xref: /dflybsd-src/sys/kern/kern_descrip.c (revision bfc09ba0a4d805c1860f88e64d6ae9a407d3567d)
1 /*
2  * Copyright (c) 2005 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Jeffrey Hsu.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *
35  * Copyright (c) 1982, 1986, 1989, 1991, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  * (c) UNIX System Laboratories, Inc.
38  * All or some portions of this file are derived from material licensed
39  * to the University of California by American Telephone and Telegraph
40  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41  * the permission of UNIX System Laboratories, Inc.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed by the University of
54  *	California, Berkeley and its contributors.
55  * 4. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
72  * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
73  * $DragonFly: src/sys/kern/kern_descrip.c,v 1.79 2008/08/31 13:18:28 aggelos Exp $
74  */
75 
76 #include "opt_compat.h"
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/malloc.h>
80 #include <sys/sysproto.h>
81 #include <sys/conf.h>
82 #include <sys/device.h>
83 #include <sys/filedesc.h>
84 #include <sys/kernel.h>
85 #include <sys/sysctl.h>
86 #include <sys/vnode.h>
87 #include <sys/proc.h>
88 #include <sys/nlookup.h>
89 #include <sys/file.h>
90 #include <sys/stat.h>
91 #include <sys/filio.h>
92 #include <sys/fcntl.h>
93 #include <sys/unistd.h>
94 #include <sys/resourcevar.h>
95 #include <sys/event.h>
96 #include <sys/kern_syscall.h>
97 #include <sys/kcore.h>
98 #include <sys/kinfo.h>
99 #include <sys/un.h>
100 
101 #include <vm/vm.h>
102 #include <vm/vm_extern.h>
103 
104 #include <sys/thread2.h>
105 #include <sys/file2.h>
106 #include <sys/spinlock2.h>
107 
108 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
109 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
110 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
111 static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp);
112 static void ffree(struct file *fp);
113 
114 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
115 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
116 		     "file desc to leader structures");
117 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
118 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
119 
120 static	 d_open_t  fdopen;
121 #define NUMFDESC 64
122 
123 #define CDEV_MAJOR 22
124 static struct dev_ops fildesc_ops = {
125 	{ "FD", CDEV_MAJOR, 0 },
126 	.d_open =	fdopen,
127 };
128 
129 static int badfo_readwrite (struct file *fp, struct uio *uio,
130     struct ucred *cred, int flags);
131 static int badfo_ioctl (struct file *fp, u_long com, caddr_t data,
132     struct ucred *cred);
133 static int badfo_poll (struct file *fp, int events, struct ucred *cred);
134 static int badfo_kqfilter (struct file *fp, struct knote *kn);
135 static int badfo_stat (struct file *fp, struct stat *sb, struct ucred *cred);
136 static int badfo_close (struct file *fp);
137 static int badfo_shutdown (struct file *fp, int how);
138 
139 /*
140  * Descriptor management.
141  */
142 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
143 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
144 static int nfiles;		/* actual number of open files */
145 extern int cmask;
146 
147 /*
148  * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
149  *
150  * MPSAFE - must be called with fdp->fd_spin exclusively held
151  */
152 static __inline
153 void
154 fdfixup_locked(struct filedesc *fdp, int fd)
155 {
156 	if (fd < fdp->fd_freefile) {
157 	       fdp->fd_freefile = fd;
158 	}
159 	while (fdp->fd_lastfile >= 0 &&
160 	       fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
161 	       fdp->fd_files[fdp->fd_lastfile].reserved == 0
162 	) {
163 		--fdp->fd_lastfile;
164 	}
165 }
166 
167 /*
168  * System calls on descriptors.
169  *
170  * MPSAFE
171  */
172 int
173 sys_getdtablesize(struct getdtablesize_args *uap)
174 {
175 	struct proc *p = curproc;
176 	struct plimit *limit = p->p_limit;
177 
178 	spin_lock_rd(&limit->p_spin);
179 	uap->sysmsg_result =
180 	    min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
181 	spin_unlock_rd(&limit->p_spin);
182 	return (0);
183 }
184 
185 /*
186  * Duplicate a file descriptor to a particular value.
187  *
188  * note: keep in mind that a potential race condition exists when closing
189  * descriptors from a shared descriptor table (via rfork).
190  *
191  * MPSAFE
192  */
193 int
194 sys_dup2(struct dup2_args *uap)
195 {
196 	int error;
197 	int fd = 0;
198 
199 	error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
200 	uap->sysmsg_fds[0] = fd;
201 
202 	return (error);
203 }
204 
205 /*
206  * Duplicate a file descriptor.
207  *
208  * MPSAFE
209  */
210 int
211 sys_dup(struct dup_args *uap)
212 {
213 	int error;
214 	int fd = 0;
215 
216 	error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
217 	uap->sysmsg_fds[0] = fd;
218 
219 	return (error);
220 }
221 
222 /*
223  * MPALMOSTSAFE - acquires mplock for fp operations
224  */
225 int
226 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
227 {
228 	struct thread *td = curthread;
229 	struct proc *p = td->td_proc;
230 	struct file *fp;
231 	struct vnode *vp;
232 	u_int newmin;
233 	u_int oflags;
234 	u_int nflags;
235 	int tmp, error, flg = F_POSIX;
236 
237 	KKASSERT(p);
238 
239 	/*
240 	 * Operations on file descriptors that do not require a file pointer.
241 	 */
242 	switch (cmd) {
243 	case F_GETFD:
244 		error = fgetfdflags(p->p_fd, fd, &tmp);
245 		if (error == 0)
246 			dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
247 		return (error);
248 
249 	case F_SETFD:
250 		if (dat->fc_cloexec & FD_CLOEXEC)
251 			error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
252 		else
253 			error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
254 		return (error);
255 	case F_DUPFD:
256 		newmin = dat->fc_fd;
257 		error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
258 		return (error);
259 	default:
260 		break;
261 	}
262 
263 	/*
264 	 * Operations on file pointers
265 	 */
266 	if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
267 		return (EBADF);
268 
269 	get_mplock();
270 	switch (cmd) {
271 	case F_GETFL:
272 		dat->fc_flags = OFLAGS(fp->f_flag);
273 		error = 0;
274 		break;
275 
276 	case F_SETFL:
277 		oflags = fp->f_flag;
278 		nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
279 		nflags |= oflags & ~FCNTLFLAGS;
280 
281 		error = 0;
282 		if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
283 			error = EINVAL;
284 		if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
285 			tmp = nflags & FASYNC;
286 			error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred);
287 		}
288 		if (error == 0)
289 			fp->f_flag = nflags;
290 		break;
291 
292 	case F_GETOWN:
293 		error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, cred);
294 		break;
295 
296 	case F_SETOWN:
297 		error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, cred);
298 		break;
299 
300 	case F_SETLKW:
301 		flg |= F_WAIT;
302 		/* Fall into F_SETLK */
303 
304 	case F_SETLK:
305 		if (fp->f_type != DTYPE_VNODE) {
306 			error = EBADF;
307 			break;
308 		}
309 		vp = (struct vnode *)fp->f_data;
310 
311 		/*
312 		 * copyin/lockop may block
313 		 */
314 		if (dat->fc_flock.l_whence == SEEK_CUR)
315 			dat->fc_flock.l_start += fp->f_offset;
316 
317 		switch (dat->fc_flock.l_type) {
318 		case F_RDLCK:
319 			if ((fp->f_flag & FREAD) == 0) {
320 				error = EBADF;
321 				break;
322 			}
323 			p->p_leader->p_flag |= P_ADVLOCK;
324 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
325 			    &dat->fc_flock, flg);
326 			break;
327 		case F_WRLCK:
328 			if ((fp->f_flag & FWRITE) == 0) {
329 				error = EBADF;
330 				break;
331 			}
332 			p->p_leader->p_flag |= P_ADVLOCK;
333 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
334 			    &dat->fc_flock, flg);
335 			break;
336 		case F_UNLCK:
337 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
338 				&dat->fc_flock, F_POSIX);
339 			break;
340 		default:
341 			error = EINVAL;
342 			break;
343 		}
344 
345 		/*
346 		 * It is possible to race a close() on the descriptor while
347 		 * we were blocked getting the lock.  If this occurs the
348 		 * close might not have caught the lock.
349 		 */
350 		if (checkfpclosed(p->p_fd, fd, fp)) {
351 			dat->fc_flock.l_whence = SEEK_SET;
352 			dat->fc_flock.l_start = 0;
353 			dat->fc_flock.l_len = 0;
354 			dat->fc_flock.l_type = F_UNLCK;
355 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
356 					   F_UNLCK, &dat->fc_flock, F_POSIX);
357 		}
358 		break;
359 
360 	case F_GETLK:
361 		if (fp->f_type != DTYPE_VNODE) {
362 			error = EBADF;
363 			break;
364 		}
365 		vp = (struct vnode *)fp->f_data;
366 		/*
367 		 * copyin/lockop may block
368 		 */
369 		if (dat->fc_flock.l_type != F_RDLCK &&
370 		    dat->fc_flock.l_type != F_WRLCK &&
371 		    dat->fc_flock.l_type != F_UNLCK) {
372 			error = EINVAL;
373 			break;
374 		}
375 		if (dat->fc_flock.l_whence == SEEK_CUR)
376 			dat->fc_flock.l_start += fp->f_offset;
377 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
378 			    &dat->fc_flock, F_POSIX);
379 		break;
380 	default:
381 		error = EINVAL;
382 		break;
383 	}
384 	rel_mplock();
385 
386 	fdrop(fp);
387 	return (error);
388 }
389 
390 /*
391  * The file control system call.
392  *
393  * MPSAFE
394  */
395 int
396 sys_fcntl(struct fcntl_args *uap)
397 {
398 	union fcntl_dat dat;
399 	int error;
400 
401 	switch (uap->cmd) {
402 	case F_DUPFD:
403 		dat.fc_fd = uap->arg;
404 		break;
405 	case F_SETFD:
406 		dat.fc_cloexec = uap->arg;
407 		break;
408 	case F_SETFL:
409 		dat.fc_flags = uap->arg;
410 		break;
411 	case F_SETOWN:
412 		dat.fc_owner = uap->arg;
413 		break;
414 	case F_SETLKW:
415 	case F_SETLK:
416 	case F_GETLK:
417 		error = copyin((caddr_t)uap->arg, &dat.fc_flock,
418 			       sizeof(struct flock));
419 		if (error)
420 			return (error);
421 		break;
422 	}
423 
424 	error = kern_fcntl(uap->fd, uap->cmd, &dat, curproc->p_ucred);
425 
426 	if (error == 0) {
427 		switch (uap->cmd) {
428 		case F_DUPFD:
429 			uap->sysmsg_result = dat.fc_fd;
430 			break;
431 		case F_GETFD:
432 			uap->sysmsg_result = dat.fc_cloexec;
433 			break;
434 		case F_GETFL:
435 			uap->sysmsg_result = dat.fc_flags;
436 			break;
437 		case F_GETOWN:
438 			uap->sysmsg_result = dat.fc_owner;
439 		case F_GETLK:
440 			error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
441 			    sizeof(struct flock));
442 			break;
443 		}
444 	}
445 
446 	return (error);
447 }
448 
449 /*
450  * Common code for dup, dup2, and fcntl(F_DUPFD).
451  *
452  * The type flag can be either DUP_FIXED or DUP_VARIABLE.  DUP_FIXED tells
453  * kern_dup() to destructively dup over an existing file descriptor if new
454  * is already open.  DUP_VARIABLE tells kern_dup() to find the lowest
455  * unused file descriptor that is greater than or equal to new.
456  *
457  * MPSAFE
458  */
459 int
460 kern_dup(enum dup_type type, int old, int new, int *res)
461 {
462 	struct thread *td = curthread;
463 	struct proc *p = td->td_proc;
464 	struct filedesc *fdp = p->p_fd;
465 	struct file *fp;
466 	struct file *delfp;
467 	int oldflags;
468 	int holdleaders;
469 	int error, newfd;
470 
471 	/*
472 	 * Verify that we have a valid descriptor to dup from and
473 	 * possibly to dup to.
474 	 */
475 retry:
476 	spin_lock_wr(&fdp->fd_spin);
477 	if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
478 	    new >= maxfilesperproc) {
479 		spin_unlock_wr(&fdp->fd_spin);
480 		return (EINVAL);
481 	}
482 	if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
483 		spin_unlock_wr(&fdp->fd_spin);
484 		return (EBADF);
485 	}
486 	if (type == DUP_FIXED && old == new) {
487 		*res = new;
488 		spin_unlock_wr(&fdp->fd_spin);
489 		return (0);
490 	}
491 	fp = fdp->fd_files[old].fp;
492 	oldflags = fdp->fd_files[old].fileflags;
493 	fhold(fp);	/* MPSAFE - can be called with a spinlock held */
494 
495 	/*
496 	 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
497 	 * if the requested descriptor is beyond the current table size.
498 	 *
499 	 * This can block.  Retry if the source descriptor no longer matches
500 	 * or if our expectation in the expansion case races.
501 	 *
502 	 * If we are not expanding or allocating a new decriptor, then reset
503 	 * the target descriptor to a reserved state so we have a uniform
504 	 * setup for the next code block.
505 	 */
506 	if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
507 		spin_unlock_wr(&fdp->fd_spin);
508 		error = fdalloc(p, new, &newfd);
509 		spin_lock_wr(&fdp->fd_spin);
510 		if (error) {
511 			spin_unlock_wr(&fdp->fd_spin);
512 			fdrop(fp);
513 			return (error);
514 		}
515 		/*
516 		 * Check for ripout
517 		 */
518 		if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
519 			fsetfd_locked(fdp, NULL, newfd);
520 			spin_unlock_wr(&fdp->fd_spin);
521 			fdrop(fp);
522 			goto retry;
523 		}
524 		/*
525 		 * Check for expansion race
526 		 */
527 		if (type != DUP_VARIABLE && new != newfd) {
528 			fsetfd_locked(fdp, NULL, newfd);
529 			spin_unlock_wr(&fdp->fd_spin);
530 			fdrop(fp);
531 			goto retry;
532 		}
533 		/*
534 		 * Check for ripout, newfd reused old (this case probably
535 		 * can't occur).
536 		 */
537 		if (old == newfd) {
538 			fsetfd_locked(fdp, NULL, newfd);
539 			spin_unlock_wr(&fdp->fd_spin);
540 			fdrop(fp);
541 			goto retry;
542 		}
543 		new = newfd;
544 		delfp = NULL;
545 	} else {
546 		if (fdp->fd_files[new].reserved) {
547 			spin_unlock_wr(&fdp->fd_spin);
548 			fdrop(fp);
549 			kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
550 			tsleep(fdp, 0, "fdres", hz);
551 			goto retry;
552 		}
553 
554 		/*
555 		 * If the target descriptor was never allocated we have
556 		 * to allocate it.  If it was we have to clean out the
557 		 * old descriptor.  delfp inherits the ref from the
558 		 * descriptor table.
559 		 */
560 		delfp = fdp->fd_files[new].fp;
561 		fdp->fd_files[new].fp = NULL;
562 		fdp->fd_files[new].reserved = 1;
563 		if (delfp == NULL) {
564 			fdreserve_locked(fdp, new, 1);
565 			if (new > fdp->fd_lastfile)
566 				fdp->fd_lastfile = new;
567 		}
568 
569 	}
570 
571 	/*
572 	 * NOTE: still holding an exclusive spinlock
573 	 */
574 
575 	/*
576 	 * If a descriptor is being overwritten we may hve to tell
577 	 * fdfree() to sleep to ensure that all relevant process
578 	 * leaders can be traversed in closef().
579 	 */
580 	if (delfp != NULL && p->p_fdtol != NULL) {
581 		fdp->fd_holdleaderscount++;
582 		holdleaders = 1;
583 	} else {
584 		holdleaders = 0;
585 	}
586 	KASSERT(delfp == NULL || type == DUP_FIXED,
587 		("dup() picked an open file"));
588 
589 	/*
590 	 * Duplicate the source descriptor, update lastfile.  If the new
591 	 * descriptor was not allocated and we aren't replacing an existing
592 	 * descriptor we have to mark the descriptor as being in use.
593 	 *
594 	 * The fd_files[] array inherits fp's hold reference.
595 	 */
596 	fsetfd_locked(fdp, fp, new);
597 	fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
598 	spin_unlock_wr(&fdp->fd_spin);
599 	fdrop(fp);
600 	*res = new;
601 
602 	/*
603 	 * If we dup'd over a valid file, we now own the reference to it
604 	 * and must dispose of it using closef() semantics (as if a
605 	 * close() were performed on it).
606 	 */
607 	if (delfp) {
608 		closef(delfp, p);
609 		if (holdleaders) {
610 			spin_lock_wr(&fdp->fd_spin);
611 			fdp->fd_holdleaderscount--;
612 			if (fdp->fd_holdleaderscount == 0 &&
613 			    fdp->fd_holdleaderswakeup != 0) {
614 				fdp->fd_holdleaderswakeup = 0;
615 				spin_unlock_wr(&fdp->fd_spin);
616 				wakeup(&fdp->fd_holdleaderscount);
617 			} else {
618 				spin_unlock_wr(&fdp->fd_spin);
619 			}
620 		}
621 	}
622 	return (0);
623 }
624 
625 /*
626  * If sigio is on the list associated with a process or process group,
627  * disable signalling from the device, remove sigio from the list and
628  * free sigio.
629  */
630 void
631 funsetown(struct sigio *sigio)
632 {
633 	if (sigio == NULL)
634 		return;
635 	crit_enter();
636 	*(sigio->sio_myref) = NULL;
637 	crit_exit();
638 	if (sigio->sio_pgid < 0) {
639 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
640 			     sigio, sio_pgsigio);
641 	} else /* if ((*sigiop)->sio_pgid > 0) */ {
642 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
643 			     sigio, sio_pgsigio);
644 	}
645 	crfree(sigio->sio_ucred);
646 	kfree(sigio, M_SIGIO);
647 }
648 
649 /* Free a list of sigio structures. */
650 void
651 funsetownlst(struct sigiolst *sigiolst)
652 {
653 	struct sigio *sigio;
654 
655 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
656 		funsetown(sigio);
657 }
658 
659 /*
660  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
661  *
662  * After permission checking, add a sigio structure to the sigio list for
663  * the process or process group.
664  */
665 int
666 fsetown(pid_t pgid, struct sigio **sigiop)
667 {
668 	struct proc *proc;
669 	struct pgrp *pgrp;
670 	struct sigio *sigio;
671 
672 	if (pgid == 0) {
673 		funsetown(*sigiop);
674 		return (0);
675 	}
676 	if (pgid > 0) {
677 		proc = pfind(pgid);
678 		if (proc == NULL)
679 			return (ESRCH);
680 
681 		/*
682 		 * Policy - Don't allow a process to FSETOWN a process
683 		 * in another session.
684 		 *
685 		 * Remove this test to allow maximum flexibility or
686 		 * restrict FSETOWN to the current process or process
687 		 * group for maximum safety.
688 		 */
689 		if (proc->p_session != curproc->p_session)
690 			return (EPERM);
691 
692 		pgrp = NULL;
693 	} else /* if (pgid < 0) */ {
694 		pgrp = pgfind(-pgid);
695 		if (pgrp == NULL)
696 			return (ESRCH);
697 
698 		/*
699 		 * Policy - Don't allow a process to FSETOWN a process
700 		 * in another session.
701 		 *
702 		 * Remove this test to allow maximum flexibility or
703 		 * restrict FSETOWN to the current process or process
704 		 * group for maximum safety.
705 		 */
706 		if (pgrp->pg_session != curproc->p_session)
707 			return (EPERM);
708 
709 		proc = NULL;
710 	}
711 	funsetown(*sigiop);
712 	sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
713 	if (pgid > 0) {
714 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
715 		sigio->sio_proc = proc;
716 	} else {
717 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
718 		sigio->sio_pgrp = pgrp;
719 	}
720 	sigio->sio_pgid = pgid;
721 	sigio->sio_ucred = crhold(curproc->p_ucred);
722 	/* It would be convenient if p_ruid was in ucred. */
723 	sigio->sio_ruid = curproc->p_ucred->cr_ruid;
724 	sigio->sio_myref = sigiop;
725 	crit_enter();
726 	*sigiop = sigio;
727 	crit_exit();
728 	return (0);
729 }
730 
731 /*
732  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
733  */
734 pid_t
735 fgetown(struct sigio *sigio)
736 {
737 	return (sigio != NULL ? sigio->sio_pgid : 0);
738 }
739 
740 /*
741  * Close many file descriptors.
742  *
743  * MPSAFE
744  */
745 int
746 sys_closefrom(struct closefrom_args *uap)
747 {
748 	return(kern_closefrom(uap->fd));
749 }
750 
751 /*
752  * Close all file descriptors greater then or equal to fd
753  *
754  * MPSAFE
755  */
756 int
757 kern_closefrom(int fd)
758 {
759 	struct thread *td = curthread;
760 	struct proc *p = td->td_proc;
761 	struct filedesc *fdp;
762 
763 	KKASSERT(p);
764 	fdp = p->p_fd;
765 
766 	if (fd < 0)
767 		return (EINVAL);
768 
769 	/*
770 	 * NOTE: This function will skip unassociated descriptors and
771 	 * reserved descriptors that have not yet been assigned.
772 	 * fd_lastfile can change as a side effect of kern_close().
773 	 */
774 	spin_lock_wr(&fdp->fd_spin);
775 	while (fd <= fdp->fd_lastfile) {
776 		if (fdp->fd_files[fd].fp != NULL) {
777 			spin_unlock_wr(&fdp->fd_spin);
778 			/* ok if this races another close */
779 			if (kern_close(fd) == EINTR)
780 				return (EINTR);
781 			spin_lock_wr(&fdp->fd_spin);
782 		}
783 		++fd;
784 	}
785 	spin_unlock_wr(&fdp->fd_spin);
786 	return (0);
787 }
788 
789 /*
790  * Close a file descriptor.
791  *
792  * MPSAFE
793  */
794 int
795 sys_close(struct close_args *uap)
796 {
797 	return(kern_close(uap->fd));
798 }
799 
800 /*
801  * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
802  */
803 int
804 kern_close(int fd)
805 {
806 	struct thread *td = curthread;
807 	struct proc *p = td->td_proc;
808 	struct filedesc *fdp;
809 	struct file *fp;
810 	int error;
811 	int holdleaders;
812 
813 	KKASSERT(p);
814 	fdp = p->p_fd;
815 
816 	spin_lock_wr(&fdp->fd_spin);
817 	if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
818 		spin_unlock_wr(&fdp->fd_spin);
819 		return (EBADF);
820 	}
821 	holdleaders = 0;
822 	if (p->p_fdtol != NULL) {
823 		/*
824 		 * Ask fdfree() to sleep to ensure that all relevant
825 		 * process leaders can be traversed in closef().
826 		 */
827 		fdp->fd_holdleaderscount++;
828 		holdleaders = 1;
829 	}
830 
831 	/*
832 	 * we now hold the fp reference that used to be owned by the descriptor
833 	 * array.
834 	 */
835 	spin_unlock_wr(&fdp->fd_spin);
836 	if (fd < fdp->fd_knlistsize) {
837 		get_mplock();
838 		if (fd < fdp->fd_knlistsize)
839 			knote_fdclose(p, fd);
840 		rel_mplock();
841 	}
842 	error = closef(fp, p);
843 	if (holdleaders) {
844 		spin_lock_wr(&fdp->fd_spin);
845 		fdp->fd_holdleaderscount--;
846 		if (fdp->fd_holdleaderscount == 0 &&
847 		    fdp->fd_holdleaderswakeup != 0) {
848 			fdp->fd_holdleaderswakeup = 0;
849 			spin_unlock_wr(&fdp->fd_spin);
850 			wakeup(&fdp->fd_holdleaderscount);
851 		} else {
852 			spin_unlock_wr(&fdp->fd_spin);
853 		}
854 	}
855 	return (error);
856 }
857 
858 /*
859  * shutdown_args(int fd, int how)
860  */
861 int
862 kern_shutdown(int fd, int how)
863 {
864 	struct thread *td = curthread;
865 	struct proc *p = td->td_proc;
866 	struct file *fp;
867 	int error;
868 
869 	KKASSERT(p);
870 
871 	if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
872 		return (EBADF);
873 	error = fo_shutdown(fp, how);
874 	fdrop(fp);
875 
876 	return (error);
877 }
878 
879 int
880 sys_shutdown(struct shutdown_args *uap)
881 {
882 	int error;
883 
884 	error = kern_shutdown(uap->s, uap->how);
885 
886 	return (error);
887 }
888 
889 /*
890  * MPSAFE
891  */
892 int
893 kern_fstat(int fd, struct stat *ub)
894 {
895 	struct thread *td = curthread;
896 	struct proc *p = td->td_proc;
897 	struct file *fp;
898 	int error;
899 
900 	KKASSERT(p);
901 
902 	if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
903 		return (EBADF);
904 	error = fo_stat(fp, ub, p->p_ucred);
905 	fdrop(fp);
906 
907 	return (error);
908 }
909 
910 /*
911  * Return status information about a file descriptor.
912  *
913  * MPSAFE
914  */
915 int
916 sys_fstat(struct fstat_args *uap)
917 {
918 	struct stat st;
919 	int error;
920 
921 	error = kern_fstat(uap->fd, &st);
922 
923 	if (error == 0)
924 		error = copyout(&st, uap->sb, sizeof(st));
925 	return (error);
926 }
927 
928 /*
929  * Return pathconf information about a file descriptor.
930  */
931 /* ARGSUSED */
932 int
933 sys_fpathconf(struct fpathconf_args *uap)
934 {
935 	struct thread *td = curthread;
936 	struct proc *p = td->td_proc;
937 	struct file *fp;
938 	struct vnode *vp;
939 	int error = 0;
940 
941 	KKASSERT(p);
942 
943 	if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
944 		return (EBADF);
945 
946 	switch (fp->f_type) {
947 	case DTYPE_PIPE:
948 	case DTYPE_SOCKET:
949 		if (uap->name != _PC_PIPE_BUF) {
950 			error = EINVAL;
951 		} else {
952 			uap->sysmsg_result = PIPE_BUF;
953 			error = 0;
954 		}
955 		break;
956 	case DTYPE_FIFO:
957 	case DTYPE_VNODE:
958 		vp = (struct vnode *)fp->f_data;
959 		error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
960 		break;
961 	default:
962 		error = EOPNOTSUPP;
963 		break;
964 	}
965 	fdrop(fp);
966 	return(error);
967 }
968 
969 static int fdexpand;
970 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
971 
972 /*
973  * Grow the file table so it can hold through descriptor (want).
974  *
975  * The fdp's spinlock must be held exclusively on entry and may be held
976  * exclusively on return.  The spinlock may be cycled by the routine.
977  *
978  * MPSAFE
979  */
980 static void
981 fdgrow_locked(struct filedesc *fdp, int want)
982 {
983 	struct fdnode *newfiles;
984 	struct fdnode *oldfiles;
985 	int nf, extra;
986 
987 	nf = fdp->fd_nfiles;
988 	do {
989 		/* nf has to be of the form 2^n - 1 */
990 		nf = 2 * nf + 1;
991 	} while (nf <= want);
992 
993 	spin_unlock_wr(&fdp->fd_spin);
994 	newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
995 	spin_lock_wr(&fdp->fd_spin);
996 
997 	/*
998 	 * We could have raced another extend while we were not holding
999 	 * the spinlock.
1000 	 */
1001 	if (fdp->fd_nfiles >= nf) {
1002 		spin_unlock_wr(&fdp->fd_spin);
1003 		kfree(newfiles, M_FILEDESC);
1004 		spin_lock_wr(&fdp->fd_spin);
1005 		return;
1006 	}
1007 	/*
1008 	 * Copy the existing ofile and ofileflags arrays
1009 	 * and zero the new portion of each array.
1010 	 */
1011 	extra = nf - fdp->fd_nfiles;
1012 	bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1013 	bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1014 
1015 	oldfiles = fdp->fd_files;
1016 	fdp->fd_files = newfiles;
1017 	fdp->fd_nfiles = nf;
1018 
1019 	if (oldfiles != fdp->fd_builtin_files) {
1020 		spin_unlock_wr(&fdp->fd_spin);
1021 		kfree(oldfiles, M_FILEDESC);
1022 		spin_lock_wr(&fdp->fd_spin);
1023 	}
1024 	fdexpand++;
1025 }
1026 
1027 /*
1028  * Number of nodes in right subtree, including the root.
1029  */
1030 static __inline int
1031 right_subtree_size(int n)
1032 {
1033 	return (n ^ (n | (n + 1)));
1034 }
1035 
1036 /*
1037  * Bigger ancestor.
1038  */
1039 static __inline int
1040 right_ancestor(int n)
1041 {
1042 	return (n | (n + 1));
1043 }
1044 
1045 /*
1046  * Smaller ancestor.
1047  */
1048 static __inline int
1049 left_ancestor(int n)
1050 {
1051 	return ((n & (n + 1)) - 1);
1052 }
1053 
1054 /*
1055  * Traverse the in-place binary tree buttom-up adjusting the allocation
1056  * count so scans can determine where free descriptors are located.
1057  *
1058  * MPSAFE - caller must be holding an exclusive spinlock on fdp
1059  */
1060 static
1061 void
1062 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1063 {
1064 	while (fd >= 0) {
1065 		fdp->fd_files[fd].allocated += incr;
1066 		KKASSERT(fdp->fd_files[fd].allocated >= 0);
1067 		fd = left_ancestor(fd);
1068 	}
1069 }
1070 
1071 /*
1072  * Reserve a file descriptor for the process.  If no error occurs, the
1073  * caller MUST at some point call fsetfd() or assign a file pointer
1074  * or dispose of the reservation.
1075  *
1076  * MPSAFE
1077  */
1078 int
1079 fdalloc(struct proc *p, int want, int *result)
1080 {
1081 	struct filedesc *fdp = p->p_fd;
1082 	int fd, rsize, rsum, node, lim;
1083 
1084 	spin_lock_rd(&p->p_limit->p_spin);
1085 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1086 	spin_unlock_rd(&p->p_limit->p_spin);
1087 	if (want >= lim)
1088 		return (EMFILE);
1089 	spin_lock_wr(&fdp->fd_spin);
1090 	if (want >= fdp->fd_nfiles)
1091 		fdgrow_locked(fdp, want);
1092 
1093 	/*
1094 	 * Search for a free descriptor starting at the higher
1095 	 * of want or fd_freefile.  If that fails, consider
1096 	 * expanding the ofile array.
1097 	 *
1098 	 * NOTE! the 'allocated' field is a cumulative recursive allocation
1099 	 * count.  If we happen to see a value of 0 then we can shortcut
1100 	 * our search.  Otherwise we run through through the tree going
1101 	 * down branches we know have free descriptor(s) until we hit a
1102 	 * leaf node.  The leaf node will be free but will not necessarily
1103 	 * have an allocated field of 0.
1104 	 */
1105 retry:
1106 	/* move up the tree looking for a subtree with a free node */
1107 	for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1108 	     fd = right_ancestor(fd)) {
1109 		if (fdp->fd_files[fd].allocated == 0)
1110 			goto found;
1111 
1112 		rsize = right_subtree_size(fd);
1113 		if (fdp->fd_files[fd].allocated == rsize)
1114 			continue;	/* right subtree full */
1115 
1116 		/*
1117 		 * Free fd is in the right subtree of the tree rooted at fd.
1118 		 * Call that subtree R.  Look for the smallest (leftmost)
1119 		 * subtree of R with an unallocated fd: continue moving
1120 		 * down the left branch until encountering a full left
1121 		 * subtree, then move to the right.
1122 		 */
1123 		for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1124 			node = fd + rsize;
1125 			rsum += fdp->fd_files[node].allocated;
1126 			if (fdp->fd_files[fd].allocated == rsum + rsize) {
1127 				fd = node;	/* move to the right */
1128 				if (fdp->fd_files[node].allocated == 0)
1129 					goto found;
1130 				rsum = 0;
1131 			}
1132 		}
1133 		goto found;
1134 	}
1135 
1136 	/*
1137 	 * No space in current array.  Expand?
1138 	 */
1139 	if (fdp->fd_nfiles >= lim) {
1140 		spin_unlock_wr(&fdp->fd_spin);
1141 		return (EMFILE);
1142 	}
1143 	fdgrow_locked(fdp, want);
1144 	goto retry;
1145 
1146 found:
1147 	KKASSERT(fd < fdp->fd_nfiles);
1148 	if (fd > fdp->fd_lastfile)
1149 		fdp->fd_lastfile = fd;
1150 	if (want <= fdp->fd_freefile)
1151 		fdp->fd_freefile = fd;
1152 	*result = fd;
1153 	KKASSERT(fdp->fd_files[fd].fp == NULL);
1154 	KKASSERT(fdp->fd_files[fd].reserved == 0);
1155 	fdp->fd_files[fd].fileflags = 0;
1156 	fdp->fd_files[fd].reserved = 1;
1157 	fdreserve_locked(fdp, fd, 1);
1158 	spin_unlock_wr(&fdp->fd_spin);
1159 	return (0);
1160 }
1161 
1162 /*
1163  * Check to see whether n user file descriptors
1164  * are available to the process p.
1165  *
1166  * MPSAFE
1167  */
1168 int
1169 fdavail(struct proc *p, int n)
1170 {
1171 	struct filedesc *fdp = p->p_fd;
1172 	struct fdnode *fdnode;
1173 	int i, lim, last;
1174 
1175 	spin_lock_rd(&p->p_limit->p_spin);
1176 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1177 	spin_unlock_rd(&p->p_limit->p_spin);
1178 
1179 	spin_lock_rd(&fdp->fd_spin);
1180 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1181 		spin_unlock_rd(&fdp->fd_spin);
1182 		return (1);
1183 	}
1184 	last = min(fdp->fd_nfiles, lim);
1185 	fdnode = &fdp->fd_files[fdp->fd_freefile];
1186 	for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1187 		if (fdnode->fp == NULL && --n <= 0) {
1188 			spin_unlock_rd(&fdp->fd_spin);
1189 			return (1);
1190 		}
1191 	}
1192 	spin_unlock_rd(&fdp->fd_spin);
1193 	return (0);
1194 }
1195 
1196 /*
1197  * Revoke open descriptors referencing (f_data, f_type)
1198  *
1199  * Any revoke executed within a prison is only able to
1200  * revoke descriptors for processes within that prison.
1201  *
1202  * Returns 0 on success or an error code.
1203  */
1204 struct fdrevoke_info {
1205 	void *data;
1206 	short type;
1207 	short unused;
1208 	int count;
1209 	int intransit;
1210 	struct ucred *cred;
1211 	struct file *nfp;
1212 };
1213 
1214 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1215 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1216 
1217 int
1218 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1219 {
1220 	struct fdrevoke_info info;
1221 	int error;
1222 
1223 	bzero(&info, sizeof(info));
1224 	info.data = f_data;
1225 	info.type = f_type;
1226 	info.cred = cred;
1227 	error = falloc(NULL, &info.nfp, NULL);
1228 	if (error)
1229 		return (error);
1230 
1231 	/*
1232 	 * Scan the file pointer table once.  dups do not dup file pointers,
1233 	 * only descriptors, so there is no leak.  Set FREVOKED on the fps
1234 	 * being revoked.
1235 	 */
1236 	allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1237 
1238 	/*
1239 	 * If any fps were marked track down the related descriptors
1240 	 * and close them.  Any dup()s at this point will notice
1241 	 * the FREVOKED already set in the fp and do the right thing.
1242 	 *
1243 	 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1244 	 * socket) bumped the intransit counter and will require a
1245 	 * scan.  Races against fps leaving the socket are closed by
1246 	 * the socket code checking for FREVOKED.
1247 	 */
1248 	if (info.count)
1249 		allproc_scan(fdrevoke_proc_callback, &info);
1250 	if (info.intransit)
1251 		unp_revoke_gc(info.nfp);
1252 	fdrop(info.nfp);
1253 	return(0);
1254 }
1255 
1256 /*
1257  * Locate matching file pointers directly.
1258  */
1259 static int
1260 fdrevoke_check_callback(struct file *fp, void *vinfo)
1261 {
1262 	struct fdrevoke_info *info = vinfo;
1263 
1264 	/*
1265 	 * File pointers already flagged for revokation are skipped.
1266 	 */
1267 	if (fp->f_flag & FREVOKED)
1268 		return(0);
1269 
1270 	/*
1271 	 * If revoking from a prison file pointers created outside of
1272 	 * that prison, or file pointers without creds, cannot be revoked.
1273 	 */
1274 	if (info->cred->cr_prison &&
1275 	    (fp->f_cred == NULL ||
1276 	     info->cred->cr_prison != fp->f_cred->cr_prison)) {
1277 		return(0);
1278 	}
1279 
1280 	/*
1281 	 * If the file pointer matches then mark it for revocation.  The
1282 	 * flag is currently only used by unp_revoke_gc().
1283 	 *
1284 	 * info->count is a heuristic and can race in a SMP environment.
1285 	 */
1286 	if (info->data == fp->f_data && info->type == fp->f_type) {
1287 		atomic_set_int(&fp->f_flag, FREVOKED);
1288 		info->count += fp->f_count;
1289 		if (fp->f_msgcount)
1290 			++info->intransit;
1291 	}
1292 	return(0);
1293 }
1294 
1295 /*
1296  * Locate matching file pointers via process descriptor tables.
1297  */
1298 static int
1299 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1300 {
1301 	struct fdrevoke_info *info = vinfo;
1302 	struct filedesc *fdp;
1303 	struct file *fp;
1304 	int n;
1305 
1306 	if (p->p_stat == SIDL || p->p_stat == SZOMB)
1307 		return(0);
1308 	if (info->cred->cr_prison &&
1309 	    info->cred->cr_prison != p->p_ucred->cr_prison) {
1310 		return(0);
1311 	}
1312 
1313 	/*
1314 	 * If the controlling terminal of the process matches the
1315 	 * vnode being revoked we clear the controlling terminal.
1316 	 *
1317 	 * The normal spec_close() may not catch this because it
1318 	 * uses curproc instead of p.
1319 	 */
1320 	if (p->p_session && info->type == DTYPE_VNODE &&
1321 	    info->data == p->p_session->s_ttyvp) {
1322 		p->p_session->s_ttyvp = NULL;
1323 		vrele(info->data);
1324 	}
1325 
1326 	/*
1327 	 * Softref the fdp to prevent it from being destroyed
1328 	 */
1329 	spin_lock_wr(&p->p_spin);
1330 	if ((fdp = p->p_fd) == NULL) {
1331 		spin_unlock_wr(&p->p_spin);
1332 		return(0);
1333 	}
1334 	atomic_add_int(&fdp->fd_softrefs, 1);
1335 	spin_unlock_wr(&p->p_spin);
1336 
1337 	/*
1338 	 * Locate and close any matching file descriptors.
1339 	 */
1340 	spin_lock_wr(&fdp->fd_spin);
1341 	for (n = 0; n < fdp->fd_nfiles; ++n) {
1342 		if ((fp = fdp->fd_files[n].fp) == NULL)
1343 			continue;
1344 		if (fp->f_flag & FREVOKED) {
1345 			fhold(info->nfp);
1346 			fdp->fd_files[n].fp = info->nfp;
1347 			spin_unlock_wr(&fdp->fd_spin);
1348 			closef(fp, p);
1349 			spin_lock_wr(&fdp->fd_spin);
1350 			--info->count;
1351 		}
1352 	}
1353 	spin_unlock_wr(&fdp->fd_spin);
1354 	atomic_subtract_int(&fdp->fd_softrefs, 1);
1355 	return(0);
1356 }
1357 
1358 /*
1359  * falloc:
1360  *	Create a new open file structure and reserve a file decriptor
1361  *	for the process that refers to it.
1362  *
1363  *	Root creds are checked using p, or assumed if p is NULL.  If
1364  *	resultfd is non-NULL then p must also be non-NULL.  No file
1365  *	descriptor is reserved if resultfd is NULL.
1366  *
1367  *	A file pointer with a refcount of 1 is returned.  Note that the
1368  *	file pointer is NOT associated with the descriptor.  If falloc
1369  *	returns success, fsetfd() MUST be called to either associate the
1370  *	file pointer or clear the reservation.
1371  *
1372  * MPSAFE
1373  */
1374 int
1375 falloc(struct proc *p, struct file **resultfp, int *resultfd)
1376 {
1377 	static struct timeval lastfail;
1378 	static int curfail;
1379 	struct file *fp;
1380 	int error;
1381 
1382 	fp = NULL;
1383 
1384 	/*
1385 	 * Handle filetable full issues and root overfill.
1386 	 */
1387 	if (nfiles >= maxfiles - maxfilesrootres &&
1388 	    ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) {
1389 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1390 			kprintf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n",
1391 				(p ? p->p_ucred->cr_ruid : -1));
1392 		}
1393 		error = ENFILE;
1394 		goto done;
1395 	}
1396 
1397 	/*
1398 	 * Allocate a new file descriptor.
1399 	 */
1400 	fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1401 	spin_init(&fp->f_spin);
1402 	fp->f_count = 1;
1403 	fp->f_ops = &badfileops;
1404 	fp->f_seqcount = 1;
1405 	if (p)
1406 		fp->f_cred = crhold(p->p_ucred);
1407 	else
1408 		fp->f_cred = crhold(proc0.p_ucred);
1409 	spin_lock_wr(&filehead_spin);
1410 	nfiles++;
1411 	LIST_INSERT_HEAD(&filehead, fp, f_list);
1412 	spin_unlock_wr(&filehead_spin);
1413 	if (resultfd) {
1414 		if ((error = fdalloc(p, 0, resultfd)) != 0) {
1415 			fdrop(fp);
1416 			fp = NULL;
1417 		}
1418 	} else {
1419 		error = 0;
1420 	}
1421 done:
1422 	*resultfp = fp;
1423 	return (error);
1424 }
1425 
1426 /*
1427  * MPSAFE
1428  */
1429 static
1430 int
1431 checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1432 {
1433 	int error;
1434 
1435 	spin_lock_rd(&fdp->fd_spin);
1436 	if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1437 		error = EBADF;
1438 	else
1439 		error = 0;
1440 	spin_unlock_rd(&fdp->fd_spin);
1441 	return (error);
1442 }
1443 
1444 /*
1445  * Associate a file pointer with a previously reserved file descriptor.
1446  * This function always succeeds.
1447  *
1448  * If fp is NULL, the file descriptor is returned to the pool.
1449  */
1450 
1451 /*
1452  * MPSAFE (exclusive spinlock must be held on call)
1453  */
1454 static void
1455 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1456 {
1457 	KKASSERT((unsigned)fd < fdp->fd_nfiles);
1458 	KKASSERT(fdp->fd_files[fd].reserved != 0);
1459 	if (fp) {
1460 		fhold(fp);
1461 		fdp->fd_files[fd].fp = fp;
1462 		fdp->fd_files[fd].reserved = 0;
1463 		if (fp->f_type == DTYPE_KQUEUE) {
1464 			if (fdp->fd_knlistsize < 0)
1465 				fdp->fd_knlistsize = 0;
1466 		}
1467 	} else {
1468 		fdp->fd_files[fd].reserved = 0;
1469 		fdreserve_locked(fdp, fd, -1);
1470 		fdfixup_locked(fdp, fd);
1471 	}
1472 }
1473 
1474 /*
1475  * MPSAFE
1476  */
1477 void
1478 fsetfd(struct proc *p, struct file *fp, int fd)
1479 {
1480 	struct filedesc *fdp = p->p_fd;
1481 
1482 	spin_lock_wr(&fdp->fd_spin);
1483 	fsetfd_locked(fdp, fp, fd);
1484 	spin_unlock_wr(&fdp->fd_spin);
1485 }
1486 
1487 /*
1488  * MPSAFE (exclusive spinlock must be held on call)
1489  */
1490 static
1491 struct file *
1492 funsetfd_locked(struct filedesc *fdp, int fd)
1493 {
1494 	struct file *fp;
1495 
1496 	if ((unsigned)fd >= fdp->fd_nfiles)
1497 		return (NULL);
1498 	if ((fp = fdp->fd_files[fd].fp) == NULL)
1499 		return (NULL);
1500 	fdp->fd_files[fd].fp = NULL;
1501 	fdp->fd_files[fd].fileflags = 0;
1502 
1503 	fdreserve_locked(fdp, fd, -1);
1504 	fdfixup_locked(fdp, fd);
1505 	return(fp);
1506 }
1507 
1508 /*
1509  * MPSAFE
1510  */
1511 int
1512 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1513 {
1514 	int error;
1515 
1516 	spin_lock_rd(&fdp->fd_spin);
1517 	if (((u_int)fd) >= fdp->fd_nfiles) {
1518 		error = EBADF;
1519 	} else if (fdp->fd_files[fd].fp == NULL) {
1520 		error = EBADF;
1521 	} else {
1522 		*flagsp = fdp->fd_files[fd].fileflags;
1523 		error = 0;
1524 	}
1525 	spin_unlock_rd(&fdp->fd_spin);
1526 	return (error);
1527 }
1528 
1529 /*
1530  * MPSAFE
1531  */
1532 int
1533 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1534 {
1535 	int error;
1536 
1537 	spin_lock_wr(&fdp->fd_spin);
1538 	if (((u_int)fd) >= fdp->fd_nfiles) {
1539 		error = EBADF;
1540 	} else if (fdp->fd_files[fd].fp == NULL) {
1541 		error = EBADF;
1542 	} else {
1543 		fdp->fd_files[fd].fileflags |= add_flags;
1544 		error = 0;
1545 	}
1546 	spin_unlock_wr(&fdp->fd_spin);
1547 	return (error);
1548 }
1549 
1550 /*
1551  * MPSAFE
1552  */
1553 int
1554 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1555 {
1556 	int error;
1557 
1558 	spin_lock_wr(&fdp->fd_spin);
1559 	if (((u_int)fd) >= fdp->fd_nfiles) {
1560 		error = EBADF;
1561 	} else if (fdp->fd_files[fd].fp == NULL) {
1562 		error = EBADF;
1563 	} else {
1564 		fdp->fd_files[fd].fileflags &= ~rem_flags;
1565 		error = 0;
1566 	}
1567 	spin_unlock_wr(&fdp->fd_spin);
1568 	return (error);
1569 }
1570 
1571 void
1572 fsetcred(struct file *fp, struct ucred *cr)
1573 {
1574 	crhold(cr);
1575 	crfree(fp->f_cred);
1576 	fp->f_cred = cr;
1577 }
1578 
1579 /*
1580  * Free a file descriptor.
1581  */
1582 static
1583 void
1584 ffree(struct file *fp)
1585 {
1586 	KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1587 	spin_lock_wr(&filehead_spin);
1588 	LIST_REMOVE(fp, f_list);
1589 	nfiles--;
1590 	spin_unlock_wr(&filehead_spin);
1591 	crfree(fp->f_cred);
1592 	if (fp->f_nchandle.ncp)
1593 	    cache_drop(&fp->f_nchandle);
1594 	kfree(fp, M_FILE);
1595 }
1596 
1597 /*
1598  * called from init_main, initialize filedesc0 for proc0.
1599  */
1600 void
1601 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1602 {
1603 	p0->p_fd = fdp0;
1604 	p0->p_fdtol = NULL;
1605 	fdp0->fd_refcnt = 1;
1606 	fdp0->fd_cmask = cmask;
1607 	fdp0->fd_files = fdp0->fd_builtin_files;
1608 	fdp0->fd_nfiles = NDFILE;
1609 	fdp0->fd_lastfile = -1;
1610 	spin_init(&fdp0->fd_spin);
1611 }
1612 
1613 /*
1614  * Build a new filedesc structure.
1615  *
1616  * NOT MPSAFE (vref)
1617  */
1618 struct filedesc *
1619 fdinit(struct proc *p)
1620 {
1621 	struct filedesc *newfdp;
1622 	struct filedesc *fdp = p->p_fd;
1623 
1624 	newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1625 	spin_lock_rd(&fdp->fd_spin);
1626 	if (fdp->fd_cdir) {
1627 		newfdp->fd_cdir = fdp->fd_cdir;
1628 		vref(newfdp->fd_cdir);
1629 		cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1630 	}
1631 
1632 	/*
1633 	 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1634 	 * proc0, but should unconditionally exist in other processes.
1635 	 */
1636 	if (fdp->fd_rdir) {
1637 		newfdp->fd_rdir = fdp->fd_rdir;
1638 		vref(newfdp->fd_rdir);
1639 		cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1640 	}
1641 	if (fdp->fd_jdir) {
1642 		newfdp->fd_jdir = fdp->fd_jdir;
1643 		vref(newfdp->fd_jdir);
1644 		cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1645 	}
1646 	spin_unlock_rd(&fdp->fd_spin);
1647 
1648 	/* Create the file descriptor table. */
1649 	newfdp->fd_refcnt = 1;
1650 	newfdp->fd_cmask = cmask;
1651 	newfdp->fd_files = newfdp->fd_builtin_files;
1652 	newfdp->fd_nfiles = NDFILE;
1653 	newfdp->fd_knlistsize = -1;
1654 	newfdp->fd_lastfile = -1;
1655 	spin_init(&newfdp->fd_spin);
1656 
1657 	return (newfdp);
1658 }
1659 
1660 /*
1661  * Share a filedesc structure.
1662  *
1663  * MPSAFE
1664  */
1665 struct filedesc *
1666 fdshare(struct proc *p)
1667 {
1668 	struct filedesc *fdp;
1669 
1670 	fdp = p->p_fd;
1671 	spin_lock_wr(&fdp->fd_spin);
1672 	fdp->fd_refcnt++;
1673 	spin_unlock_wr(&fdp->fd_spin);
1674 	return (fdp);
1675 }
1676 
1677 /*
1678  * Copy a filedesc structure.
1679  *
1680  * MPSAFE
1681  */
1682 struct filedesc *
1683 fdcopy(struct proc *p)
1684 {
1685 	struct filedesc *fdp = p->p_fd;
1686 	struct filedesc *newfdp;
1687 	struct fdnode *fdnode;
1688 	int i;
1689 	int ni;
1690 
1691 	/*
1692 	 * Certain daemons might not have file descriptors.
1693 	 */
1694 	if (fdp == NULL)
1695 		return (NULL);
1696 
1697 	/*
1698 	 * Allocate the new filedesc and fd_files[] array.  This can race
1699 	 * with operations by other threads on the fdp so we have to be
1700 	 * careful.
1701 	 */
1702 	newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1703 again:
1704 	spin_lock_rd(&fdp->fd_spin);
1705 	if (fdp->fd_lastfile < NDFILE) {
1706 		newfdp->fd_files = newfdp->fd_builtin_files;
1707 		i = NDFILE;
1708 	} else {
1709 		/*
1710 		 * We have to allocate (N^2-1) entries for our in-place
1711 		 * binary tree.  Allow the table to shrink.
1712 		 */
1713 		i = fdp->fd_nfiles;
1714 		ni = (i - 1) / 2;
1715 		while (ni > fdp->fd_lastfile && ni > NDFILE) {
1716 			i = ni;
1717 			ni = (i - 1) / 2;
1718 		}
1719 		spin_unlock_rd(&fdp->fd_spin);
1720 		newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1721 					  M_FILEDESC, M_WAITOK | M_ZERO);
1722 
1723 		/*
1724 		 * Check for race, retry
1725 		 */
1726 		spin_lock_rd(&fdp->fd_spin);
1727 		if (i <= fdp->fd_lastfile) {
1728 			spin_unlock_rd(&fdp->fd_spin);
1729 			kfree(newfdp->fd_files, M_FILEDESC);
1730 			goto again;
1731 		}
1732 	}
1733 
1734 	/*
1735 	 * Dup the remaining fields. vref() and cache_hold() can be
1736 	 * safely called while holding the read spinlock on fdp.
1737 	 *
1738 	 * The read spinlock on fdp is still being held.
1739 	 *
1740 	 * NOTE: vref and cache_hold calls for the case where the vnode
1741 	 * or cache entry already has at least one ref may be called
1742 	 * while holding spin locks.
1743 	 */
1744 	if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1745 		vref(newfdp->fd_cdir);
1746 		cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1747 	}
1748 	/*
1749 	 * We must check for fd_rdir here, at least for now because
1750 	 * the init process is created before we have access to the
1751 	 * rootvode to take a reference to it.
1752 	 */
1753 	if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1754 		vref(newfdp->fd_rdir);
1755 		cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1756 	}
1757 	if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1758 		vref(newfdp->fd_jdir);
1759 		cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1760 	}
1761 	newfdp->fd_refcnt = 1;
1762 	newfdp->fd_nfiles = i;
1763 	newfdp->fd_lastfile = fdp->fd_lastfile;
1764 	newfdp->fd_freefile = fdp->fd_freefile;
1765 	newfdp->fd_cmask = fdp->fd_cmask;
1766 	newfdp->fd_knlist = NULL;
1767 	newfdp->fd_knlistsize = -1;
1768 	newfdp->fd_knhash = NULL;
1769 	newfdp->fd_knhashmask = 0;
1770 	spin_init(&newfdp->fd_spin);
1771 
1772 	/*
1773 	 * Copy the descriptor table through (i).  This also copies the
1774 	 * allocation state.   Then go through and ref the file pointers
1775 	 * and clean up any KQ descriptors.
1776 	 *
1777 	 * kq descriptors cannot be copied.  Since we haven't ref'd the
1778 	 * copied files yet we can ignore the return value from funsetfd().
1779 	 *
1780 	 * The read spinlock on fdp is still being held.
1781 	 */
1782 	bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1783 	for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1784 		fdnode = &newfdp->fd_files[i];
1785 		if (fdnode->reserved) {
1786 			fdreserve_locked(newfdp, i, -1);
1787 			fdnode->reserved = 0;
1788 			fdfixup_locked(newfdp, i);
1789 		} else if (fdnode->fp) {
1790 			if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1791 				(void)funsetfd_locked(newfdp, i);
1792 			} else {
1793 				fhold(fdnode->fp);
1794 			}
1795 		}
1796 	}
1797 	spin_unlock_rd(&fdp->fd_spin);
1798 	return (newfdp);
1799 }
1800 
1801 /*
1802  * Release a filedesc structure.
1803  *
1804  * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1805  */
1806 void
1807 fdfree(struct proc *p, struct filedesc *repl)
1808 {
1809 	struct filedesc *fdp;
1810 	struct fdnode *fdnode;
1811 	int i;
1812 	struct filedesc_to_leader *fdtol;
1813 	struct file *fp;
1814 	struct vnode *vp;
1815 	struct flock lf;
1816 
1817 	/*
1818 	 * Certain daemons might not have file descriptors.
1819 	 */
1820 	fdp = p->p_fd;
1821 	if (fdp == NULL) {
1822 		p->p_fd = repl;
1823 		return;
1824 	}
1825 
1826 	/*
1827 	 * Severe messing around to follow.
1828 	 */
1829 	spin_lock_wr(&fdp->fd_spin);
1830 
1831 	/* Check for special need to clear POSIX style locks */
1832 	fdtol = p->p_fdtol;
1833 	if (fdtol != NULL) {
1834 		KASSERT(fdtol->fdl_refcount > 0,
1835 			("filedesc_to_refcount botch: fdl_refcount=%d",
1836 			 fdtol->fdl_refcount));
1837 		if (fdtol->fdl_refcount == 1 &&
1838 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1839 			for (i = 0; i <= fdp->fd_lastfile; ++i) {
1840 				fdnode = &fdp->fd_files[i];
1841 				if (fdnode->fp == NULL ||
1842 				    fdnode->fp->f_type != DTYPE_VNODE) {
1843 					continue;
1844 				}
1845 				fp = fdnode->fp;
1846 				fhold(fp);
1847 				spin_unlock_wr(&fdp->fd_spin);
1848 
1849 				lf.l_whence = SEEK_SET;
1850 				lf.l_start = 0;
1851 				lf.l_len = 0;
1852 				lf.l_type = F_UNLCK;
1853 				vp = (struct vnode *)fp->f_data;
1854 				(void) VOP_ADVLOCK(vp,
1855 						   (caddr_t)p->p_leader,
1856 						   F_UNLCK,
1857 						   &lf,
1858 						   F_POSIX);
1859 				fdrop(fp);
1860 				spin_lock_wr(&fdp->fd_spin);
1861 			}
1862 		}
1863 	retry:
1864 		if (fdtol->fdl_refcount == 1) {
1865 			if (fdp->fd_holdleaderscount > 0 &&
1866 			    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1867 				/*
1868 				 * close() or do_dup() has cleared a reference
1869 				 * in a shared file descriptor table.
1870 				 */
1871 				fdp->fd_holdleaderswakeup = 1;
1872 				ssleep(&fdp->fd_holdleaderscount,
1873 				       &fdp->fd_spin, 0, "fdlhold", 0);
1874 				goto retry;
1875 			}
1876 			if (fdtol->fdl_holdcount > 0) {
1877 				/*
1878 				 * Ensure that fdtol->fdl_leader
1879 				 * remains valid in closef().
1880 				 */
1881 				fdtol->fdl_wakeup = 1;
1882 				ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1883 				goto retry;
1884 			}
1885 		}
1886 		fdtol->fdl_refcount--;
1887 		if (fdtol->fdl_refcount == 0 &&
1888 		    fdtol->fdl_holdcount == 0) {
1889 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1890 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1891 		} else {
1892 			fdtol = NULL;
1893 		}
1894 		p->p_fdtol = NULL;
1895 		if (fdtol != NULL) {
1896 			spin_unlock_wr(&fdp->fd_spin);
1897 			kfree(fdtol, M_FILEDESC_TO_LEADER);
1898 			spin_lock_wr(&fdp->fd_spin);
1899 		}
1900 	}
1901 	if (--fdp->fd_refcnt > 0) {
1902 		spin_unlock_wr(&fdp->fd_spin);
1903 		spin_lock_wr(&p->p_spin);
1904 		p->p_fd = repl;
1905 		spin_unlock_wr(&p->p_spin);
1906 		return;
1907 	}
1908 
1909 	/*
1910 	 * Even though we are the last reference to the structure allproc
1911 	 * scans may still reference the structure.  Maintain proper
1912 	 * locks until we can replace p->p_fd.
1913 	 *
1914 	 * Also note that kqueue's closef still needs to reference the
1915 	 * fdp via p->p_fd, so we have to close the descriptors before
1916 	 * we replace p->p_fd.
1917 	 */
1918 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1919 		if (fdp->fd_files[i].fp) {
1920 			fp = funsetfd_locked(fdp, i);
1921 			if (fp) {
1922 				spin_unlock_wr(&fdp->fd_spin);
1923 				closef(fp, p);
1924 				spin_lock_wr(&fdp->fd_spin);
1925 			}
1926 		}
1927 	}
1928 	spin_unlock_wr(&fdp->fd_spin);
1929 
1930 	/*
1931 	 * Interlock against an allproc scan operations (typically frevoke).
1932 	 */
1933 	spin_lock_wr(&p->p_spin);
1934 	p->p_fd = repl;
1935 	spin_unlock_wr(&p->p_spin);
1936 
1937 	/*
1938 	 * Wait for any softrefs to go away.  This race rarely occurs so
1939 	 * we can use a non-critical-path style poll/sleep loop.  The
1940 	 * race only occurs against allproc scans.
1941 	 *
1942 	 * No new softrefs can occur with the fdp disconnected from the
1943 	 * process.
1944 	 */
1945 	if (fdp->fd_softrefs) {
1946 		kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
1947 		while (fdp->fd_softrefs)
1948 			tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
1949 	}
1950 
1951 	if (fdp->fd_files != fdp->fd_builtin_files)
1952 		kfree(fdp->fd_files, M_FILEDESC);
1953 	if (fdp->fd_cdir) {
1954 		cache_drop(&fdp->fd_ncdir);
1955 		vrele(fdp->fd_cdir);
1956 	}
1957 	if (fdp->fd_rdir) {
1958 		cache_drop(&fdp->fd_nrdir);
1959 		vrele(fdp->fd_rdir);
1960 	}
1961 	if (fdp->fd_jdir) {
1962 		cache_drop(&fdp->fd_njdir);
1963 		vrele(fdp->fd_jdir);
1964 	}
1965 	if (fdp->fd_knlist)
1966 		kfree(fdp->fd_knlist, M_KQUEUE);
1967 	if (fdp->fd_knhash)
1968 		kfree(fdp->fd_knhash, M_KQUEUE);
1969 	kfree(fdp, M_FILEDESC);
1970 }
1971 
1972 /*
1973  * Retrieve and reference the file pointer associated with a descriptor.
1974  *
1975  * MPSAFE
1976  */
1977 struct file *
1978 holdfp(struct filedesc *fdp, int fd, int flag)
1979 {
1980 	struct file* fp;
1981 
1982 	spin_lock_rd(&fdp->fd_spin);
1983 	if (((u_int)fd) >= fdp->fd_nfiles) {
1984 		fp = NULL;
1985 		goto done;
1986 	}
1987 	if ((fp = fdp->fd_files[fd].fp) == NULL)
1988 		goto done;
1989 	if ((fp->f_flag & flag) == 0 && flag != -1) {
1990 		fp = NULL;
1991 		goto done;
1992 	}
1993 	fhold(fp);
1994 done:
1995 	spin_unlock_rd(&fdp->fd_spin);
1996 	return (fp);
1997 }
1998 
1999 /*
2000  * holdsock() - load the struct file pointer associated
2001  * with a socket into *fpp.  If an error occurs, non-zero
2002  * will be returned and *fpp will be set to NULL.
2003  *
2004  * MPSAFE
2005  */
2006 int
2007 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2008 {
2009 	struct file *fp;
2010 	int error;
2011 
2012 	spin_lock_rd(&fdp->fd_spin);
2013 	if ((unsigned)fd >= fdp->fd_nfiles) {
2014 		error = EBADF;
2015 		fp = NULL;
2016 		goto done;
2017 	}
2018 	if ((fp = fdp->fd_files[fd].fp) == NULL) {
2019 		error = EBADF;
2020 		goto done;
2021 	}
2022 	if (fp->f_type != DTYPE_SOCKET) {
2023 		error = ENOTSOCK;
2024 		goto done;
2025 	}
2026 	fhold(fp);
2027 	error = 0;
2028 done:
2029 	spin_unlock_rd(&fdp->fd_spin);
2030 	*fpp = fp;
2031 	return (error);
2032 }
2033 
2034 /*
2035  * Convert a user file descriptor to a held file pointer.
2036  *
2037  * MPSAFE
2038  */
2039 int
2040 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2041 {
2042 	struct file *fp;
2043 	int error;
2044 
2045 	spin_lock_rd(&fdp->fd_spin);
2046 	if ((unsigned)fd >= fdp->fd_nfiles) {
2047 		error = EBADF;
2048 		fp = NULL;
2049 		goto done;
2050 	}
2051 	if ((fp = fdp->fd_files[fd].fp) == NULL) {
2052 		error = EBADF;
2053 		goto done;
2054 	}
2055 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2056 		fp = NULL;
2057 		error = EINVAL;
2058 		goto done;
2059 	}
2060 	fhold(fp);
2061 	error = 0;
2062 done:
2063 	spin_unlock_rd(&fdp->fd_spin);
2064 	*fpp = fp;
2065 	return (error);
2066 }
2067 
2068 /*
2069  * For setugid programs, we don't want to people to use that setugidness
2070  * to generate error messages which write to a file which otherwise would
2071  * otherwise be off-limits to the process.
2072  *
2073  * This is a gross hack to plug the hole.  A better solution would involve
2074  * a special vop or other form of generalized access control mechanism.  We
2075  * go ahead and just reject all procfs file systems accesses as dangerous.
2076  *
2077  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2078  * sufficient.  We also don't for check setugidness since we know we are.
2079  */
2080 static int
2081 is_unsafe(struct file *fp)
2082 {
2083 	if (fp->f_type == DTYPE_VNODE &&
2084 	    ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2085 		return (1);
2086 	return (0);
2087 }
2088 
2089 /*
2090  * Make this setguid thing safe, if at all possible.
2091  *
2092  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2093  */
2094 void
2095 setugidsafety(struct proc *p)
2096 {
2097 	struct filedesc *fdp = p->p_fd;
2098 	int i;
2099 
2100 	/* Certain daemons might not have file descriptors. */
2101 	if (fdp == NULL)
2102 		return;
2103 
2104 	/*
2105 	 * note: fdp->fd_files may be reallocated out from under us while
2106 	 * we are blocked in a close.  Be careful!
2107 	 */
2108 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2109 		if (i > 2)
2110 			break;
2111 		if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2112 			struct file *fp;
2113 
2114 			if (i < fdp->fd_knlistsize)
2115 				knote_fdclose(p, i);
2116 			/*
2117 			 * NULL-out descriptor prior to close to avoid
2118 			 * a race while close blocks.
2119 			 */
2120 			if ((fp = funsetfd_locked(fdp, i)) != NULL)
2121 				closef(fp, p);
2122 		}
2123 	}
2124 }
2125 
2126 /*
2127  * Close any files on exec?
2128  *
2129  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2130  */
2131 void
2132 fdcloseexec(struct proc *p)
2133 {
2134 	struct filedesc *fdp = p->p_fd;
2135 	int i;
2136 
2137 	/* Certain daemons might not have file descriptors. */
2138 	if (fdp == NULL)
2139 		return;
2140 
2141 	/*
2142 	 * We cannot cache fd_files since operations may block and rip
2143 	 * them out from under us.
2144 	 */
2145 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2146 		if (fdp->fd_files[i].fp != NULL &&
2147 		    (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2148 			struct file *fp;
2149 
2150 			if (i < fdp->fd_knlistsize)
2151 				knote_fdclose(p, i);
2152 			/*
2153 			 * NULL-out descriptor prior to close to avoid
2154 			 * a race while close blocks.
2155 			 */
2156 			if ((fp = funsetfd_locked(fdp, i)) != NULL)
2157 				closef(fp, p);
2158 		}
2159 	}
2160 }
2161 
2162 /*
2163  * It is unsafe for set[ug]id processes to be started with file
2164  * descriptors 0..2 closed, as these descriptors are given implicit
2165  * significance in the Standard C library.  fdcheckstd() will create a
2166  * descriptor referencing /dev/null for each of stdin, stdout, and
2167  * stderr that is not already open.
2168  *
2169  * NOT MPSAFE - calls falloc, vn_open, etc
2170  */
2171 int
2172 fdcheckstd(struct proc *p)
2173 {
2174 	struct nlookupdata nd;
2175 	struct filedesc *fdp;
2176 	struct file *fp;
2177 	int retval;
2178 	int i, error, flags, devnull;
2179 
2180 	fdp = p->p_fd;
2181 	if (fdp == NULL)
2182 		return (0);
2183 	devnull = -1;
2184 	error = 0;
2185 	for (i = 0; i < 3; i++) {
2186 		if (fdp->fd_files[i].fp != NULL)
2187 			continue;
2188 		if (devnull < 0) {
2189 			if ((error = falloc(p, &fp, &devnull)) != 0)
2190 				break;
2191 
2192 			error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2193 						NLC_FOLLOW|NLC_LOCKVP);
2194 			flags = FREAD | FWRITE;
2195 			if (error == 0)
2196 				error = vn_open(&nd, fp, flags, 0);
2197 			if (error == 0)
2198 				fsetfd(p, fp, devnull);
2199 			else
2200 				fsetfd(p, NULL, devnull);
2201 			fdrop(fp);
2202 			nlookup_done(&nd);
2203 			if (error)
2204 				break;
2205 			KKASSERT(i == devnull);
2206 		} else {
2207 			error = kern_dup(DUP_FIXED, devnull, i, &retval);
2208 			if (error != 0)
2209 				break;
2210 		}
2211 	}
2212 	return (error);
2213 }
2214 
2215 /*
2216  * Internal form of close.
2217  * Decrement reference count on file structure.
2218  * Note: td and/or p may be NULL when closing a file
2219  * that was being passed in a message.
2220  *
2221  * MPALMOSTSAFE - acquires mplock for VOP operations
2222  */
2223 int
2224 closef(struct file *fp, struct proc *p)
2225 {
2226 	struct vnode *vp;
2227 	struct flock lf;
2228 	struct filedesc_to_leader *fdtol;
2229 
2230 	if (fp == NULL)
2231 		return (0);
2232 
2233 	/*
2234 	 * POSIX record locking dictates that any close releases ALL
2235 	 * locks owned by this process.  This is handled by setting
2236 	 * a flag in the unlock to free ONLY locks obeying POSIX
2237 	 * semantics, and not to free BSD-style file locks.
2238 	 * If the descriptor was in a message, POSIX-style locks
2239 	 * aren't passed with the descriptor.
2240 	 */
2241 	if (p != NULL && fp->f_type == DTYPE_VNODE &&
2242 	    (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2243 	) {
2244 		get_mplock();
2245 		if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2246 			lf.l_whence = SEEK_SET;
2247 			lf.l_start = 0;
2248 			lf.l_len = 0;
2249 			lf.l_type = F_UNLCK;
2250 			vp = (struct vnode *)fp->f_data;
2251 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2252 					   &lf, F_POSIX);
2253 		}
2254 		fdtol = p->p_fdtol;
2255 		if (fdtol != NULL) {
2256 			/*
2257 			 * Handle special case where file descriptor table
2258 			 * is shared between multiple process leaders.
2259 			 */
2260 			for (fdtol = fdtol->fdl_next;
2261 			     fdtol != p->p_fdtol;
2262 			     fdtol = fdtol->fdl_next) {
2263 				if ((fdtol->fdl_leader->p_flag &
2264 				     P_ADVLOCK) == 0)
2265 					continue;
2266 				fdtol->fdl_holdcount++;
2267 				lf.l_whence = SEEK_SET;
2268 				lf.l_start = 0;
2269 				lf.l_len = 0;
2270 				lf.l_type = F_UNLCK;
2271 				vp = (struct vnode *)fp->f_data;
2272 				(void) VOP_ADVLOCK(vp,
2273 						   (caddr_t)fdtol->fdl_leader,
2274 						   F_UNLCK, &lf, F_POSIX);
2275 				fdtol->fdl_holdcount--;
2276 				if (fdtol->fdl_holdcount == 0 &&
2277 				    fdtol->fdl_wakeup != 0) {
2278 					fdtol->fdl_wakeup = 0;
2279 					wakeup(fdtol);
2280 				}
2281 			}
2282 		}
2283 		rel_mplock();
2284 	}
2285 	return (fdrop(fp));
2286 }
2287 
2288 /*
2289  * MPSAFE
2290  *
2291  * fhold() can only be called if f_count is already at least 1 (i.e. the
2292  * caller of fhold() already has a reference to the file pointer in some
2293  * manner or other).
2294  *
2295  * f_count is not spin-locked.  Instead, atomic ops are used for
2296  * incrementing, decrementing, and handling the 1->0 transition.
2297  */
2298 void
2299 fhold(struct file *fp)
2300 {
2301 	atomic_add_int(&fp->f_count, 1);
2302 }
2303 
2304 /*
2305  * fdrop() - drop a reference to a descriptor
2306  *
2307  * MPALMOSTSAFE - acquires mplock for final close sequence
2308  */
2309 int
2310 fdrop(struct file *fp)
2311 {
2312 	struct flock lf;
2313 	struct vnode *vp;
2314 	int error;
2315 
2316 	/*
2317 	 * A combined fetch and subtract is needed to properly detect
2318 	 * 1->0 transitions, otherwise two cpus dropping from a ref
2319 	 * count of 2 might both try to run the 1->0 code.
2320 	 */
2321 	if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2322 		return (0);
2323 
2324 	get_mplock();
2325 
2326 	/*
2327 	 * The last reference has gone away, we own the fp structure free
2328 	 * and clear.
2329 	 */
2330 	if (fp->f_count < 0)
2331 		panic("fdrop: count < 0");
2332 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2333 	    (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2334 	) {
2335 		lf.l_whence = SEEK_SET;
2336 		lf.l_start = 0;
2337 		lf.l_len = 0;
2338 		lf.l_type = F_UNLCK;
2339 		vp = (struct vnode *)fp->f_data;
2340 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2341 	}
2342 	if (fp->f_ops != &badfileops)
2343 		error = fo_close(fp);
2344 	else
2345 		error = 0;
2346 	ffree(fp);
2347 	rel_mplock();
2348 	return (error);
2349 }
2350 
2351 /*
2352  * Apply an advisory lock on a file descriptor.
2353  *
2354  * Just attempt to get a record lock of the requested type on
2355  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2356  */
2357 int
2358 sys_flock(struct flock_args *uap)
2359 {
2360 	struct proc *p = curproc;
2361 	struct file *fp;
2362 	struct vnode *vp;
2363 	struct flock lf;
2364 	int error;
2365 
2366 	if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2367 		return (EBADF);
2368 	if (fp->f_type != DTYPE_VNODE) {
2369 		error = EOPNOTSUPP;
2370 		goto done;
2371 	}
2372 	vp = (struct vnode *)fp->f_data;
2373 	lf.l_whence = SEEK_SET;
2374 	lf.l_start = 0;
2375 	lf.l_len = 0;
2376 	if (uap->how & LOCK_UN) {
2377 		lf.l_type = F_UNLCK;
2378 		fp->f_flag &= ~FHASLOCK;
2379 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2380 		goto done;
2381 	}
2382 	if (uap->how & LOCK_EX)
2383 		lf.l_type = F_WRLCK;
2384 	else if (uap->how & LOCK_SH)
2385 		lf.l_type = F_RDLCK;
2386 	else {
2387 		error = EBADF;
2388 		goto done;
2389 	}
2390 	fp->f_flag |= FHASLOCK;
2391 	if (uap->how & LOCK_NB)
2392 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2393 	else
2394 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2395 done:
2396 	fdrop(fp);
2397 	return (error);
2398 }
2399 
2400 /*
2401  * File Descriptor pseudo-device driver (/dev/fd/).
2402  *
2403  * Opening minor device N dup()s the file (if any) connected to file
2404  * descriptor N belonging to the calling process.  Note that this driver
2405  * consists of only the ``open()'' routine, because all subsequent
2406  * references to this file will be direct to the other driver.
2407  */
2408 /* ARGSUSED */
2409 static int
2410 fdopen(struct dev_open_args *ap)
2411 {
2412 	thread_t td = curthread;
2413 
2414 	KKASSERT(td->td_lwp != NULL);
2415 
2416 	/*
2417 	 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2418 	 * the file descriptor being sought for duplication. The error
2419 	 * return ensures that the vnode for this device will be released
2420 	 * by vn_open. Open will detect this special error and take the
2421 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2422 	 * will simply report the error.
2423 	 */
2424 	td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2425 	return (ENODEV);
2426 }
2427 
2428 /*
2429  * The caller has reserved the file descriptor dfd for us.  On success we
2430  * must fsetfd() it.  On failure the caller will clean it up.
2431  *
2432  * NOT MPSAFE - isn't getting spinlocks, possibly other things
2433  */
2434 int
2435 dupfdopen(struct proc *p, int dfd, int sfd, int mode, int error)
2436 {
2437 	struct filedesc *fdp = p->p_fd;
2438 	struct file *wfp;
2439 	struct file *xfp;
2440 	int werror;
2441 
2442 	if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2443 		return (EBADF);
2444 
2445 	/*
2446 	 * Close a revoke/dup race.  Duping a descriptor marked as revoked
2447 	 * will dup a dummy descriptor instead of the real one.
2448 	 */
2449 	if (wfp->f_flag & FREVOKED) {
2450 		kprintf("Warning: attempt to dup() a revoked descriptor\n");
2451 		fdrop(wfp);
2452 		wfp = NULL;
2453 		werror = falloc(NULL, &wfp, NULL);
2454 		if (werror)
2455 			return (werror);
2456 	}
2457 
2458 	/*
2459 	 * There are two cases of interest here.
2460 	 *
2461 	 * For ENODEV simply dup sfd to file descriptor dfd and return.
2462 	 *
2463 	 * For ENXIO steal away the file structure from sfd and store it
2464 	 * dfd.  sfd is effectively closed by this operation.
2465 	 *
2466 	 * Any other error code is just returned.
2467 	 */
2468 	switch (error) {
2469 	case ENODEV:
2470 		/*
2471 		 * Check that the mode the file is being opened for is a
2472 		 * subset of the mode of the existing descriptor.
2473 		 */
2474 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2475 			error = EACCES;
2476 			break;
2477 		}
2478 		fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2479 		fsetfd(p, wfp, dfd);
2480 		error = 0;
2481 		break;
2482 	case ENXIO:
2483 		/*
2484 		 * Steal away the file pointer from dfd, and stuff it into indx.
2485 		 */
2486 		fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2487 		fsetfd(p, wfp, dfd);
2488 		if ((xfp = funsetfd_locked(fdp, sfd)) != NULL)
2489 			fdrop(xfp);
2490 		error = 0;
2491 		break;
2492 	default:
2493 		break;
2494 	}
2495 	fdrop(wfp);
2496 	return (error);
2497 }
2498 
2499 /*
2500  * NOT MPSAFE - I think these refer to a common file descriptor table
2501  * and we need to spinlock that to link fdtol in.
2502  */
2503 struct filedesc_to_leader *
2504 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2505 			 struct proc *leader)
2506 {
2507 	struct filedesc_to_leader *fdtol;
2508 
2509 	fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2510 			M_FILEDESC_TO_LEADER, M_WAITOK);
2511 	fdtol->fdl_refcount = 1;
2512 	fdtol->fdl_holdcount = 0;
2513 	fdtol->fdl_wakeup = 0;
2514 	fdtol->fdl_leader = leader;
2515 	if (old != NULL) {
2516 		fdtol->fdl_next = old->fdl_next;
2517 		fdtol->fdl_prev = old;
2518 		old->fdl_next = fdtol;
2519 		fdtol->fdl_next->fdl_prev = fdtol;
2520 	} else {
2521 		fdtol->fdl_next = fdtol;
2522 		fdtol->fdl_prev = fdtol;
2523 	}
2524 	return fdtol;
2525 }
2526 
2527 /*
2528  * Scan all file pointers in the system.  The callback is made with
2529  * the master list spinlock held exclusively.
2530  *
2531  * MPSAFE
2532  */
2533 void
2534 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2535 {
2536 	struct file *fp;
2537 	int res;
2538 
2539 	spin_lock_wr(&filehead_spin);
2540 	LIST_FOREACH(fp, &filehead, f_list) {
2541 		res = callback(fp, data);
2542 		if (res < 0)
2543 			break;
2544 	}
2545 	spin_unlock_wr(&filehead_spin);
2546 }
2547 
2548 /*
2549  * Get file structures.
2550  *
2551  * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2552  */
2553 
2554 struct sysctl_kern_file_info {
2555 	int count;
2556 	int error;
2557 	struct sysctl_req *req;
2558 };
2559 
2560 static int sysctl_kern_file_callback(struct proc *p, void *data);
2561 
2562 static int
2563 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2564 {
2565 	struct sysctl_kern_file_info info;
2566 
2567 	/*
2568 	 * Note: because the number of file descriptors is calculated
2569 	 * in different ways for sizing vs returning the data,
2570 	 * there is information leakage from the first loop.  However,
2571 	 * it is of a similar order of magnitude to the leakage from
2572 	 * global system statistics such as kern.openfiles.
2573 	 *
2574 	 * When just doing a count, note that we cannot just count
2575 	 * the elements and add f_count via the filehead list because
2576 	 * threaded processes share their descriptor table and f_count might
2577 	 * still be '1' in that case.
2578 	 *
2579 	 * Since the SYSCTL op can block, we must hold the process to
2580 	 * prevent it being ripped out from under us either in the
2581 	 * file descriptor loop or in the greater LIST_FOREACH.  The
2582 	 * process may be in varying states of disrepair.  If the process
2583 	 * is in SZOMB we may have caught it just as it is being removed
2584 	 * from the allproc list, we must skip it in that case to maintain
2585 	 * an unbroken chain through the allproc list.
2586 	 */
2587 	info.count = 0;
2588 	info.error = 0;
2589 	info.req = req;
2590 	allproc_scan(sysctl_kern_file_callback, &info);
2591 
2592 	/*
2593 	 * When just calculating the size, overestimate a bit to try to
2594 	 * prevent system activity from causing the buffer-fill call
2595 	 * to fail later on.
2596 	 */
2597 	if (req->oldptr == NULL) {
2598 		info.count = (info.count + 16) + (info.count / 10);
2599 		info.error = SYSCTL_OUT(req, NULL,
2600 					info.count * sizeof(struct kinfo_file));
2601 	}
2602 	return (info.error);
2603 }
2604 
2605 static int
2606 sysctl_kern_file_callback(struct proc *p, void *data)
2607 {
2608 	struct sysctl_kern_file_info *info = data;
2609 	struct kinfo_file kf;
2610 	struct filedesc *fdp;
2611 	struct file *fp;
2612 	uid_t uid;
2613 	int n;
2614 
2615 	if (p->p_stat == SIDL || p->p_stat == SZOMB)
2616 		return(0);
2617 	if (!PRISON_CHECK(info->req->td->td_proc->p_ucred, p->p_ucred) != 0)
2618 		return(0);
2619 
2620 	/*
2621 	 * Softref the fdp to prevent it from being destroyed
2622 	 */
2623 	spin_lock_wr(&p->p_spin);
2624 	if ((fdp = p->p_fd) == NULL) {
2625 		spin_unlock_wr(&p->p_spin);
2626 		return(0);
2627 	}
2628 	atomic_add_int(&fdp->fd_softrefs, 1);
2629 	spin_unlock_wr(&p->p_spin);
2630 
2631 	/*
2632 	 * The fdp's own spinlock prevents the contents from being
2633 	 * modified.
2634 	 */
2635 	spin_lock_rd(&fdp->fd_spin);
2636 	for (n = 0; n < fdp->fd_nfiles; ++n) {
2637 		if ((fp = fdp->fd_files[n].fp) == NULL)
2638 			continue;
2639 		if (info->req->oldptr == NULL) {
2640 			++info->count;
2641 		} else {
2642 			uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2643 			kcore_make_file(&kf, fp, p->p_pid, uid, n);
2644 			spin_unlock_rd(&fdp->fd_spin);
2645 			info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2646 			spin_lock_rd(&fdp->fd_spin);
2647 			if (info->error)
2648 				break;
2649 		}
2650 	}
2651 	spin_unlock_rd(&fdp->fd_spin);
2652 	atomic_subtract_int(&fdp->fd_softrefs, 1);
2653 	if (info->error)
2654 		return(-1);
2655 	return(0);
2656 }
2657 
2658 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2659     0, 0, sysctl_kern_file, "S,file", "Entire file table");
2660 
2661 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2662     &maxfilesperproc, 0, "Maximum files allowed open per process");
2663 
2664 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2665     &maxfiles, 0, "Maximum number of files");
2666 
2667 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2668     &maxfilesrootres, 0, "Descriptors reserved for root use");
2669 
2670 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2671 	&nfiles, 0, "System-wide number of open files");
2672 
2673 static void
2674 fildesc_drvinit(void *unused)
2675 {
2676 	int fd;
2677 
2678 	for (fd = 0; fd < NUMFDESC; fd++) {
2679 		make_dev(&fildesc_ops, fd,
2680 			 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2681 	}
2682 
2683 	kprintf("fildesc_drvinit() building stdin, stdout, stderr: \n");
2684 
2685 	make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2686 	make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2687 	make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2688 }
2689 
2690 /*
2691  * MPSAFE
2692  */
2693 struct fileops badfileops = {
2694 	.fo_read = badfo_readwrite,
2695 	.fo_write = badfo_readwrite,
2696 	.fo_ioctl = badfo_ioctl,
2697 	.fo_poll = badfo_poll,
2698 	.fo_kqfilter = badfo_kqfilter,
2699 	.fo_stat = badfo_stat,
2700 	.fo_close = badfo_close,
2701 	.fo_shutdown = badfo_shutdown
2702 };
2703 
2704 /*
2705  * MPSAFE
2706  */
2707 static int
2708 badfo_readwrite(
2709 	struct file *fp,
2710 	struct uio *uio,
2711 	struct ucred *cred,
2712 	int flags
2713 ) {
2714 	return (EBADF);
2715 }
2716 
2717 /*
2718  * MPSAFE
2719  */
2720 static int
2721 badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *cred)
2722 {
2723 	return (EBADF);
2724 }
2725 
2726 /*
2727  * MPSAFE
2728  */
2729 static int
2730 badfo_poll(struct file *fp, int events, struct ucred *cred)
2731 {
2732 	return (0);
2733 }
2734 
2735 /*
2736  * MPSAFE
2737  */
2738 static int
2739 badfo_kqfilter(struct file *fp, struct knote *kn)
2740 {
2741 	return (0);
2742 }
2743 
2744 static int
2745 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2746 {
2747 	return (EBADF);
2748 }
2749 
2750 /*
2751  * MPSAFE
2752  */
2753 static int
2754 badfo_close(struct file *fp)
2755 {
2756 	return (EBADF);
2757 }
2758 
2759 /*
2760  * MPSAFE
2761  */
2762 static int
2763 badfo_shutdown(struct file *fp, int how)
2764 {
2765 	return (EBADF);
2766 }
2767 
2768 /*
2769  * MPSAFE
2770  */
2771 int
2772 nofo_shutdown(struct file *fp, int how)
2773 {
2774 	return (EOPNOTSUPP);
2775 }
2776 
2777 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2778 					fildesc_drvinit,NULL)
2779