to sbuf.
3165  *
3166  * Takes a locked proc as argument, and returns with the proc unlocked.
3167  */
3168 int
3169 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3170 {
3171 	struct file *fp;
3172 	struct filedesc *fdp;
3173 	struct export_fd_buf *efbuf;
3174 	struct vnode *cttyvp, *textvp, *tracevp;
3175 	int error, i;
3176 	cap_rights_t rights;
3177 
3178 	PROC_LOCK_ASSERT(p, MA_OWNED);
3179 
3180 	/* ktrace vnode */
3181 	tracevp = p->p_tracevp;
3182 	if (tracevp != NULL)
3183 		vref(tracevp);
3184 	/* text vnode */
3185 	textvp = p->p_textvp;
3186 	if (textvp != NULL)
3187 		vref(textvp);
3188 	/* Controlling tty. */
3189 	cttyvp = NULL;
3190 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3191 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3192 		if (cttyvp != NULL)
3193 			vref(cttyvp);
3194 	}
3195 	fdp = fdhold(p);
3196 	PROC_UNLOCK(p);
3197 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3198 	efbuf->fdp = NULL;
3199 	efbuf->sb = sb;
3200 	efbuf->remainder = maxlen;
3201 	if (tracevp != NULL)
3202 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3203 		    efbuf);
3204 	if (textvp != NULL)
3205 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3206 	if (cttyvp != NULL)
3207 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3208 		    efbuf);
3209 	error = 0;
3210 	if (fdp == NULL)
3211 		goto fail;
3212 	efbuf->fdp = fdp;
3213 	FILEDESC_SLOCK(fdp);
3214 	/* working directory */
3215 	if (fdp->fd_cdir != NULL) {
3216 		vref(fdp->fd_cdir);
3217 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3218 	}
3219 	/* root directory */
3220 	if (fdp->fd_rdir != NULL) {
3221 		vref(fdp->fd_rdir);
3222 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3223 	}
3224 	/* jail directory */
3225 	if (fdp->fd_jdir != NULL) {
3226 		vref(fdp->fd_jdir);
3227 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3228 	}
3229 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3230 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3231 			continue;
3232 #ifdef CAPABILITIES
3233 		rights = *cap_rights(fdp, i);
3234 #else /* !CAPABILITIES */
3235 		cap_rights_init(&rights);
3236 #endif
3237 		/*
3238 		 * Create sysctl entry.  It is OK to drop the filedesc
3239 		 * lock inside of export_file_to_sb() as we will
3240 		 * re-validate and re-evaluate its properties when the
3241 		 * loop continues.
3242 		 */
3243 		error = export_file_to_sb(fp, i, &rights, efbuf);
3244 		if (error != 0 || efbuf->remainder == 0)
3245 			break;
3246 	}
3247 	FILEDESC_SUNLOCK(fdp);
3248 	fddrop(fdp);
3249 fail:
3250 	free(efbuf, M_TEMP);
3251 	return (error);
3252 }
3253 
3254 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3255 
3256 /*
3257  * Get per-process file descriptors for use by procstat(1), et al.
3258  */
3259 static int
3260 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3261 {
3262 	struct sbuf sb;
3263 	struct proc *p;
3264 	ssize_t maxlen;
3265 	int error, error2, *name;
3266 
3267 	name = (int *)arg1;
3268 
3269 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3270 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3271 	if (error != 0) {
3272 		sbuf_delete(&sb);
3273 		return (error);
3274 	}
3275 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3276 	error = kern_proc_filedesc_out(p, &sb, maxlen);
3277 	error2 = sbuf_finish(&sb);
3278 	sbuf_delete(&sb);
3279 	return (error != 0 ? error : error2);
3280 }
3281 
3282 #ifdef KINFO_OFILE_SIZE
3283 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3284 #endif
3285 
3286 #ifdef COMPAT_FREEBSD7
3287 static void
3288 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3289 {
3290 
3291 	okif->kf_structsize = sizeof(*okif);
3292 	okif->kf_type = kif->kf_type;
3293 	okif->kf_fd = kif->kf_fd;
3294 	okif->kf_ref_count = kif->kf_ref_count;
3295 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3296 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3297 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3298 	okif->kf_offset = kif->kf_offset;
3299 	okif->kf_vnode_type = kif->kf_vnode_type;
3300 	okif->kf_sock_domain = kif->kf_sock_domain;
3301 	okif->kf_sock_type = kif->kf_sock_type;
3302 	okif->kf_sock_protocol = kif->kf_sock_protocol;
3303 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3304 	okif->kf_sa_local = kif->kf_sa_local;
3305 	okif->kf_sa_peer = kif->kf_sa_peer;
3306 }
3307 
3308 static int
3309 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3310     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3311 {
3312 	int error;
3313 
3314 	vref(vp);
3315 	FILEDESC_SUNLOCK(fdp);
3316 	export_vnode_to_kinfo(vp, type, 0, kif);
3317 	kinfo_to_okinfo(kif, okif);
3318 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3319 	FILEDESC_SLOCK(fdp);
3320 	return (error);
3321 }
3322 
3323 /*
3324  * Get per-process file descriptors for use by procstat(1), et al.
3325  */
3326 static int
3327 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3328 {
3329 	struct kinfo_ofile *okif;
3330 	struct kinfo_file *kif;
3331 	struct filedesc *fdp;
3332 	int error, i, *name;
3333 	struct file *fp;
3334 	struct proc *p;
3335 
3336 	name = (int *)arg1;
3337 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3338 	if (error != 0)
3339 		return (error);
3340 	fdp = fdhold(p);
3341 	PROC_UNLOCK(p);
3342 	if (fdp == NULL)
3343 		return (ENOENT);
3344 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3345 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3346 	FILEDESC_SLOCK(fdp);
3347 	if (fdp->fd_cdir != NULL)
3348 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3349 		    okif, fdp, req);
3350 	if (fdp->fd_rdir != NULL)
3351 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3352 		    okif, fdp, req);
3353 	if (fdp->fd_jdir != NULL)
3354 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3355 		    okif, fdp, req);
3356 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3357 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3358 			continue;
3359 		export_file_to_kinfo(fp, i, NULL, kif, fdp);
3360 		FILEDESC_SUNLOCK(fdp);
3361 		kinfo_to_okinfo(kif, okif);
3362 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3363 		FILEDESC_SLOCK(fdp);
3364 		if (error)
3365 			break;
3366 	}
3367 	FILEDESC_SUNLOCK(fdp);
3368 	fddrop(fdp);
3369 	free(kif, M_TEMP);
3370 	free(okif, M_TEMP);
3371 	return (0);
3372 }
3373 
3374 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3375     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3376     "Process ofiledesc entries");
3377 #endif	/* COMPAT_FREEBSD7 */
3378 
3379 int
3380 vntype_to_kinfo(int vtype)
3381 {
3382 	struct {
3383 		int	vtype;
3384 		int	kf_vtype;
3385 	} vtypes_table[] = {
3386 		{ VBAD, KF_VTYPE_VBAD },
3387 		{ VBLK, KF_VTYPE_VBLK },
3388 		{ VCHR, KF_VTYPE_VCHR },
3389 		{ VDIR, KF_VTYPE_VDIR },
3390 		{ VFIFO, KF_VTYPE_VFIFO },
3391 		{ VLNK, KF_VTYPE_VLNK },
3392 		{ VNON, KF_VTYPE_VNON },
3393 		{ VREG, KF_VTYPE_VREG },
3394 		{ VSOCK, KF_VTYPE_VSOCK }
3395 	};
3396 	unsigned int i;
3397 
3398 	/*
3399 	 * Perform vtype translation.
3400 	 */
3401 	for (i = 0; i < nitems(vtypes_table); i++)
3402 		if (vtypes_table[i].vtype == vtype)
3403 			return (vtypes_table[i].kf_vtype);
3404 
3405 	return (KF_VTYPE_UNKNOWN);
3406 }
3407 
3408 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3409     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3410     "Process filedesc entries");
3411 
3412 /*
3413  * Store a process current working directory information to sbuf.
3414  *
3415  * Takes a locked proc as argument, and returns with the proc unlocked.
3416  */
3417 int
3418 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3419 {
3420 	struct filedesc *fdp;
3421 	struct export_fd_buf *efbuf;
3422 	int error;
3423 
3424 	PROC_LOCK_ASSERT(p, MA_OWNED);
3425 
3426 	fdp = fdhold(p);
3427 	PROC_UNLOCK(p);
3428 	if (fdp == NULL)
3429 		return (EINVAL);
3430 
3431 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3432 	efbuf->fdp = fdp;
3433 	efbuf->sb = sb;
3434 	efbuf->remainder = maxlen;
3435 
3436 	FILEDESC_SLOCK(fdp);
3437 	if (fdp->fd_cdir == NULL)
3438 		error = EINVAL;
3439 	else {
3440 		vref(fdp->fd_cdir);
3441 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3442 		    FREAD, efbuf);
3443 	}
3444 	FILEDESC_SUNLOCK(fdp);
3445 	fddrop(fdp);
3446 	free(efbuf, M_TEMP);
3447 	return (error);
3448 }
3449 
3450 /*
3451  * Get per-process current working directory.
3452  */
3453 static int
3454 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3455 {
3456 	struct sbuf sb;
3457 	struct proc *p;
3458 	ssize_t maxlen;
3459 	int error, error2, *name;
3460 
3461 	name = (int *)arg1;
3462 
3463 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3464 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3465 	if (error != 0) {
3466 		sbuf_delete(&sb);
3467 		return (error);
3468 	}
3469 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3470 	error = kern_proc_cwd_out(p, &sb, maxlen);
3471 	error2 = sbuf_finish(&sb);
3472 	sbuf_delete(&sb);
3473 	return (error != 0 ? error : error2);
3474 }
3475 
3476 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3477     sysctl_kern_proc_cwd, "Process current working directory");
3478 
3479 #ifdef DDB
3480 /*
3481  * For the purposes of debugging, generate a human-readable string for the
3482  * file type.
3483  */
3484 static const char *
3485 file_type_to_name(short type)
3486 {
3487 
3488 	switch (type) {
3489 	case 0:
3490 		return ("zero");
3491 	case DTYPE_VNODE:
3492 		return ("vnod");
3493 	case DTYPE_SOCKET:
3494 		return ("sock");
3495 	case DTYPE_PIPE:
3496 		return ("pipe");
3497 	case DTYPE_FIFO:
3498 		return ("fifo");
3499 	case DTYPE_KQUEUE:
3500 		return ("kque");
3501 	case DTYPE_CRYPTO:
3502 		return ("crpt");
3503 	case DTYPE_MQUEUE:
3504 		return ("mque");
3505 	case DTYPE_SHM:
3506 		return ("shm");
3507 	case DTYPE_SEM:
3508 		return ("ksem");
3509 	default:
3510 		return ("unkn");
3511 	}
3512 }
3513 
3514 /*
3515  * For the purposes of debugging, identify a process (if any, perhaps one of
3516  * many) that references the passed file in its file descriptor array. Return
3517  * NULL if none.
3518  */
3519 static struct proc *
3520 file_to_first_proc(struct file *fp)
3521 {
3522 	struct filedesc *fdp;
3523 	struct proc *p;
3524 	int n;
3525 
3526 	FOREACH_PROC_IN_SYSTEM(p) {
3527 		if (p->p_state == PRS_NEW)
3528 			continue;
3529 		fdp = p->p_fd;
3530 		if (fdp == NULL)
3531 			continue;
3532 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3533 			if (fp == fdp->fd_ofiles[n].fde_file)
3534 				return (p);
3535 		}
3536 	}
3537 	return (NULL);
3538 }
3539 
3540 static void
3541 db_print_file(struct file *fp, int header)
3542 {
3543 	struct proc *p;
3544 
3545 	if (header)
3546 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3547 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3548 		    "MCount", "Vnode", "FPID", "FCmd");
3549 	p = file_to_first_proc(fp);
3550 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3551 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3552 	    0, fp->f_count, 0, fp->f_vnode,
3553 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3554 }
3555 
3556 DB_SHOW_COMMAND(file, db_show_file)
3557 {
3558 	struct file *fp;
3559 
3560 	if (!have_addr) {
3561 		db_printf("usage: show file <addr>\n");
3562 		return;
3563 	}
3564 	fp = (struct file *)addr;
3565 	db_print_file(fp, 1);
3566 }
3567 
3568 DB_SHOW_COMMAND(files, db_show_files)
3569 {
3570 	struct filedesc *fdp;
3571 	struct file *fp;
3572 	struct proc *p;
3573 	int header;
3574 	int n;
3575 
3576 	header = 1;
3577 	FOREACH_PROC_IN_SYSTEM(p) {
3578 		if (p->p_state == PRS_NEW)
3579 			continue;
3580 		if ((fdp = p->p_fd) == NULL)
3581 			continue;
3582 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
3583 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3584 				continue;
3585 			db_print_file(fp, header);
3586 			header = 0;
3587 		}
3588 	}
3589 }
3590 #endif
3591 
3592 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3593     &maxfilesperproc, 0, "Maximum files allowed open per process");
3594 
3595 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3596     &maxfiles, 0, "Maximum number of files");
3597 
3598 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3599     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3600 
3601 /* ARGSUSED*/
3602 static void
3603 filelistinit(void *dummy)
3604 {
3605 
3606 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3607 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3608 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3609 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3610 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3611 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3612 }
3613 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3614 
3615 /*-------------------------------------------------------------------*/
3616 
3617 static int
3618 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3619     int flags, struct thread *td)
3620 {
3621 
3622 	return (EBADF);
3623 }
3624 
3625 static int
3626 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3627     struct thread *td)
3628 {
3629 
3630 	return (EINVAL);
3631 }
3632 
3633 static int
3634 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3635     struct thread *td)
3636 {
3637 
3638 	return (EBADF);
3639 }
3640 
3641 static int
3642 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3643     struct thread *td)
3644 {
3645 
3646 	return (0);
3647 }
3648 
3649 static int
3650 badfo_kqfilter(struct file *fp, struct knote *kn)
3651 {
3652 
3653 	return (EBADF);
3654 }
3655 
3656 static int
3657 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3658     struct thread *td)
3659 {
3660 
3661 	return (EBADF);
3662 }
3663 
3664 static int
3665 badfo_close(struct file *fp, struct thread *td)
3666 {
3667 
3668 	return (EBADF);
3669 }
3670 
3671 static int
3672 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3673     struct thread *td)
3674 {
3675 
3676 	return (EBADF);
3677 }
3678 
3679 static int
3680 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3681     struct thread *td)
3682 {
3683 
3684 	return (EBADF);
3685 }
3686 
3687 static int
3688 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3689     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3690     int kflags, struct thread *td)
3691 {
3692 
3693 	return (EBADF);
3694 }
3695 
3696 static int
3697 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
3698 {
3699 
3700 	return (0);
3701 }
3702 
3703 struct fileops badfileops = {
3704 	.fo_read = badfo_readwrite,
3705 	.fo_write = badfo_readwrite,
3706 	.fo_truncate = badfo_truncate,
3707 	.fo_ioctl = badfo_ioctl,
3708 	.fo_poll = badfo_poll,
3709 	.fo_kqfilter = badfo_kqfilter,
3710 	.fo_stat = badfo_stat,
3711 	.fo_close = badfo_close,
3712 	.fo_chmod = badfo_chmod,
3713 	.fo_chown = badfo_chown,
3714 	.fo_sendfile = badfo_sendfile,
3715 	.fo_fill_kinfo = badfo_fill_kinfo,
3716 };
3717 
3718 int
3719 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
3720     int flags, struct thread *td)
3721 {
3722 
3723 	return (EOPNOTSUPP);
3724 }
3725 
3726 int
3727 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3728     struct thread *td)
3729 {
3730 
3731 	return (EINVAL);
3732 }
3733 
3734 int
3735 invfo_ioctl(struct file *fp, u_long com, void *data,
3736     struct ucred *active_cred, struct thread *td)
3737 {
3738 
3739 	return (ENOTTY);
3740 }
3741 
3742 int
3743 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
3744     struct thread *td)
3745 {
3746 
3747 	return (poll_no_poll(events));
3748 }
3749 
3750 int
3751 invfo_kqfilter(struct file *fp, struct knote *kn)
3752 {
3753 
3754 	return (EINVAL);
3755 }
3756 
3757 int
3758 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3759     struct thread *td)
3760 {
3761 
3762 	return (EINVAL);
3763 }
3764 
3765 int
3766 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3767     struct thread *td)
3768 {
3769 
3770 	return (EINVAL);
3771 }
3772 
3773 int
3774 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3775     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3776     int kflags, struct thread *td)
3777 {
3778 
3779 	return (EINVAL);
3780 }
3781 
3782 /*-------------------------------------------------------------------*/
3783 
3784 /*
3785  * File Descriptor pseudo-device driver (/dev/fd/).
3786  *
3787  * Opening minor device N dup()s the file (if any) connected to file
3788  * descriptor N belonging to the calling process.  Note that this driver
3789  * consists of only the ``open()'' routine, because all subsequent
3790  * references to this file will be direct to the other driver.
3791  *
3792  * XXX: we could give this one a cloning event handler if necessary.
3793  */
3794 
3795 /* ARGSUSED */
3796 static int
3797 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3798 {
3799 
3800 	/*
3801 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3802 	 * the file descriptor being sought for duplication. The error
3803 	 * return ensures that the vnode for this device will be released
3804 	 * by vn_open. Open will detect this special error and take the
3805 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3806 	 * will simply report the error.
3807 	 */
3808 	td->td_dupfd = dev2unit(dev);
3809 	return (ENODEV);
3810 }
3811 
3812 static struct cdevsw fildesc_cdevsw = {
3813 	.d_version =	D_VERSION,
3814 	.d_open =	fdopen,
3815 	.d_name =	"FD",
3816 };
3817 
3818 static void
3819 fildesc_drvinit(void *unused)
3820 {
3821 	struct cdev *dev;
3822 
3823 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
3824 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
3825 	make_dev_alias(dev, "stdin");
3826 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
3827 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
3828 	make_dev_alias(dev, "stdout");
3829 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
3830 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
3831 	make_dev_alias(dev, "stderr");
3832 }
3833 
3834 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
3835