xref: /netbsd-src/sys/kern/kern_descrip.c (revision b7b7574d3bf8eeb51a1fa3977b59142ec6434a55)
1 /*	$NetBSD: kern_descrip.c,v 1.224 2014/03/16 05:20:30 dholland Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1982, 1986, 1989, 1991, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  * (c) UNIX System Laboratories, Inc.
36  * All or some portions of this file are derived from material licensed
37  * to the University of California by American Telephone and Telegraph
38  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
39  * the permission of UNIX System Laboratories, Inc.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)kern_descrip.c	8.8 (Berkeley) 2/14/95
66  */
67 
68 /*
69  * File descriptor management.
70  */
71 
72 #include <sys/cdefs.h>
73 __KERNEL_RCSID(0, "$NetBSD: kern_descrip.c,v 1.224 2014/03/16 05:20:30 dholland Exp $");
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/filedesc.h>
78 #include <sys/kernel.h>
79 #include <sys/proc.h>
80 #include <sys/file.h>
81 #include <sys/socket.h>
82 #include <sys/socketvar.h>
83 #include <sys/stat.h>
84 #include <sys/ioctl.h>
85 #include <sys/fcntl.h>
86 #include <sys/pool.h>
87 #include <sys/unistd.h>
88 #include <sys/resourcevar.h>
89 #include <sys/conf.h>
90 #include <sys/event.h>
91 #include <sys/kauth.h>
92 #include <sys/atomic.h>
93 #include <sys/syscallargs.h>
94 #include <sys/cpu.h>
95 #include <sys/kmem.h>
96 #include <sys/vnode.h>
97 #include <sys/sysctl.h>
98 #include <sys/ktrace.h>
99 
100 /*
101  * A list (head) of open files, counter, and lock protecting them.
102  */
103 struct filelist		filehead	__cacheline_aligned;
104 static u_int		nfiles		__cacheline_aligned;
105 kmutex_t		filelist_lock	__cacheline_aligned;
106 
107 static pool_cache_t	filedesc_cache	__read_mostly;
108 static pool_cache_t	file_cache	__read_mostly;
109 static pool_cache_t	fdfile_cache	__read_mostly;
110 
111 static int	file_ctor(void *, void *, int);
112 static void	file_dtor(void *, void *);
113 static int	fdfile_ctor(void *, void *, int);
114 static void	fdfile_dtor(void *, void *);
115 static int	filedesc_ctor(void *, void *, int);
116 static void	filedesc_dtor(void *, void *);
117 static int	filedescopen(dev_t, int, int, lwp_t *);
118 
119 static int sysctl_kern_file(SYSCTLFN_PROTO);
120 static int sysctl_kern_file2(SYSCTLFN_PROTO);
121 static void fill_file(struct kinfo_file *, const file_t *, const fdfile_t *,
122 		      int, pid_t);
123 
124 const struct cdevsw filedesc_cdevsw = {
125 	.d_open = filedescopen,
126 	.d_close = noclose,
127 	.d_read = noread,
128 	.d_write = nowrite,
129 	.d_ioctl = noioctl,
130 	.d_stop = nostop,
131 	.d_tty = notty,
132 	.d_poll = nopoll,
133 	.d_mmap = nommap,
134 	.d_kqfilter = nokqfilter,
135 	.d_flag = D_OTHER | D_MPSAFE
136 };
137 
138 /* For ease of reading. */
139 __strong_alias(fd_putvnode,fd_putfile)
140 __strong_alias(fd_putsock,fd_putfile)
141 
142 /*
143  * Initialize the descriptor system.
144  */
145 void
146 fd_sys_init(void)
147 {
148 	static struct sysctllog *clog;
149 
150 	mutex_init(&filelist_lock, MUTEX_DEFAULT, IPL_NONE);
151 
152 	file_cache = pool_cache_init(sizeof(file_t), coherency_unit, 0,
153 	    0, "file", NULL, IPL_NONE, file_ctor, file_dtor, NULL);
154 	KASSERT(file_cache != NULL);
155 
156 	fdfile_cache = pool_cache_init(sizeof(fdfile_t), coherency_unit, 0,
157 	    PR_LARGECACHE, "fdfile", NULL, IPL_NONE, fdfile_ctor, fdfile_dtor,
158 	    NULL);
159 	KASSERT(fdfile_cache != NULL);
160 
161 	filedesc_cache = pool_cache_init(sizeof(filedesc_t), coherency_unit,
162 	    0, 0, "filedesc", NULL, IPL_NONE, filedesc_ctor, filedesc_dtor,
163 	    NULL);
164 	KASSERT(filedesc_cache != NULL);
165 
166 	sysctl_createv(&clog, 0, NULL, NULL,
167 		       CTLFLAG_PERMANENT,
168 		       CTLTYPE_STRUCT, "file",
169 		       SYSCTL_DESCR("System open file table"),
170 		       sysctl_kern_file, 0, NULL, 0,
171 		       CTL_KERN, KERN_FILE, CTL_EOL);
172 	sysctl_createv(&clog, 0, NULL, NULL,
173 		       CTLFLAG_PERMANENT,
174 		       CTLTYPE_STRUCT, "file2",
175 		       SYSCTL_DESCR("System open file table"),
176 		       sysctl_kern_file2, 0, NULL, 0,
177 		       CTL_KERN, KERN_FILE2, CTL_EOL);
178 }
179 
180 static bool
181 fd_isused(filedesc_t *fdp, unsigned fd)
182 {
183 	u_int off = fd >> NDENTRYSHIFT;
184 
185 	KASSERT(fd < fdp->fd_dt->dt_nfiles);
186 
187 	return (fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0;
188 }
189 
190 /*
191  * Verify that the bitmaps match the descriptor table.
192  */
193 static inline void
194 fd_checkmaps(filedesc_t *fdp)
195 {
196 #ifdef DEBUG
197 	fdtab_t *dt;
198 	u_int fd;
199 
200 	dt = fdp->fd_dt;
201 	if (fdp->fd_refcnt == -1) {
202 		/*
203 		 * fd_free tears down the table without maintaining its bitmap.
204 		 */
205 		return;
206 	}
207 	for (fd = 0; fd < dt->dt_nfiles; fd++) {
208 		if (fd < NDFDFILE) {
209 			KASSERT(dt->dt_ff[fd] ==
210 			    (fdfile_t *)fdp->fd_dfdfile[fd]);
211 		}
212 		if (dt->dt_ff[fd] == NULL) {
213 			KASSERT(!fd_isused(fdp, fd));
214 		} else if (dt->dt_ff[fd]->ff_file != NULL) {
215 			KASSERT(fd_isused(fdp, fd));
216 		}
217 	}
218 #endif
219 }
220 
221 static int
222 fd_next_zero(filedesc_t *fdp, uint32_t *bitmap, int want, u_int bits)
223 {
224 	int i, off, maxoff;
225 	uint32_t sub;
226 
227 	KASSERT(mutex_owned(&fdp->fd_lock));
228 
229 	fd_checkmaps(fdp);
230 
231 	if (want > bits)
232 		return -1;
233 
234 	off = want >> NDENTRYSHIFT;
235 	i = want & NDENTRYMASK;
236 	if (i) {
237 		sub = bitmap[off] | ((u_int)~0 >> (NDENTRIES - i));
238 		if (sub != ~0)
239 			goto found;
240 		off++;
241 	}
242 
243 	maxoff = NDLOSLOTS(bits);
244 	while (off < maxoff) {
245 		if ((sub = bitmap[off]) != ~0)
246 			goto found;
247 		off++;
248 	}
249 
250 	return -1;
251 
252  found:
253 	return (off << NDENTRYSHIFT) + ffs(~sub) - 1;
254 }
255 
256 static int
257 fd_last_set(filedesc_t *fd, int last)
258 {
259 	int off, i;
260 	fdfile_t **ff = fd->fd_dt->dt_ff;
261 	uint32_t *bitmap = fd->fd_lomap;
262 
263 	KASSERT(mutex_owned(&fd->fd_lock));
264 
265 	fd_checkmaps(fd);
266 
267 	off = (last - 1) >> NDENTRYSHIFT;
268 
269 	while (off >= 0 && !bitmap[off])
270 		off--;
271 
272 	if (off < 0)
273 		return -1;
274 
275 	i = ((off + 1) << NDENTRYSHIFT) - 1;
276 	if (i >= last)
277 		i = last - 1;
278 
279 	/* XXX should use bitmap */
280 	while (i > 0 && (ff[i] == NULL || !ff[i]->ff_allocated))
281 		i--;
282 
283 	return i;
284 }
285 
286 static inline void
287 fd_used(filedesc_t *fdp, unsigned fd)
288 {
289 	u_int off = fd >> NDENTRYSHIFT;
290 	fdfile_t *ff;
291 
292 	ff = fdp->fd_dt->dt_ff[fd];
293 
294 	KASSERT(mutex_owned(&fdp->fd_lock));
295 	KASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) == 0);
296 	KASSERT(ff != NULL);
297 	KASSERT(ff->ff_file == NULL);
298 	KASSERT(!ff->ff_allocated);
299 
300 	ff->ff_allocated = true;
301 	fdp->fd_lomap[off] |= 1 << (fd & NDENTRYMASK);
302 	if (__predict_false(fdp->fd_lomap[off] == ~0)) {
303 		KASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
304 		    (1 << (off & NDENTRYMASK))) == 0);
305 		fdp->fd_himap[off >> NDENTRYSHIFT] |= 1 << (off & NDENTRYMASK);
306 	}
307 
308 	if ((int)fd > fdp->fd_lastfile) {
309 		fdp->fd_lastfile = fd;
310 	}
311 
312 	fd_checkmaps(fdp);
313 }
314 
315 static inline void
316 fd_unused(filedesc_t *fdp, unsigned fd)
317 {
318 	u_int off = fd >> NDENTRYSHIFT;
319 	fdfile_t *ff;
320 
321 	ff = fdp->fd_dt->dt_ff[fd];
322 
323 	/*
324 	 * Don't assert the lock is held here, as we may be copying
325 	 * the table during exec() and it is not needed there.
326 	 * procfs and sysctl are locked out by proc::p_reflock.
327 	 *
328 	 * KASSERT(mutex_owned(&fdp->fd_lock));
329 	 */
330 	KASSERT(ff != NULL);
331 	KASSERT(ff->ff_file == NULL);
332 	KASSERT(ff->ff_allocated);
333 
334 	if (fd < fdp->fd_freefile) {
335 		fdp->fd_freefile = fd;
336 	}
337 
338 	if (fdp->fd_lomap[off] == ~0) {
339 		KASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
340 		    (1 << (off & NDENTRYMASK))) != 0);
341 		fdp->fd_himap[off >> NDENTRYSHIFT] &=
342 		    ~(1 << (off & NDENTRYMASK));
343 	}
344 	KASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0);
345 	fdp->fd_lomap[off] &= ~(1 << (fd & NDENTRYMASK));
346 	ff->ff_allocated = false;
347 
348 	KASSERT(fd <= fdp->fd_lastfile);
349 	if (fd == fdp->fd_lastfile) {
350 		fdp->fd_lastfile = fd_last_set(fdp, fd);
351 	}
352 	fd_checkmaps(fdp);
353 }
354 
355 /*
356  * Look up the file structure corresponding to a file descriptor
357  * and return the file, holding a reference on the descriptor.
358  */
359 file_t *
360 fd_getfile(unsigned fd)
361 {
362 	filedesc_t *fdp;
363 	fdfile_t *ff;
364 	file_t *fp;
365 	fdtab_t *dt;
366 
367 	/*
368 	 * Look up the fdfile structure representing this descriptor.
369 	 * We are doing this unlocked.  See fd_tryexpand().
370 	 */
371 	fdp = curlwp->l_fd;
372 	dt = fdp->fd_dt;
373 	if (__predict_false(fd >= dt->dt_nfiles)) {
374 		return NULL;
375 	}
376 	ff = dt->dt_ff[fd];
377 	KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
378 	if (__predict_false(ff == NULL)) {
379 		return NULL;
380 	}
381 
382 	/* Now get a reference to the descriptor. */
383 	if (fdp->fd_refcnt == 1) {
384 		/*
385 		 * Single threaded: don't need to worry about concurrent
386 		 * access (other than earlier calls to kqueue, which may
387 		 * hold a reference to the descriptor).
388 		 */
389 		ff->ff_refcnt++;
390 	} else {
391 		/*
392 		 * Multi threaded: issue a memory barrier to ensure that we
393 		 * acquire the file pointer _after_ adding a reference.  If
394 		 * no memory barrier, we could fetch a stale pointer.
395 		 */
396 		atomic_inc_uint(&ff->ff_refcnt);
397 #ifndef __HAVE_ATOMIC_AS_MEMBAR
398 		membar_enter();
399 #endif
400 	}
401 
402 	/*
403 	 * If the file is not open or is being closed then put the
404 	 * reference back.
405 	 */
406 	fp = ff->ff_file;
407 	if (__predict_true(fp != NULL)) {
408 		return fp;
409 	}
410 	fd_putfile(fd);
411 	return NULL;
412 }
413 
414 /*
415  * Release a reference to a file descriptor acquired with fd_getfile().
416  */
417 void
418 fd_putfile(unsigned fd)
419 {
420 	filedesc_t *fdp;
421 	fdfile_t *ff;
422 	u_int u, v;
423 
424 	fdp = curlwp->l_fd;
425 	ff = fdp->fd_dt->dt_ff[fd];
426 
427 	KASSERT(fd < fdp->fd_dt->dt_nfiles);
428 	KASSERT(ff != NULL);
429 	KASSERT((ff->ff_refcnt & FR_MASK) > 0);
430 	KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
431 
432 	if (fdp->fd_refcnt == 1) {
433 		/*
434 		 * Single threaded: don't need to worry about concurrent
435 		 * access (other than earlier calls to kqueue, which may
436 		 * hold a reference to the descriptor).
437 		 */
438 		if (__predict_false((ff->ff_refcnt & FR_CLOSING) != 0)) {
439 			fd_close(fd);
440 			return;
441 		}
442 		ff->ff_refcnt--;
443 		return;
444 	}
445 
446 	/*
447 	 * Ensure that any use of the file is complete and globally
448 	 * visible before dropping the final reference.  If no membar,
449 	 * the current CPU could still access memory associated with
450 	 * the file after it has been freed or recycled by another
451 	 * CPU.
452 	 */
453 #ifndef __HAVE_ATOMIC_AS_MEMBAR
454 	membar_exit();
455 #endif
456 
457 	/*
458 	 * Be optimistic and start out with the assumption that no other
459 	 * threads are trying to close the descriptor.  If the CAS fails,
460 	 * we lost a race and/or it's being closed.
461 	 */
462 	for (u = ff->ff_refcnt & FR_MASK;; u = v) {
463 		v = atomic_cas_uint(&ff->ff_refcnt, u, u - 1);
464 		if (__predict_true(u == v)) {
465 			return;
466 		}
467 		if (__predict_false((v & FR_CLOSING) != 0)) {
468 			break;
469 		}
470 	}
471 
472 	/* Another thread is waiting to close the file: join it. */
473 	(void)fd_close(fd);
474 }
475 
476 /*
477  * Convenience wrapper around fd_getfile() that returns reference
478  * to a vnode.
479  */
480 int
481 fd_getvnode(unsigned fd, file_t **fpp)
482 {
483 	vnode_t *vp;
484 	file_t *fp;
485 
486 	fp = fd_getfile(fd);
487 	if (__predict_false(fp == NULL)) {
488 		return EBADF;
489 	}
490 	if (__predict_false(fp->f_type != DTYPE_VNODE)) {
491 		fd_putfile(fd);
492 		return EINVAL;
493 	}
494 	vp = fp->f_data;
495 	if (__predict_false(vp->v_type == VBAD)) {
496 		/* XXX Is this case really necessary? */
497 		fd_putfile(fd);
498 		return EBADF;
499 	}
500 	*fpp = fp;
501 	return 0;
502 }
503 
504 /*
505  * Convenience wrapper around fd_getfile() that returns reference
506  * to a socket.
507  */
508 int
509 fd_getsock1(unsigned fd, struct socket **sop, file_t **fp)
510 {
511 	*fp = fd_getfile(fd);
512 	if (__predict_false(*fp == NULL)) {
513 		return EBADF;
514 	}
515 	if (__predict_false((*fp)->f_type != DTYPE_SOCKET)) {
516 		fd_putfile(fd);
517 		return ENOTSOCK;
518 	}
519 	*sop = (*fp)->f_data;
520 	return 0;
521 }
522 
523 int
524 fd_getsock(unsigned fd, struct socket **sop)
525 {
526 	file_t *fp;
527 	return fd_getsock1(fd, sop, &fp);
528 }
529 
530 /*
531  * Look up the file structure corresponding to a file descriptor
532  * and return it with a reference held on the file, not the
533  * descriptor.
534  *
535  * This is heavyweight and only used when accessing descriptors
536  * from a foreign process.  The caller must ensure that `p' does
537  * not exit or fork across this call.
538  *
539  * To release the file (not descriptor) reference, use closef().
540  */
541 file_t *
542 fd_getfile2(proc_t *p, unsigned fd)
543 {
544 	filedesc_t *fdp;
545 	fdfile_t *ff;
546 	file_t *fp;
547 	fdtab_t *dt;
548 
549 	fdp = p->p_fd;
550 	mutex_enter(&fdp->fd_lock);
551 	dt = fdp->fd_dt;
552 	if (fd >= dt->dt_nfiles) {
553 		mutex_exit(&fdp->fd_lock);
554 		return NULL;
555 	}
556 	if ((ff = dt->dt_ff[fd]) == NULL) {
557 		mutex_exit(&fdp->fd_lock);
558 		return NULL;
559 	}
560 	if ((fp = ff->ff_file) == NULL) {
561 		mutex_exit(&fdp->fd_lock);
562 		return NULL;
563 	}
564 	mutex_enter(&fp->f_lock);
565 	fp->f_count++;
566 	mutex_exit(&fp->f_lock);
567 	mutex_exit(&fdp->fd_lock);
568 
569 	return fp;
570 }
571 
572 /*
573  * Internal form of close.  Must be called with a reference to the
574  * descriptor, and will drop the reference.  When all descriptor
575  * references are dropped, releases the descriptor slot and a single
576  * reference to the file structure.
577  */
578 int
579 fd_close(unsigned fd)
580 {
581 	struct flock lf;
582 	filedesc_t *fdp;
583 	fdfile_t *ff;
584 	file_t *fp;
585 	proc_t *p;
586 	lwp_t *l;
587 	u_int refcnt;
588 
589 	l = curlwp;
590 	p = l->l_proc;
591 	fdp = l->l_fd;
592 	ff = fdp->fd_dt->dt_ff[fd];
593 
594 	KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
595 
596 	mutex_enter(&fdp->fd_lock);
597 	KASSERT((ff->ff_refcnt & FR_MASK) > 0);
598 	if (__predict_false(ff->ff_file == NULL)) {
599 		/*
600 		 * Another user of the file is already closing, and is
601 		 * waiting for other users of the file to drain.  Release
602 		 * our reference, and wake up the closer.
603 		 */
604 		atomic_dec_uint(&ff->ff_refcnt);
605 		cv_broadcast(&ff->ff_closing);
606 		mutex_exit(&fdp->fd_lock);
607 
608 		/*
609 		 * An application error, so pretend that the descriptor
610 		 * was already closed.  We can't safely wait for it to
611 		 * be closed without potentially deadlocking.
612 		 */
613 		return (EBADF);
614 	}
615 	KASSERT((ff->ff_refcnt & FR_CLOSING) == 0);
616 
617 	/*
618 	 * There may be multiple users of this file within the process.
619 	 * Notify existing and new users that the file is closing.  This
620 	 * will prevent them from adding additional uses to this file
621 	 * while we are closing it.
622 	 */
623 	fp = ff->ff_file;
624 	ff->ff_file = NULL;
625 	ff->ff_exclose = false;
626 
627 	/*
628 	 * We expect the caller to hold a descriptor reference - drop it.
629 	 * The reference count may increase beyond zero at this point due
630 	 * to an erroneous descriptor reference by an application, but
631 	 * fd_getfile() will notice that the file is being closed and drop
632 	 * the reference again.
633 	 */
634 	if (fdp->fd_refcnt == 1) {
635 		/* Single threaded. */
636 		refcnt = --(ff->ff_refcnt);
637 	} else {
638 		/* Multi threaded. */
639 #ifndef __HAVE_ATOMIC_AS_MEMBAR
640 		membar_producer();
641 #endif
642 		refcnt = atomic_dec_uint_nv(&ff->ff_refcnt);
643 	}
644 	if (__predict_false(refcnt != 0)) {
645 		/*
646 		 * Wait for other references to drain.  This is typically
647 		 * an application error - the descriptor is being closed
648 		 * while still in use.
649 		 * (Or just a threaded application trying to unblock its
650 		 * thread that sleeps in (say) accept()).
651 		 */
652 		atomic_or_uint(&ff->ff_refcnt, FR_CLOSING);
653 
654 		/*
655 		 * Remove any knotes attached to the file.  A knote
656 		 * attached to the descriptor can hold references on it.
657 		 */
658 		mutex_exit(&fdp->fd_lock);
659 		if (!SLIST_EMPTY(&ff->ff_knlist)) {
660 			knote_fdclose(fd);
661 		}
662 
663 		/*
664 		 * Since the file system code doesn't know which fd
665 		 * each request came from (think dup()), we have to
666 		 * ask it to return ERESTART for any long-term blocks.
667 		 * The re-entry through read/write/etc will detect the
668 		 * closed fd and return EBAFD.
669 		 * Blocked partial writes may return a short length.
670 		 */
671 		(*fp->f_ops->fo_restart)(fp);
672 		mutex_enter(&fdp->fd_lock);
673 
674 		/*
675 		 * We need to see the count drop to zero at least once,
676 		 * in order to ensure that all pre-existing references
677 		 * have been drained.  New references past this point are
678 		 * of no interest.
679 		 * XXX (dsl) this may need to call fo_restart() after a
680 		 * timeout to guarantee that all the system calls exit.
681 		 */
682 		while ((ff->ff_refcnt & FR_MASK) != 0) {
683 			cv_wait(&ff->ff_closing, &fdp->fd_lock);
684 		}
685 		atomic_and_uint(&ff->ff_refcnt, ~FR_CLOSING);
686 	} else {
687 		/* If no references, there must be no knotes. */
688 		KASSERT(SLIST_EMPTY(&ff->ff_knlist));
689 	}
690 
691 	/*
692 	 * POSIX record locking dictates that any close releases ALL
693 	 * locks owned by this process.  This is handled by setting
694 	 * a flag in the unlock to free ONLY locks obeying POSIX
695 	 * semantics, and not to free BSD-style file locks.
696 	 * If the descriptor was in a message, POSIX-style locks
697 	 * aren't passed with the descriptor.
698 	 */
699 	if (__predict_false((p->p_flag & PK_ADVLOCK) != 0 &&
700 	    fp->f_type == DTYPE_VNODE)) {
701 		lf.l_whence = SEEK_SET;
702 		lf.l_start = 0;
703 		lf.l_len = 0;
704 		lf.l_type = F_UNLCK;
705 		mutex_exit(&fdp->fd_lock);
706 		(void)VOP_ADVLOCK(fp->f_data, p, F_UNLCK, &lf, F_POSIX);
707 		mutex_enter(&fdp->fd_lock);
708 	}
709 
710 	/* Free descriptor slot. */
711 	fd_unused(fdp, fd);
712 	mutex_exit(&fdp->fd_lock);
713 
714 	/* Now drop reference to the file itself. */
715 	return closef(fp);
716 }
717 
718 /*
719  * Duplicate a file descriptor.
720  */
721 int
722 fd_dup(file_t *fp, int minfd, int *newp, bool exclose)
723 {
724 	proc_t *p = curproc;
725 	int error;
726 
727 	while ((error = fd_alloc(p, minfd, newp)) != 0) {
728 		if (error != ENOSPC) {
729 			return error;
730 		}
731 		fd_tryexpand(p);
732 	}
733 
734 	curlwp->l_fd->fd_dt->dt_ff[*newp]->ff_exclose = exclose;
735 	fd_affix(p, fp, *newp);
736 	return 0;
737 }
738 
739 /*
740  * dup2 operation.
741  */
742 int
743 fd_dup2(file_t *fp, unsigned new, int flags)
744 {
745 	filedesc_t *fdp = curlwp->l_fd;
746 	fdfile_t *ff;
747 	fdtab_t *dt;
748 
749 	if (flags & ~(O_CLOEXEC|O_NONBLOCK))
750 		return EINVAL;
751 	/*
752 	 * Ensure there are enough slots in the descriptor table,
753 	 * and allocate an fdfile_t up front in case we need it.
754 	 */
755 	while (new >= fdp->fd_dt->dt_nfiles) {
756 		fd_tryexpand(curproc);
757 	}
758 	ff = pool_cache_get(fdfile_cache, PR_WAITOK);
759 
760 	/*
761 	 * If there is already a file open, close it.  If the file is
762 	 * half open, wait for it to be constructed before closing it.
763 	 * XXX Potential for deadlock here?
764 	 */
765 	mutex_enter(&fdp->fd_lock);
766 	while (fd_isused(fdp, new)) {
767 		mutex_exit(&fdp->fd_lock);
768 		if (fd_getfile(new) != NULL) {
769 			(void)fd_close(new);
770 		} else {
771 			/*
772 			 * Crummy, but unlikely to happen.
773 			 * Can occur if we interrupt another
774 			 * thread while it is opening a file.
775 			 */
776 			kpause("dup2", false, 1, NULL);
777 		}
778 		mutex_enter(&fdp->fd_lock);
779 	}
780 	dt = fdp->fd_dt;
781 	if (dt->dt_ff[new] == NULL) {
782 		KASSERT(new >= NDFDFILE);
783 		dt->dt_ff[new] = ff;
784 		ff = NULL;
785 	}
786 	fd_used(fdp, new);
787 	mutex_exit(&fdp->fd_lock);
788 
789 	dt->dt_ff[new]->ff_exclose = (flags & O_CLOEXEC) != 0;
790 	fp->f_flag |= flags & FNONBLOCK;
791 	/* Slot is now allocated.  Insert copy of the file. */
792 	fd_affix(curproc, fp, new);
793 	if (ff != NULL) {
794 		pool_cache_put(fdfile_cache, ff);
795 	}
796 	return 0;
797 }
798 
799 /*
800  * Drop reference to a file structure.
801  */
802 int
803 closef(file_t *fp)
804 {
805 	struct flock lf;
806 	int error;
807 
808 	/*
809 	 * Drop reference.  If referenced elsewhere it's still open
810 	 * and we have nothing more to do.
811 	 */
812 	mutex_enter(&fp->f_lock);
813 	KASSERT(fp->f_count > 0);
814 	if (--fp->f_count > 0) {
815 		mutex_exit(&fp->f_lock);
816 		return 0;
817 	}
818 	KASSERT(fp->f_count == 0);
819 	mutex_exit(&fp->f_lock);
820 
821 	/* We held the last reference - release locks, close and free. */
822 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
823 		lf.l_whence = SEEK_SET;
824 		lf.l_start = 0;
825 		lf.l_len = 0;
826 		lf.l_type = F_UNLCK;
827 		(void)VOP_ADVLOCK(fp->f_data, fp, F_UNLCK, &lf, F_FLOCK);
828 	}
829 	if (fp->f_ops != NULL) {
830 		error = (*fp->f_ops->fo_close)(fp);
831 	} else {
832 		error = 0;
833 	}
834 	KASSERT(fp->f_count == 0);
835 	KASSERT(fp->f_cred != NULL);
836 	pool_cache_put(file_cache, fp);
837 
838 	return error;
839 }
840 
841 /*
842  * Allocate a file descriptor for the process.
843  */
844 int
845 fd_alloc(proc_t *p, int want, int *result)
846 {
847 	filedesc_t *fdp = p->p_fd;
848 	int i, lim, last, error;
849 	u_int off, new;
850 	fdtab_t *dt;
851 
852 	KASSERT(p == curproc || p == &proc0);
853 
854 	/*
855 	 * Search for a free descriptor starting at the higher
856 	 * of want or fd_freefile.
857 	 */
858 	mutex_enter(&fdp->fd_lock);
859 	fd_checkmaps(fdp);
860 	dt = fdp->fd_dt;
861 	KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
862 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
863 	last = min(dt->dt_nfiles, lim);
864 	for (;;) {
865 		if ((i = want) < fdp->fd_freefile)
866 			i = fdp->fd_freefile;
867 		off = i >> NDENTRYSHIFT;
868 		new = fd_next_zero(fdp, fdp->fd_himap, off,
869 		    (last + NDENTRIES - 1) >> NDENTRYSHIFT);
870 		if (new == -1)
871 			break;
872 		i = fd_next_zero(fdp, &fdp->fd_lomap[new],
873 		    new > off ? 0 : i & NDENTRYMASK, NDENTRIES);
874 		if (i == -1) {
875 			/*
876 			 * Free file descriptor in this block was
877 			 * below want, try again with higher want.
878 			 */
879 			want = (new + 1) << NDENTRYSHIFT;
880 			continue;
881 		}
882 		i += (new << NDENTRYSHIFT);
883 		if (i >= last) {
884 			break;
885 		}
886 		if (dt->dt_ff[i] == NULL) {
887 			KASSERT(i >= NDFDFILE);
888 			dt->dt_ff[i] = pool_cache_get(fdfile_cache, PR_WAITOK);
889 		}
890 		KASSERT(dt->dt_ff[i]->ff_file == NULL);
891 		fd_used(fdp, i);
892 		if (want <= fdp->fd_freefile) {
893 			fdp->fd_freefile = i;
894 		}
895 		*result = i;
896 		KASSERT(i >= NDFDFILE ||
897 		    dt->dt_ff[i] == (fdfile_t *)fdp->fd_dfdfile[i]);
898 		fd_checkmaps(fdp);
899 		mutex_exit(&fdp->fd_lock);
900 		return 0;
901 	}
902 
903 	/* No space in current array.  Let the caller expand and retry. */
904 	error = (dt->dt_nfiles >= lim) ? EMFILE : ENOSPC;
905 	mutex_exit(&fdp->fd_lock);
906 	return error;
907 }
908 
909 /*
910  * Allocate memory for a descriptor table.
911  */
912 static fdtab_t *
913 fd_dtab_alloc(int n)
914 {
915 	fdtab_t *dt;
916 	size_t sz;
917 
918 	KASSERT(n > NDFILE);
919 
920 	sz = sizeof(*dt) + (n - NDFILE) * sizeof(dt->dt_ff[0]);
921 	dt = kmem_alloc(sz, KM_SLEEP);
922 #ifdef DIAGNOSTIC
923 	memset(dt, 0xff, sz);
924 #endif
925 	dt->dt_nfiles = n;
926 	dt->dt_link = NULL;
927 	return dt;
928 }
929 
930 /*
931  * Free a descriptor table, and all tables linked for deferred free.
932  */
933 static void
934 fd_dtab_free(fdtab_t *dt)
935 {
936 	fdtab_t *next;
937 	size_t sz;
938 
939 	do {
940 		next = dt->dt_link;
941 		KASSERT(dt->dt_nfiles > NDFILE);
942 		sz = sizeof(*dt) +
943 		    (dt->dt_nfiles - NDFILE) * sizeof(dt->dt_ff[0]);
944 #ifdef DIAGNOSTIC
945 		memset(dt, 0xff, sz);
946 #endif
947 		kmem_free(dt, sz);
948 		dt = next;
949 	} while (dt != NULL);
950 }
951 
952 /*
953  * Allocate descriptor bitmap.
954  */
955 static void
956 fd_map_alloc(int n, uint32_t **lo, uint32_t **hi)
957 {
958 	uint8_t *ptr;
959 	size_t szlo, szhi;
960 
961 	KASSERT(n > NDENTRIES);
962 
963 	szlo = NDLOSLOTS(n) * sizeof(uint32_t);
964 	szhi = NDHISLOTS(n) * sizeof(uint32_t);
965 	ptr = kmem_alloc(szlo + szhi, KM_SLEEP);
966 	*lo = (uint32_t *)ptr;
967 	*hi = (uint32_t *)(ptr + szlo);
968 }
969 
970 /*
971  * Free descriptor bitmap.
972  */
973 static void
974 fd_map_free(int n, uint32_t *lo, uint32_t *hi)
975 {
976 	size_t szlo, szhi;
977 
978 	KASSERT(n > NDENTRIES);
979 
980 	szlo = NDLOSLOTS(n) * sizeof(uint32_t);
981 	szhi = NDHISLOTS(n) * sizeof(uint32_t);
982 	KASSERT(hi == (uint32_t *)((uint8_t *)lo + szlo));
983 	kmem_free(lo, szlo + szhi);
984 }
985 
986 /*
987  * Expand a process' descriptor table.
988  */
989 void
990 fd_tryexpand(proc_t *p)
991 {
992 	filedesc_t *fdp;
993 	int i, numfiles, oldnfiles;
994 	fdtab_t *newdt, *dt;
995 	uint32_t *newhimap, *newlomap;
996 
997 	KASSERT(p == curproc || p == &proc0);
998 
999 	fdp = p->p_fd;
1000 	newhimap = NULL;
1001 	newlomap = NULL;
1002 	oldnfiles = fdp->fd_dt->dt_nfiles;
1003 
1004 	if (oldnfiles < NDEXTENT)
1005 		numfiles = NDEXTENT;
1006 	else
1007 		numfiles = 2 * oldnfiles;
1008 
1009 	newdt = fd_dtab_alloc(numfiles);
1010 	if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
1011 		fd_map_alloc(numfiles, &newlomap, &newhimap);
1012 	}
1013 
1014 	mutex_enter(&fdp->fd_lock);
1015 	dt = fdp->fd_dt;
1016 	KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
1017 	if (dt->dt_nfiles != oldnfiles) {
1018 		/* fdp changed; caller must retry */
1019 		mutex_exit(&fdp->fd_lock);
1020 		fd_dtab_free(newdt);
1021 		if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
1022 			fd_map_free(numfiles, newlomap, newhimap);
1023 		}
1024 		return;
1025 	}
1026 
1027 	/* Copy the existing descriptor table and zero the new portion. */
1028 	i = sizeof(fdfile_t *) * oldnfiles;
1029 	memcpy(newdt->dt_ff, dt->dt_ff, i);
1030 	memset((uint8_t *)newdt->dt_ff + i, 0,
1031 	    numfiles * sizeof(fdfile_t *) - i);
1032 
1033 	/*
1034 	 * Link old descriptor array into list to be discarded.  We defer
1035 	 * freeing until the last reference to the descriptor table goes
1036 	 * away (usually process exit).  This allows us to do lockless
1037 	 * lookups in fd_getfile().
1038 	 */
1039 	if (oldnfiles > NDFILE) {
1040 		if (fdp->fd_refcnt > 1) {
1041 			newdt->dt_link = dt;
1042 		} else {
1043 			fd_dtab_free(dt);
1044 		}
1045 	}
1046 
1047 	if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
1048 		i = NDHISLOTS(oldnfiles) * sizeof(uint32_t);
1049 		memcpy(newhimap, fdp->fd_himap, i);
1050 		memset((uint8_t *)newhimap + i, 0,
1051 		    NDHISLOTS(numfiles) * sizeof(uint32_t) - i);
1052 
1053 		i = NDLOSLOTS(oldnfiles) * sizeof(uint32_t);
1054 		memcpy(newlomap, fdp->fd_lomap, i);
1055 		memset((uint8_t *)newlomap + i, 0,
1056 		    NDLOSLOTS(numfiles) * sizeof(uint32_t) - i);
1057 
1058 		if (NDHISLOTS(oldnfiles) > NDHISLOTS(NDFILE)) {
1059 			fd_map_free(oldnfiles, fdp->fd_lomap, fdp->fd_himap);
1060 		}
1061 		fdp->fd_himap = newhimap;
1062 		fdp->fd_lomap = newlomap;
1063 	}
1064 
1065 	/*
1066 	 * All other modifications must become globally visible before
1067 	 * the change to fd_dt.  See fd_getfile().
1068 	 */
1069 	membar_producer();
1070 	fdp->fd_dt = newdt;
1071 	KASSERT(newdt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
1072 	fd_checkmaps(fdp);
1073 	mutex_exit(&fdp->fd_lock);
1074 }
1075 
1076 /*
1077  * Create a new open file structure and allocate a file descriptor
1078  * for the current process.
1079  */
1080 int
1081 fd_allocfile(file_t **resultfp, int *resultfd)
1082 {
1083 	proc_t *p = curproc;
1084 	kauth_cred_t cred;
1085 	file_t *fp;
1086 	int error;
1087 
1088 	while ((error = fd_alloc(p, 0, resultfd)) != 0) {
1089 		if (error != ENOSPC) {
1090 			return error;
1091 		}
1092 		fd_tryexpand(p);
1093 	}
1094 
1095 	fp = pool_cache_get(file_cache, PR_WAITOK);
1096 	if (fp == NULL) {
1097 		fd_abort(p, NULL, *resultfd);
1098 		return ENFILE;
1099 	}
1100 	KASSERT(fp->f_count == 0);
1101 	KASSERT(fp->f_msgcount == 0);
1102 	KASSERT(fp->f_unpcount == 0);
1103 
1104 	/* Replace cached credentials if not what we need. */
1105 	cred = curlwp->l_cred;
1106 	if (__predict_false(cred != fp->f_cred)) {
1107 		kauth_cred_free(fp->f_cred);
1108 		kauth_cred_hold(cred);
1109 		fp->f_cred = cred;
1110 	}
1111 
1112 	/*
1113 	 * Don't allow recycled files to be scanned.
1114 	 * See uipc_usrreq.c.
1115 	 */
1116 	if (__predict_false((fp->f_flag & FSCAN) != 0)) {
1117 		mutex_enter(&fp->f_lock);
1118 		atomic_and_uint(&fp->f_flag, ~FSCAN);
1119 		mutex_exit(&fp->f_lock);
1120 	}
1121 
1122 	fp->f_advice = 0;
1123 	fp->f_offset = 0;
1124 	*resultfp = fp;
1125 
1126 	return 0;
1127 }
1128 
1129 /*
1130  * Successful creation of a new descriptor: make visible to the process.
1131  */
1132 void
1133 fd_affix(proc_t *p, file_t *fp, unsigned fd)
1134 {
1135 	fdfile_t *ff;
1136 	filedesc_t *fdp;
1137 
1138 	KASSERT(p == curproc || p == &proc0);
1139 
1140 	/* Add a reference to the file structure. */
1141 	mutex_enter(&fp->f_lock);
1142 	fp->f_count++;
1143 	mutex_exit(&fp->f_lock);
1144 
1145 	/*
1146 	 * Insert the new file into the descriptor slot.
1147 	 *
1148 	 * The memory barriers provided by lock activity in this routine
1149 	 * ensure that any updates to the file structure become globally
1150 	 * visible before the file becomes visible to other LWPs in the
1151 	 * current process.
1152 	 */
1153 	fdp = p->p_fd;
1154 	ff = fdp->fd_dt->dt_ff[fd];
1155 
1156 	KASSERT(ff != NULL);
1157 	KASSERT(ff->ff_file == NULL);
1158 	KASSERT(ff->ff_allocated);
1159 	KASSERT(fd_isused(fdp, fd));
1160 	KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
1161 
1162 	/* No need to lock in order to make file initially visible. */
1163 	ff->ff_file = fp;
1164 }
1165 
1166 /*
1167  * Abort creation of a new descriptor: free descriptor slot and file.
1168  */
1169 void
1170 fd_abort(proc_t *p, file_t *fp, unsigned fd)
1171 {
1172 	filedesc_t *fdp;
1173 	fdfile_t *ff;
1174 
1175 	KASSERT(p == curproc || p == &proc0);
1176 
1177 	fdp = p->p_fd;
1178 	ff = fdp->fd_dt->dt_ff[fd];
1179 	ff->ff_exclose = false;
1180 
1181 	KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
1182 
1183 	mutex_enter(&fdp->fd_lock);
1184 	KASSERT(fd_isused(fdp, fd));
1185 	fd_unused(fdp, fd);
1186 	mutex_exit(&fdp->fd_lock);
1187 
1188 	if (fp != NULL) {
1189 		KASSERT(fp->f_count == 0);
1190 		KASSERT(fp->f_cred != NULL);
1191 		pool_cache_put(file_cache, fp);
1192 	}
1193 }
1194 
1195 static int
1196 file_ctor(void *arg, void *obj, int flags)
1197 {
1198 	file_t *fp = obj;
1199 
1200 	memset(fp, 0, sizeof(*fp));
1201 
1202 	mutex_enter(&filelist_lock);
1203 	if (__predict_false(nfiles >= maxfiles)) {
1204 		mutex_exit(&filelist_lock);
1205 		tablefull("file", "increase kern.maxfiles or MAXFILES");
1206 		return ENFILE;
1207 	}
1208 	nfiles++;
1209 	LIST_INSERT_HEAD(&filehead, fp, f_list);
1210 	mutex_init(&fp->f_lock, MUTEX_DEFAULT, IPL_NONE);
1211 	fp->f_cred = curlwp->l_cred;
1212 	kauth_cred_hold(fp->f_cred);
1213 	mutex_exit(&filelist_lock);
1214 
1215 	return 0;
1216 }
1217 
1218 static void
1219 file_dtor(void *arg, void *obj)
1220 {
1221 	file_t *fp = obj;
1222 
1223 	mutex_enter(&filelist_lock);
1224 	nfiles--;
1225 	LIST_REMOVE(fp, f_list);
1226 	mutex_exit(&filelist_lock);
1227 
1228 	kauth_cred_free(fp->f_cred);
1229 	mutex_destroy(&fp->f_lock);
1230 }
1231 
1232 static int
1233 fdfile_ctor(void *arg, void *obj, int flags)
1234 {
1235 	fdfile_t *ff = obj;
1236 
1237 	memset(ff, 0, sizeof(*ff));
1238 	cv_init(&ff->ff_closing, "fdclose");
1239 
1240 	return 0;
1241 }
1242 
1243 static void
1244 fdfile_dtor(void *arg, void *obj)
1245 {
1246 	fdfile_t *ff = obj;
1247 
1248 	cv_destroy(&ff->ff_closing);
1249 }
1250 
1251 file_t *
1252 fgetdummy(void)
1253 {
1254 	file_t *fp;
1255 
1256 	fp = kmem_zalloc(sizeof(*fp), KM_SLEEP);
1257 	if (fp != NULL) {
1258 		mutex_init(&fp->f_lock, MUTEX_DEFAULT, IPL_NONE);
1259 	}
1260 	return fp;
1261 }
1262 
1263 void
1264 fputdummy(file_t *fp)
1265 {
1266 
1267 	mutex_destroy(&fp->f_lock);
1268 	kmem_free(fp, sizeof(*fp));
1269 }
1270 
1271 /*
1272  * Create an initial filedesc structure.
1273  */
1274 filedesc_t *
1275 fd_init(filedesc_t *fdp)
1276 {
1277 #ifdef DIAGNOSTIC
1278 	unsigned fd;
1279 #endif
1280 
1281 	if (__predict_true(fdp == NULL)) {
1282 		fdp = pool_cache_get(filedesc_cache, PR_WAITOK);
1283 	} else {
1284 		KASSERT(fdp == &filedesc0);
1285 		filedesc_ctor(NULL, fdp, PR_WAITOK);
1286 	}
1287 
1288 #ifdef DIAGNOSTIC
1289 	KASSERT(fdp->fd_lastfile == -1);
1290 	KASSERT(fdp->fd_lastkqfile == -1);
1291 	KASSERT(fdp->fd_knhash == NULL);
1292 	KASSERT(fdp->fd_freefile == 0);
1293 	KASSERT(fdp->fd_exclose == false);
1294 	KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin);
1295 	KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE);
1296 	for (fd = 0; fd < NDFDFILE; fd++) {
1297 		KASSERT(fdp->fd_dtbuiltin.dt_ff[fd] ==
1298 		    (fdfile_t *)fdp->fd_dfdfile[fd]);
1299 	}
1300 	for (fd = NDFDFILE; fd < NDFILE; fd++) {
1301 		KASSERT(fdp->fd_dtbuiltin.dt_ff[fd] == NULL);
1302 	}
1303 	KASSERT(fdp->fd_himap == fdp->fd_dhimap);
1304 	KASSERT(fdp->fd_lomap == fdp->fd_dlomap);
1305 #endif	/* DIAGNOSTIC */
1306 
1307 	fdp->fd_refcnt = 1;
1308 	fd_checkmaps(fdp);
1309 
1310 	return fdp;
1311 }
1312 
1313 /*
1314  * Initialize a file descriptor table.
1315  */
1316 static int
1317 filedesc_ctor(void *arg, void *obj, int flag)
1318 {
1319 	filedesc_t *fdp = obj;
1320 	fdfile_t **ffp;
1321 	int i;
1322 
1323 	memset(fdp, 0, sizeof(*fdp));
1324 	mutex_init(&fdp->fd_lock, MUTEX_DEFAULT, IPL_NONE);
1325 	fdp->fd_lastfile = -1;
1326 	fdp->fd_lastkqfile = -1;
1327 	fdp->fd_dt = &fdp->fd_dtbuiltin;
1328 	fdp->fd_dtbuiltin.dt_nfiles = NDFILE;
1329 	fdp->fd_himap = fdp->fd_dhimap;
1330 	fdp->fd_lomap = fdp->fd_dlomap;
1331 
1332 	CTASSERT(sizeof(fdp->fd_dfdfile[0]) >= sizeof(fdfile_t));
1333 	for (i = 0, ffp = fdp->fd_dt->dt_ff; i < NDFDFILE; i++, ffp++) {
1334 		*ffp = (fdfile_t *)fdp->fd_dfdfile[i];
1335 		(void)fdfile_ctor(NULL, fdp->fd_dfdfile[i], PR_WAITOK);
1336 	}
1337 
1338 	return 0;
1339 }
1340 
1341 static void
1342 filedesc_dtor(void *arg, void *obj)
1343 {
1344 	filedesc_t *fdp = obj;
1345 	int i;
1346 
1347 	for (i = 0; i < NDFDFILE; i++) {
1348 		fdfile_dtor(NULL, fdp->fd_dfdfile[i]);
1349 	}
1350 
1351 	mutex_destroy(&fdp->fd_lock);
1352 }
1353 
1354 /*
1355  * Make p share curproc's filedesc structure.
1356  */
1357 void
1358 fd_share(struct proc *p)
1359 {
1360 	filedesc_t *fdp;
1361 
1362 	fdp = curlwp->l_fd;
1363 	p->p_fd = fdp;
1364 	atomic_inc_uint(&fdp->fd_refcnt);
1365 }
1366 
1367 /*
1368  * Acquire a hold on a filedesc structure.
1369  */
1370 void
1371 fd_hold(lwp_t *l)
1372 {
1373 	filedesc_t *fdp = l->l_fd;
1374 
1375 	atomic_inc_uint(&fdp->fd_refcnt);
1376 }
1377 
1378 /*
1379  * Copy a filedesc structure.
1380  */
1381 filedesc_t *
1382 fd_copy(void)
1383 {
1384 	filedesc_t *newfdp, *fdp;
1385 	fdfile_t *ff, **ffp, **nffp, *ff2;
1386 	int i, j, numfiles, lastfile, newlast;
1387 	file_t *fp;
1388 	fdtab_t *newdt;
1389 
1390 	fdp = curproc->p_fd;
1391 	newfdp = pool_cache_get(filedesc_cache, PR_WAITOK);
1392 	newfdp->fd_refcnt = 1;
1393 
1394 #ifdef DIAGNOSTIC
1395 	KASSERT(newfdp->fd_lastfile == -1);
1396 	KASSERT(newfdp->fd_lastkqfile == -1);
1397 	KASSERT(newfdp->fd_knhash == NULL);
1398 	KASSERT(newfdp->fd_freefile == 0);
1399 	KASSERT(newfdp->fd_exclose == false);
1400 	KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin);
1401 	KASSERT(newfdp->fd_dtbuiltin.dt_nfiles == NDFILE);
1402 	for (i = 0; i < NDFDFILE; i++) {
1403 		KASSERT(newfdp->fd_dtbuiltin.dt_ff[i] ==
1404 		    (fdfile_t *)&newfdp->fd_dfdfile[i]);
1405 	}
1406 	for (i = NDFDFILE; i < NDFILE; i++) {
1407 		KASSERT(newfdp->fd_dtbuiltin.dt_ff[i] == NULL);
1408 	}
1409 #endif	/* DIAGNOSTIC */
1410 
1411 	mutex_enter(&fdp->fd_lock);
1412 	fd_checkmaps(fdp);
1413 	numfiles = fdp->fd_dt->dt_nfiles;
1414 	lastfile = fdp->fd_lastfile;
1415 
1416 	/*
1417 	 * If the number of open files fits in the internal arrays
1418 	 * of the open file structure, use them, otherwise allocate
1419 	 * additional memory for the number of descriptors currently
1420 	 * in use.
1421 	 */
1422 	if (lastfile < NDFILE) {
1423 		i = NDFILE;
1424 		newdt = newfdp->fd_dt;
1425 		KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin);
1426 	} else {
1427 		/*
1428 		 * Compute the smallest multiple of NDEXTENT needed
1429 		 * for the file descriptors currently in use,
1430 		 * allowing the table to shrink.
1431 		 */
1432 		i = numfiles;
1433 		while (i >= 2 * NDEXTENT && i > lastfile * 2) {
1434 			i /= 2;
1435 		}
1436 		KASSERT(i > NDFILE);
1437 		newdt = fd_dtab_alloc(i);
1438 		newfdp->fd_dt = newdt;
1439 		memcpy(newdt->dt_ff, newfdp->fd_dtbuiltin.dt_ff,
1440 		    NDFDFILE * sizeof(fdfile_t **));
1441 		memset(newdt->dt_ff + NDFDFILE, 0,
1442 		    (i - NDFDFILE) * sizeof(fdfile_t **));
1443 	}
1444 	if (NDHISLOTS(i) <= NDHISLOTS(NDFILE)) {
1445 		newfdp->fd_himap = newfdp->fd_dhimap;
1446 		newfdp->fd_lomap = newfdp->fd_dlomap;
1447 	} else {
1448 		fd_map_alloc(i, &newfdp->fd_lomap, &newfdp->fd_himap);
1449 		KASSERT(i >= NDENTRIES * NDENTRIES);
1450 		memset(newfdp->fd_himap, 0, NDHISLOTS(i)*sizeof(uint32_t));
1451 		memset(newfdp->fd_lomap, 0, NDLOSLOTS(i)*sizeof(uint32_t));
1452 	}
1453 	newfdp->fd_freefile = fdp->fd_freefile;
1454 	newfdp->fd_exclose = fdp->fd_exclose;
1455 
1456 	ffp = fdp->fd_dt->dt_ff;
1457 	nffp = newdt->dt_ff;
1458 	newlast = -1;
1459 	for (i = 0; i <= (int)lastfile; i++, ffp++, nffp++) {
1460 		KASSERT(i >= NDFDFILE ||
1461 		    *nffp == (fdfile_t *)newfdp->fd_dfdfile[i]);
1462 		ff = *ffp;
1463 		if (ff == NULL || (fp = ff->ff_file) == NULL) {
1464 			/* Descriptor unused, or descriptor half open. */
1465 			KASSERT(!fd_isused(newfdp, i));
1466 			continue;
1467 		}
1468 		if (__predict_false(fp->f_type == DTYPE_KQUEUE)) {
1469 			/* kqueue descriptors cannot be copied. */
1470 			if (i < newfdp->fd_freefile) {
1471 				newfdp->fd_freefile = i;
1472 			}
1473 			continue;
1474 		}
1475 		/* It's active: add a reference to the file. */
1476 		mutex_enter(&fp->f_lock);
1477 		fp->f_count++;
1478 		mutex_exit(&fp->f_lock);
1479 
1480 		/* Allocate an fdfile_t to represent it. */
1481 		if (i >= NDFDFILE) {
1482 			ff2 = pool_cache_get(fdfile_cache, PR_WAITOK);
1483 			*nffp = ff2;
1484 		} else {
1485 			ff2 = newdt->dt_ff[i];
1486 		}
1487 		ff2->ff_file = fp;
1488 		ff2->ff_exclose = ff->ff_exclose;
1489 		ff2->ff_allocated = true;
1490 
1491 		/* Fix up bitmaps. */
1492 		j = i >> NDENTRYSHIFT;
1493 		KASSERT((newfdp->fd_lomap[j] & (1 << (i & NDENTRYMASK))) == 0);
1494 		newfdp->fd_lomap[j] |= 1 << (i & NDENTRYMASK);
1495 		if (__predict_false(newfdp->fd_lomap[j] == ~0)) {
1496 			KASSERT((newfdp->fd_himap[j >> NDENTRYSHIFT] &
1497 			    (1 << (j & NDENTRYMASK))) == 0);
1498 			newfdp->fd_himap[j >> NDENTRYSHIFT] |=
1499 			    1 << (j & NDENTRYMASK);
1500 		}
1501 		newlast = i;
1502 	}
1503 	KASSERT(newdt->dt_ff[0] == (fdfile_t *)newfdp->fd_dfdfile[0]);
1504 	newfdp->fd_lastfile = newlast;
1505 	fd_checkmaps(newfdp);
1506 	mutex_exit(&fdp->fd_lock);
1507 
1508 	return newfdp;
1509 }
1510 
1511 /*
1512  * Release a filedesc structure.
1513  */
1514 void
1515 fd_free(void)
1516 {
1517 	fdfile_t *ff;
1518 	file_t *fp;
1519 	int fd, nf;
1520 	fdtab_t *dt;
1521 	lwp_t * const l = curlwp;
1522 	filedesc_t * const fdp = l->l_fd;
1523 	const bool noadvlock = (l->l_proc->p_flag & PK_ADVLOCK) == 0;
1524 
1525 	KASSERT(fdp->fd_dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
1526 	KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE);
1527 	KASSERT(fdp->fd_dtbuiltin.dt_link == NULL);
1528 
1529 #ifndef __HAVE_ATOMIC_AS_MEMBAR
1530 	membar_exit();
1531 #endif
1532 	if (atomic_dec_uint_nv(&fdp->fd_refcnt) > 0)
1533 		return;
1534 
1535 	/*
1536 	 * Close any files that the process holds open.
1537 	 */
1538 	dt = fdp->fd_dt;
1539 	fd_checkmaps(fdp);
1540 #ifdef DEBUG
1541 	fdp->fd_refcnt = -1; /* see fd_checkmaps */
1542 #endif
1543 	for (fd = 0, nf = dt->dt_nfiles; fd < nf; fd++) {
1544 		ff = dt->dt_ff[fd];
1545 		KASSERT(fd >= NDFDFILE ||
1546 		    ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
1547 		if (ff == NULL)
1548 			continue;
1549 		if ((fp = ff->ff_file) != NULL) {
1550 			/*
1551 			 * Must use fd_close() here if there is
1552 			 * a reference from kqueue or we might have posix
1553 			 * advisory locks.
1554 			 */
1555 			if (__predict_true(ff->ff_refcnt == 0) &&
1556 			    (noadvlock || fp->f_type != DTYPE_VNODE)) {
1557 				ff->ff_file = NULL;
1558 				ff->ff_exclose = false;
1559 				ff->ff_allocated = false;
1560 				closef(fp);
1561 			} else {
1562 				ff->ff_refcnt++;
1563 				fd_close(fd);
1564 			}
1565 		}
1566 		KASSERT(ff->ff_refcnt == 0);
1567 		KASSERT(ff->ff_file == NULL);
1568 		KASSERT(!ff->ff_exclose);
1569 		KASSERT(!ff->ff_allocated);
1570 		if (fd >= NDFDFILE) {
1571 			pool_cache_put(fdfile_cache, ff);
1572 			dt->dt_ff[fd] = NULL;
1573 		}
1574 	}
1575 
1576 	/*
1577 	 * Clean out the descriptor table for the next user and return
1578 	 * to the cache.
1579 	 */
1580 	if (__predict_false(dt != &fdp->fd_dtbuiltin)) {
1581 		fd_dtab_free(fdp->fd_dt);
1582 		/* Otherwise, done above. */
1583 		memset(&fdp->fd_dtbuiltin.dt_ff[NDFDFILE], 0,
1584 		    (NDFILE - NDFDFILE) * sizeof(fdp->fd_dtbuiltin.dt_ff[0]));
1585 		fdp->fd_dt = &fdp->fd_dtbuiltin;
1586 	}
1587 	if (__predict_false(NDHISLOTS(nf) > NDHISLOTS(NDFILE))) {
1588 		KASSERT(fdp->fd_himap != fdp->fd_dhimap);
1589 		KASSERT(fdp->fd_lomap != fdp->fd_dlomap);
1590 		fd_map_free(nf, fdp->fd_lomap, fdp->fd_himap);
1591 	}
1592 	if (__predict_false(fdp->fd_knhash != NULL)) {
1593 		hashdone(fdp->fd_knhash, HASH_LIST, fdp->fd_knhashmask);
1594 		fdp->fd_knhash = NULL;
1595 		fdp->fd_knhashmask = 0;
1596 	} else {
1597 		KASSERT(fdp->fd_knhashmask == 0);
1598 	}
1599 	fdp->fd_dt = &fdp->fd_dtbuiltin;
1600 	fdp->fd_lastkqfile = -1;
1601 	fdp->fd_lastfile = -1;
1602 	fdp->fd_freefile = 0;
1603 	fdp->fd_exclose = false;
1604 	memset(&fdp->fd_startzero, 0, sizeof(*fdp) -
1605 	    offsetof(filedesc_t, fd_startzero));
1606 	fdp->fd_himap = fdp->fd_dhimap;
1607 	fdp->fd_lomap = fdp->fd_dlomap;
1608 	KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE);
1609 	KASSERT(fdp->fd_dtbuiltin.dt_link == NULL);
1610 	KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin);
1611 #ifdef DEBUG
1612 	fdp->fd_refcnt = 0; /* see fd_checkmaps */
1613 #endif
1614 	fd_checkmaps(fdp);
1615 	pool_cache_put(filedesc_cache, fdp);
1616 }
1617 
1618 /*
1619  * File Descriptor pseudo-device driver (/dev/fd/).
1620  *
1621  * Opening minor device N dup()s the file (if any) connected to file
1622  * descriptor N belonging to the calling process.  Note that this driver
1623  * consists of only the ``open()'' routine, because all subsequent
1624  * references to this file will be direct to the other driver.
1625  */
1626 static int
1627 filedescopen(dev_t dev, int mode, int type, lwp_t *l)
1628 {
1629 
1630 	/*
1631 	 * XXX Kludge: set dupfd to contain the value of the
1632 	 * the file descriptor being sought for duplication. The error
1633 	 * return ensures that the vnode for this device will be released
1634 	 * by vn_open. Open will detect this special error and take the
1635 	 * actions in fd_dupopen below. Other callers of vn_open or VOP_OPEN
1636 	 * will simply report the error.
1637 	 */
1638 	l->l_dupfd = minor(dev);	/* XXX */
1639 	return EDUPFD;
1640 }
1641 
1642 /*
1643  * Duplicate the specified descriptor to a free descriptor.
1644  */
1645 int
1646 fd_dupopen(int old, int *new, int mode, int error)
1647 {
1648 	filedesc_t *fdp;
1649 	fdfile_t *ff;
1650 	file_t *fp;
1651 	fdtab_t *dt;
1652 
1653 	if ((fp = fd_getfile(old)) == NULL) {
1654 		return EBADF;
1655 	}
1656 	fdp = curlwp->l_fd;
1657 	dt = fdp->fd_dt;
1658 	ff = dt->dt_ff[old];
1659 
1660 	/*
1661 	 * There are two cases of interest here.
1662 	 *
1663 	 * For EDUPFD simply dup (old) to file descriptor
1664 	 * (new) and return.
1665 	 *
1666 	 * For EMOVEFD steal away the file structure from (old) and
1667 	 * store it in (new).  (old) is effectively closed by
1668 	 * this operation.
1669 	 *
1670 	 * Any other error code is just returned.
1671 	 */
1672 	switch (error) {
1673 	case EDUPFD:
1674 		/*
1675 		 * Check that the mode the file is being opened for is a
1676 		 * subset of the mode of the existing descriptor.
1677 		 */
1678 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
1679 			error = EACCES;
1680 			break;
1681 		}
1682 
1683 		/* Copy it. */
1684 		error = fd_dup(fp, 0, new, ff->ff_exclose);
1685 		break;
1686 
1687 	case EMOVEFD:
1688 		/* Copy it. */
1689 		error = fd_dup(fp, 0, new, ff->ff_exclose);
1690 		if (error != 0) {
1691 			break;
1692 		}
1693 
1694 		/* Steal away the file pointer from 'old'. */
1695 		(void)fd_close(old);
1696 		return 0;
1697 	}
1698 
1699 	fd_putfile(old);
1700 	return error;
1701 }
1702 
1703 /*
1704  * Close open files on exec.
1705  */
1706 void
1707 fd_closeexec(void)
1708 {
1709 	proc_t *p;
1710 	filedesc_t *fdp;
1711 	fdfile_t *ff;
1712 	lwp_t *l;
1713 	fdtab_t *dt;
1714 	int fd;
1715 
1716 	l = curlwp;
1717 	p = l->l_proc;
1718 	fdp = p->p_fd;
1719 
1720 	if (fdp->fd_refcnt > 1) {
1721 		fdp = fd_copy();
1722 		fd_free();
1723 		p->p_fd = fdp;
1724 		l->l_fd = fdp;
1725 	}
1726 	if (!fdp->fd_exclose) {
1727 		return;
1728 	}
1729 	fdp->fd_exclose = false;
1730 	dt = fdp->fd_dt;
1731 
1732 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
1733 		if ((ff = dt->dt_ff[fd]) == NULL) {
1734 			KASSERT(fd >= NDFDFILE);
1735 			continue;
1736 		}
1737 		KASSERT(fd >= NDFDFILE ||
1738 		    ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
1739 		if (ff->ff_file == NULL)
1740 			continue;
1741 		if (ff->ff_exclose) {
1742 			/*
1743 			 * We need a reference to close the file.
1744 			 * No other threads can see the fdfile_t at
1745 			 * this point, so don't bother locking.
1746 			 */
1747 			KASSERT((ff->ff_refcnt & FR_CLOSING) == 0);
1748 			ff->ff_refcnt++;
1749 			fd_close(fd);
1750 		}
1751 	}
1752 }
1753 
1754 /*
1755  * Sets descriptor owner. If the owner is a process, 'pgid'
1756  * is set to positive value, process ID. If the owner is process group,
1757  * 'pgid' is set to -pg_id.
1758  */
1759 int
1760 fsetown(pid_t *pgid, u_long cmd, const void *data)
1761 {
1762 	pid_t id = *(const pid_t *)data;
1763 	int error;
1764 
1765 	switch (cmd) {
1766 	case TIOCSPGRP:
1767 		if (id < 0)
1768 			return EINVAL;
1769 		id = -id;
1770 		break;
1771 	default:
1772 		break;
1773 	}
1774 	if (id > 0) {
1775 		mutex_enter(proc_lock);
1776 		error = proc_find(id) ? 0 : ESRCH;
1777 		mutex_exit(proc_lock);
1778 	} else if (id < 0) {
1779 		error = pgid_in_session(curproc, -id);
1780 	} else {
1781 		error = 0;
1782 	}
1783 	if (!error) {
1784 		*pgid = id;
1785 	}
1786 	return error;
1787 }
1788 
1789 void
1790 fd_set_exclose(struct lwp *l, int fd, bool exclose)
1791 {
1792 	filedesc_t *fdp = l->l_fd;
1793 	fdfile_t *ff = fdp->fd_dt->dt_ff[fd];
1794 
1795 	ff->ff_exclose = exclose;
1796 	if (exclose)
1797 		fdp->fd_exclose = true;
1798 }
1799 
1800 /*
1801  * Return descriptor owner information. If the value is positive,
1802  * it's process ID. If it's negative, it's process group ID and
1803  * needs the sign removed before use.
1804  */
1805 int
1806 fgetown(pid_t pgid, u_long cmd, void *data)
1807 {
1808 
1809 	switch (cmd) {
1810 	case TIOCGPGRP:
1811 		*(int *)data = -pgid;
1812 		break;
1813 	default:
1814 		*(int *)data = pgid;
1815 		break;
1816 	}
1817 	return 0;
1818 }
1819 
1820 /*
1821  * Send signal to descriptor owner, either process or process group.
1822  */
1823 void
1824 fownsignal(pid_t pgid, int signo, int code, int band, void *fdescdata)
1825 {
1826 	ksiginfo_t ksi;
1827 
1828 	KASSERT(!cpu_intr_p());
1829 
1830 	if (pgid == 0) {
1831 		return;
1832 	}
1833 
1834 	KSI_INIT(&ksi);
1835 	ksi.ksi_signo = signo;
1836 	ksi.ksi_code = code;
1837 	ksi.ksi_band = band;
1838 
1839 	mutex_enter(proc_lock);
1840 	if (pgid > 0) {
1841 		struct proc *p1;
1842 
1843 		p1 = proc_find(pgid);
1844 		if (p1 != NULL) {
1845 			kpsignal(p1, &ksi, fdescdata);
1846 		}
1847 	} else {
1848 		struct pgrp *pgrp;
1849 
1850 		KASSERT(pgid < 0);
1851 		pgrp = pgrp_find(-pgid);
1852 		if (pgrp != NULL) {
1853 			kpgsignal(pgrp, &ksi, fdescdata, 0);
1854 		}
1855 	}
1856 	mutex_exit(proc_lock);
1857 }
1858 
1859 int
1860 fd_clone(file_t *fp, unsigned fd, int flag, const struct fileops *fops,
1861 	 void *data)
1862 {
1863 
1864 	fp->f_flag = flag;
1865 	fp->f_type = DTYPE_MISC;
1866 	fp->f_ops = fops;
1867 	fp->f_data = data;
1868 	curlwp->l_dupfd = fd;
1869 	fd_affix(curproc, fp, fd);
1870 
1871 	return EMOVEFD;
1872 }
1873 
1874 int
1875 fnullop_fcntl(file_t *fp, u_int cmd, void *data)
1876 {
1877 
1878 	if (cmd == F_SETFL)
1879 		return 0;
1880 
1881 	return EOPNOTSUPP;
1882 }
1883 
1884 int
1885 fnullop_poll(file_t *fp, int which)
1886 {
1887 
1888 	return 0;
1889 }
1890 
1891 int
1892 fnullop_kqfilter(file_t *fp, struct knote *kn)
1893 {
1894 
1895 	return EOPNOTSUPP;
1896 }
1897 
1898 void
1899 fnullop_restart(file_t *fp)
1900 {
1901 
1902 }
1903 
1904 int
1905 fbadop_read(file_t *fp, off_t *offset, struct uio *uio,
1906 	    kauth_cred_t cred, int flags)
1907 {
1908 
1909 	return EOPNOTSUPP;
1910 }
1911 
1912 int
1913 fbadop_write(file_t *fp, off_t *offset, struct uio *uio,
1914 	     kauth_cred_t cred, int flags)
1915 {
1916 
1917 	return EOPNOTSUPP;
1918 }
1919 
1920 int
1921 fbadop_ioctl(file_t *fp, u_long com, void *data)
1922 {
1923 
1924 	return EOPNOTSUPP;
1925 }
1926 
1927 int
1928 fbadop_stat(file_t *fp, struct stat *sb)
1929 {
1930 
1931 	return EOPNOTSUPP;
1932 }
1933 
1934 int
1935 fbadop_close(file_t *fp)
1936 {
1937 
1938 	return EOPNOTSUPP;
1939 }
1940 
1941 /*
1942  * sysctl routines pertaining to file descriptors
1943  */
1944 
1945 /* Initialized in sysctl_init() for now... */
1946 extern kmutex_t sysctl_file_marker_lock;
1947 static u_int sysctl_file_marker = 1;
1948 
1949 /*
1950  * Expects to be called with proc_lock and sysctl_file_marker_lock locked.
1951  */
1952 static void
1953 sysctl_file_marker_reset(void)
1954 {
1955 	struct proc *p;
1956 
1957 	PROCLIST_FOREACH(p, &allproc) {
1958 		struct filedesc *fd = p->p_fd;
1959 		fdtab_t *dt;
1960 		u_int i;
1961 
1962 		mutex_enter(&fd->fd_lock);
1963 		dt = fd->fd_dt;
1964 		for (i = 0; i < dt->dt_nfiles; i++) {
1965 			struct file *fp;
1966 			fdfile_t *ff;
1967 
1968 			if ((ff = dt->dt_ff[i]) == NULL) {
1969 				continue;
1970 			}
1971 			if ((fp = ff->ff_file) == NULL) {
1972 				continue;
1973 			}
1974 			fp->f_marker = 0;
1975 		}
1976 		mutex_exit(&fd->fd_lock);
1977 	}
1978 }
1979 
1980 /*
1981  * sysctl helper routine for kern.file pseudo-subtree.
1982  */
1983 static int
1984 sysctl_kern_file(SYSCTLFN_ARGS)
1985 {
1986 	int error;
1987 	size_t buflen;
1988 	struct file *fp, fbuf;
1989 	char *start, *where;
1990 	struct proc *p;
1991 
1992 	start = where = oldp;
1993 	buflen = *oldlenp;
1994 
1995 	if (where == NULL) {
1996 		/*
1997 		 * overestimate by 10 files
1998 		 */
1999 		*oldlenp = sizeof(filehead) + (nfiles + 10) *
2000 		    sizeof(struct file);
2001 		return 0;
2002 	}
2003 
2004 	/*
2005 	 * first sysctl_copyout filehead
2006 	 */
2007 	if (buflen < sizeof(filehead)) {
2008 		*oldlenp = 0;
2009 		return 0;
2010 	}
2011 	sysctl_unlock();
2012 	error = sysctl_copyout(l, &filehead, where, sizeof(filehead));
2013 	if (error) {
2014 		sysctl_relock();
2015 		return error;
2016 	}
2017 	buflen -= sizeof(filehead);
2018 	where += sizeof(filehead);
2019 
2020 	/*
2021 	 * followed by an array of file structures
2022 	 */
2023 	mutex_enter(&sysctl_file_marker_lock);
2024 	mutex_enter(proc_lock);
2025 	PROCLIST_FOREACH(p, &allproc) {
2026 		struct filedesc *fd;
2027 		fdtab_t *dt;
2028 		u_int i;
2029 
2030 		if (p->p_stat == SIDL) {
2031 			/* skip embryonic processes */
2032 			continue;
2033 		}
2034 		mutex_enter(p->p_lock);
2035 		error = kauth_authorize_process(l->l_cred,
2036 		    KAUTH_PROCESS_CANSEE, p,
2037 		    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES),
2038 		    NULL, NULL);
2039 		mutex_exit(p->p_lock);
2040 		if (error != 0) {
2041 			/*
2042 			 * Don't leak kauth retval if we're silently
2043 			 * skipping this entry.
2044 			 */
2045 			error = 0;
2046 			continue;
2047 		}
2048 
2049 		/*
2050 		 * Grab a hold on the process.
2051 		 */
2052 		if (!rw_tryenter(&p->p_reflock, RW_READER)) {
2053 			continue;
2054 		}
2055 		mutex_exit(proc_lock);
2056 
2057 		fd = p->p_fd;
2058 		mutex_enter(&fd->fd_lock);
2059 		dt = fd->fd_dt;
2060 		for (i = 0; i < dt->dt_nfiles; i++) {
2061 			fdfile_t *ff;
2062 
2063 			if ((ff = dt->dt_ff[i]) == NULL) {
2064 				continue;
2065 			}
2066 			if ((fp = ff->ff_file) == NULL) {
2067 				continue;
2068 			}
2069 
2070 			mutex_enter(&fp->f_lock);
2071 
2072 			if ((fp->f_count == 0) ||
2073 			    (fp->f_marker == sysctl_file_marker)) {
2074 				mutex_exit(&fp->f_lock);
2075 				continue;
2076 			}
2077 
2078 			/* Check that we have enough space. */
2079 			if (buflen < sizeof(struct file)) {
2080 				*oldlenp = where - start;
2081 				mutex_exit(&fp->f_lock);
2082 				error = ENOMEM;
2083 				break;
2084 			}
2085 
2086 			memcpy(&fbuf, fp, sizeof(fbuf));
2087 			mutex_exit(&fp->f_lock);
2088 			error = sysctl_copyout(l, &fbuf, where, sizeof(fbuf));
2089 			if (error) {
2090 				break;
2091 			}
2092 			buflen -= sizeof(struct file);
2093 			where += sizeof(struct file);
2094 
2095 			fp->f_marker = sysctl_file_marker;
2096 		}
2097 		mutex_exit(&fd->fd_lock);
2098 
2099 		/*
2100 		 * Release reference to process.
2101 		 */
2102 		mutex_enter(proc_lock);
2103 		rw_exit(&p->p_reflock);
2104 
2105 		if (error)
2106 			break;
2107 	}
2108 
2109 	sysctl_file_marker++;
2110 	/* Reset all markers if wrapped. */
2111 	if (sysctl_file_marker == 0) {
2112 		sysctl_file_marker_reset();
2113 		sysctl_file_marker++;
2114 	}
2115 
2116 	mutex_exit(proc_lock);
2117 	mutex_exit(&sysctl_file_marker_lock);
2118 
2119 	*oldlenp = where - start;
2120 	sysctl_relock();
2121 	return error;
2122 }
2123 
2124 /*
2125  * sysctl helper function for kern.file2
2126  */
2127 static int
2128 sysctl_kern_file2(SYSCTLFN_ARGS)
2129 {
2130 	struct proc *p;
2131 	struct file *fp;
2132 	struct filedesc *fd;
2133 	struct kinfo_file kf;
2134 	char *dp;
2135 	u_int i, op;
2136 	size_t len, needed, elem_size, out_size;
2137 	int error, arg, elem_count;
2138 	fdfile_t *ff;
2139 	fdtab_t *dt;
2140 
2141 	if (namelen == 1 && name[0] == CTL_QUERY)
2142 		return sysctl_query(SYSCTLFN_CALL(rnode));
2143 
2144 	if (namelen != 4)
2145 		return EINVAL;
2146 
2147 	error = 0;
2148 	dp = oldp;
2149 	len = (oldp != NULL) ? *oldlenp : 0;
2150 	op = name[0];
2151 	arg = name[1];
2152 	elem_size = name[2];
2153 	elem_count = name[3];
2154 	out_size = MIN(sizeof(kf), elem_size);
2155 	needed = 0;
2156 
2157 	if (elem_size < 1 || elem_count < 0)
2158 		return EINVAL;
2159 
2160 	switch (op) {
2161 	case KERN_FILE_BYFILE:
2162 	case KERN_FILE_BYPID:
2163 		/*
2164 		 * We're traversing the process list in both cases; the BYFILE
2165 		 * case does additional work of keeping track of files already
2166 		 * looked at.
2167 		 */
2168 
2169 		/* doesn't use arg so it must be zero */
2170 		if ((op == KERN_FILE_BYFILE) && (arg != 0))
2171 			return EINVAL;
2172 
2173 		if ((op == KERN_FILE_BYPID) && (arg < -1))
2174 			/* -1 means all processes */
2175 			return EINVAL;
2176 
2177 		sysctl_unlock();
2178 		if (op == KERN_FILE_BYFILE)
2179 			mutex_enter(&sysctl_file_marker_lock);
2180 		mutex_enter(proc_lock);
2181 		PROCLIST_FOREACH(p, &allproc) {
2182 			if (p->p_stat == SIDL) {
2183 				/* skip embryonic processes */
2184 				continue;
2185 			}
2186 			if (arg > 0 && p->p_pid != arg) {
2187 				/* pick only the one we want */
2188 				/* XXX want 0 to mean "kernel files" */
2189 				continue;
2190 			}
2191 			mutex_enter(p->p_lock);
2192 			error = kauth_authorize_process(l->l_cred,
2193 			    KAUTH_PROCESS_CANSEE, p,
2194 			    KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES),
2195 			    NULL, NULL);
2196 			mutex_exit(p->p_lock);
2197 			if (error != 0) {
2198 				/*
2199 				 * Don't leak kauth retval if we're silently
2200 				 * skipping this entry.
2201 				 */
2202 				error = 0;
2203 				continue;
2204 			}
2205 
2206 			/*
2207 			 * Grab a hold on the process.
2208 			 */
2209 			if (!rw_tryenter(&p->p_reflock, RW_READER)) {
2210 				continue;
2211 			}
2212 			mutex_exit(proc_lock);
2213 
2214 			fd = p->p_fd;
2215 			mutex_enter(&fd->fd_lock);
2216 			dt = fd->fd_dt;
2217 			for (i = 0; i < dt->dt_nfiles; i++) {
2218 				if ((ff = dt->dt_ff[i]) == NULL) {
2219 					continue;
2220 				}
2221 				if ((fp = ff->ff_file) == NULL) {
2222 					continue;
2223 				}
2224 
2225 				if ((op == KERN_FILE_BYFILE) &&
2226 				    (fp->f_marker == sysctl_file_marker)) {
2227 					continue;
2228 				}
2229 				if (len >= elem_size && elem_count > 0) {
2230 					mutex_enter(&fp->f_lock);
2231 					fill_file(&kf, fp, ff, i, p->p_pid);
2232 					mutex_exit(&fp->f_lock);
2233 					mutex_exit(&fd->fd_lock);
2234 					error = sysctl_copyout(l,
2235 					    &kf, dp, out_size);
2236 					mutex_enter(&fd->fd_lock);
2237 					if (error)
2238 						break;
2239 					dp += elem_size;
2240 					len -= elem_size;
2241 				}
2242 				if (op == KERN_FILE_BYFILE)
2243 					fp->f_marker = sysctl_file_marker;
2244 				needed += elem_size;
2245 				if (elem_count > 0 && elem_count != INT_MAX)
2246 					elem_count--;
2247 			}
2248 			mutex_exit(&fd->fd_lock);
2249 
2250 			/*
2251 			 * Release reference to process.
2252 			 */
2253 			mutex_enter(proc_lock);
2254 			rw_exit(&p->p_reflock);
2255 		}
2256 		if (op == KERN_FILE_BYFILE) {
2257 			sysctl_file_marker++;
2258 
2259 			/* Reset all markers if wrapped. */
2260 			if (sysctl_file_marker == 0) {
2261 				sysctl_file_marker_reset();
2262 				sysctl_file_marker++;
2263 			}
2264 		}
2265 		mutex_exit(proc_lock);
2266 		if (op == KERN_FILE_BYFILE)
2267 			mutex_exit(&sysctl_file_marker_lock);
2268 		sysctl_relock();
2269 		break;
2270 	default:
2271 		return EINVAL;
2272 	}
2273 
2274 	if (oldp == NULL)
2275 		needed += KERN_FILESLOP * elem_size;
2276 	*oldlenp = needed;
2277 
2278 	return error;
2279 }
2280 
2281 static void
2282 fill_file(struct kinfo_file *kp, const file_t *fp, const fdfile_t *ff,
2283 	  int i, pid_t pid)
2284 {
2285 
2286 	memset(kp, 0, sizeof(*kp));
2287 
2288 	kp->ki_fileaddr =	PTRTOUINT64(fp);
2289 	kp->ki_flag =		fp->f_flag;
2290 	kp->ki_iflags =		0;
2291 	kp->ki_ftype =		fp->f_type;
2292 	kp->ki_count =		fp->f_count;
2293 	kp->ki_msgcount =	fp->f_msgcount;
2294 	kp->ki_fucred =		PTRTOUINT64(fp->f_cred);
2295 	kp->ki_fuid =		kauth_cred_geteuid(fp->f_cred);
2296 	kp->ki_fgid =		kauth_cred_getegid(fp->f_cred);
2297 	kp->ki_fops =		PTRTOUINT64(fp->f_ops);
2298 	kp->ki_foffset =	fp->f_offset;
2299 	kp->ki_fdata =		PTRTOUINT64(fp->f_data);
2300 
2301 	/* vnode information to glue this file to something */
2302 	if (fp->f_type == DTYPE_VNODE) {
2303 		struct vnode *vp = (struct vnode *)fp->f_data;
2304 
2305 		kp->ki_vun =	PTRTOUINT64(vp->v_un.vu_socket);
2306 		kp->ki_vsize =	vp->v_size;
2307 		kp->ki_vtype =	vp->v_type;
2308 		kp->ki_vtag =	vp->v_tag;
2309 		kp->ki_vdata =	PTRTOUINT64(vp->v_data);
2310 	}
2311 
2312 	/* process information when retrieved via KERN_FILE_BYPID */
2313 	if (ff != NULL) {
2314 		kp->ki_pid =		pid;
2315 		kp->ki_fd =		i;
2316 		kp->ki_ofileflags =	ff->ff_exclose;
2317 		kp->ki_usecount =	ff->ff_refcnt;
2318 	}
2319 }
2320