xref: /dflybsd-src/sys/kern/kern_descrip.c (revision 330d3c4b487f3fc5d0eb023645b0b2a569f7048e)
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/file.h>
84 #include <sys/filedesc.h>
85 #include <sys/kernel.h>
86 #include <sys/sysctl.h>
87 #include <sys/vnode.h>
88 #include <sys/proc.h>
89 #include <sys/nlookup.h>
90 #include <sys/file.h>
91 #include <sys/stat.h>
92 #include <sys/filio.h>
93 #include <sys/fcntl.h>
94 #include <sys/unistd.h>
95 #include <sys/resourcevar.h>
96 #include <sys/event.h>
97 #include <sys/kern_syscall.h>
98 #include <sys/kcore.h>
99 #include <sys/kinfo.h>
100 #include <sys/un.h>
101 
102 #include <vm/vm.h>
103 #include <vm/vm_extern.h>
104 
105 #include <sys/thread2.h>
106 #include <sys/file2.h>
107 #include <sys/spinlock2.h>
108 #include <sys/mplock2.h>
109 
110 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
111 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
112 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
113 static void ffree(struct file *fp);
114 
115 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
116 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
117 		     "file desc to leader structures");
118 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
119 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
120 
121 static struct krate krate_uidinfo = { .freq = 1 };
122 
123 static	 d_open_t  fdopen;
124 #define NUMFDESC 64
125 
126 #define CDEV_MAJOR 22
127 static struct dev_ops fildesc_ops = {
128 	{ "FD", CDEV_MAJOR, 0 },
129 	.d_open =	fdopen,
130 };
131 
132 /*
133  * Descriptor management.
134  */
135 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
136 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
137 static int nfiles;		/* actual number of open files */
138 extern int cmask;
139 
140 /*
141  * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
142  *
143  * MPSAFE - must be called with fdp->fd_spin exclusively held
144  */
145 static __inline
146 void
147 fdfixup_locked(struct filedesc *fdp, int fd)
148 {
149 	if (fd < fdp->fd_freefile) {
150 	       fdp->fd_freefile = fd;
151 	}
152 	while (fdp->fd_lastfile >= 0 &&
153 	       fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
154 	       fdp->fd_files[fdp->fd_lastfile].reserved == 0
155 	) {
156 		--fdp->fd_lastfile;
157 	}
158 }
159 
160 /*
161  * System calls on descriptors.
162  *
163  * MPSAFE
164  */
165 int
166 sys_getdtablesize(struct getdtablesize_args *uap)
167 {
168 	struct proc *p = curproc;
169 	struct plimit *limit = p->p_limit;
170 	int dtsize;
171 
172 	spin_lock(&limit->p_spin);
173 	if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
174 		dtsize = INT_MAX;
175 	else
176 		dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
177 	spin_unlock(&limit->p_spin);
178 
179 	if (dtsize > maxfilesperproc)
180 		dtsize = maxfilesperproc;
181 	if (dtsize < minfilesperproc)
182 		dtsize = minfilesperproc;
183 	if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
184 		dtsize = maxfilesperuser;
185 	uap->sysmsg_result = dtsize;
186 	return (0);
187 }
188 
189 /*
190  * Duplicate a file descriptor to a particular value.
191  *
192  * note: keep in mind that a potential race condition exists when closing
193  * descriptors from a shared descriptor table (via rfork).
194  *
195  * MPSAFE
196  */
197 int
198 sys_dup2(struct dup2_args *uap)
199 {
200 	int error;
201 	int fd = 0;
202 
203 	error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
204 	uap->sysmsg_fds[0] = fd;
205 
206 	return (error);
207 }
208 
209 /*
210  * Duplicate a file descriptor.
211  *
212  * MPSAFE
213  */
214 int
215 sys_dup(struct dup_args *uap)
216 {
217 	int error;
218 	int fd = 0;
219 
220 	error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
221 	uap->sysmsg_fds[0] = fd;
222 
223 	return (error);
224 }
225 
226 /*
227  * MPALMOSTSAFE - acquires mplock for fp operations
228  */
229 int
230 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
231 {
232 	struct thread *td = curthread;
233 	struct proc *p = td->td_proc;
234 	struct file *fp;
235 	struct vnode *vp;
236 	u_int newmin;
237 	u_int oflags;
238 	u_int nflags;
239 	int tmp, error, flg = F_POSIX;
240 
241 	KKASSERT(p);
242 
243 	/*
244 	 * Operations on file descriptors that do not require a file pointer.
245 	 */
246 	switch (cmd) {
247 	case F_GETFD:
248 		error = fgetfdflags(p->p_fd, fd, &tmp);
249 		if (error == 0)
250 			dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
251 		return (error);
252 
253 	case F_SETFD:
254 		if (dat->fc_cloexec & FD_CLOEXEC)
255 			error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
256 		else
257 			error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
258 		return (error);
259 	case F_DUPFD:
260 		newmin = dat->fc_fd;
261 		error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
262 		return (error);
263 	default:
264 		break;
265 	}
266 
267 	/*
268 	 * Operations on file pointers
269 	 */
270 	if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
271 		return (EBADF);
272 
273 	get_mplock();
274 	switch (cmd) {
275 	case F_GETFL:
276 		dat->fc_flags = OFLAGS(fp->f_flag);
277 		error = 0;
278 		break;
279 
280 	case F_SETFL:
281 		oflags = fp->f_flag;
282 		nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
283 		nflags |= oflags & ~FCNTLFLAGS;
284 
285 		error = 0;
286 		if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
287 			error = EINVAL;
288 		if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
289 			tmp = nflags & FASYNC;
290 			error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
291 					 cred, NULL);
292 		}
293 		if (error == 0)
294 			fp->f_flag = nflags;
295 		break;
296 
297 	case F_GETOWN:
298 		error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
299 				 cred, NULL);
300 		break;
301 
302 	case F_SETOWN:
303 		error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
304 				 cred, NULL);
305 		break;
306 
307 	case F_SETLKW:
308 		flg |= F_WAIT;
309 		/* Fall into F_SETLK */
310 
311 	case F_SETLK:
312 		if (fp->f_type != DTYPE_VNODE) {
313 			error = EBADF;
314 			break;
315 		}
316 		vp = (struct vnode *)fp->f_data;
317 
318 		/*
319 		 * copyin/lockop may block
320 		 */
321 		if (dat->fc_flock.l_whence == SEEK_CUR)
322 			dat->fc_flock.l_start += fp->f_offset;
323 
324 		switch (dat->fc_flock.l_type) {
325 		case F_RDLCK:
326 			if ((fp->f_flag & FREAD) == 0) {
327 				error = EBADF;
328 				break;
329 			}
330 			p->p_leader->p_flag |= P_ADVLOCK;
331 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
332 			    &dat->fc_flock, flg);
333 			break;
334 		case F_WRLCK:
335 			if ((fp->f_flag & FWRITE) == 0) {
336 				error = EBADF;
337 				break;
338 			}
339 			p->p_leader->p_flag |= P_ADVLOCK;
340 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
341 			    &dat->fc_flock, flg);
342 			break;
343 		case F_UNLCK:
344 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
345 				&dat->fc_flock, F_POSIX);
346 			break;
347 		default:
348 			error = EINVAL;
349 			break;
350 		}
351 
352 		/*
353 		 * It is possible to race a close() on the descriptor while
354 		 * we were blocked getting the lock.  If this occurs the
355 		 * close might not have caught the lock.
356 		 */
357 		if (checkfdclosed(p->p_fd, fd, fp)) {
358 			dat->fc_flock.l_whence = SEEK_SET;
359 			dat->fc_flock.l_start = 0;
360 			dat->fc_flock.l_len = 0;
361 			dat->fc_flock.l_type = F_UNLCK;
362 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
363 					   F_UNLCK, &dat->fc_flock, F_POSIX);
364 		}
365 		break;
366 
367 	case F_GETLK:
368 		if (fp->f_type != DTYPE_VNODE) {
369 			error = EBADF;
370 			break;
371 		}
372 		vp = (struct vnode *)fp->f_data;
373 		/*
374 		 * copyin/lockop may block
375 		 */
376 		if (dat->fc_flock.l_type != F_RDLCK &&
377 		    dat->fc_flock.l_type != F_WRLCK &&
378 		    dat->fc_flock.l_type != F_UNLCK) {
379 			error = EINVAL;
380 			break;
381 		}
382 		if (dat->fc_flock.l_whence == SEEK_CUR)
383 			dat->fc_flock.l_start += fp->f_offset;
384 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
385 			    &dat->fc_flock, F_POSIX);
386 		break;
387 	default:
388 		error = EINVAL;
389 		break;
390 	}
391 	rel_mplock();
392 
393 	fdrop(fp);
394 	return (error);
395 }
396 
397 /*
398  * The file control system call.
399  *
400  * MPSAFE
401  */
402 int
403 sys_fcntl(struct fcntl_args *uap)
404 {
405 	union fcntl_dat dat;
406 	int error;
407 
408 	switch (uap->cmd) {
409 	case F_DUPFD:
410 		dat.fc_fd = uap->arg;
411 		break;
412 	case F_SETFD:
413 		dat.fc_cloexec = uap->arg;
414 		break;
415 	case F_SETFL:
416 		dat.fc_flags = uap->arg;
417 		break;
418 	case F_SETOWN:
419 		dat.fc_owner = uap->arg;
420 		break;
421 	case F_SETLKW:
422 	case F_SETLK:
423 	case F_GETLK:
424 		error = copyin((caddr_t)uap->arg, &dat.fc_flock,
425 			       sizeof(struct flock));
426 		if (error)
427 			return (error);
428 		break;
429 	}
430 
431 	error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
432 
433 	if (error == 0) {
434 		switch (uap->cmd) {
435 		case F_DUPFD:
436 			uap->sysmsg_result = dat.fc_fd;
437 			break;
438 		case F_GETFD:
439 			uap->sysmsg_result = dat.fc_cloexec;
440 			break;
441 		case F_GETFL:
442 			uap->sysmsg_result = dat.fc_flags;
443 			break;
444 		case F_GETOWN:
445 			uap->sysmsg_result = dat.fc_owner;
446 		case F_GETLK:
447 			error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
448 			    sizeof(struct flock));
449 			break;
450 		}
451 	}
452 
453 	return (error);
454 }
455 
456 /*
457  * Common code for dup, dup2, and fcntl(F_DUPFD).
458  *
459  * The type flag can be either DUP_FIXED or DUP_VARIABLE.  DUP_FIXED tells
460  * kern_dup() to destructively dup over an existing file descriptor if new
461  * is already open.  DUP_VARIABLE tells kern_dup() to find the lowest
462  * unused file descriptor that is greater than or equal to new.
463  *
464  * MPSAFE
465  */
466 int
467 kern_dup(enum dup_type type, int old, int new, int *res)
468 {
469 	struct thread *td = curthread;
470 	struct proc *p = td->td_proc;
471 	struct filedesc *fdp = p->p_fd;
472 	struct file *fp;
473 	struct file *delfp;
474 	int oldflags;
475 	int holdleaders;
476 	int dtsize;
477 	int error, newfd;
478 
479 	/*
480 	 * Verify that we have a valid descriptor to dup from and
481 	 * possibly to dup to.
482 	 *
483 	 * NOTE: maxfilesperuser is not applicable to dup()
484 	 */
485 retry:
486 	if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
487 		dtsize = INT_MAX;
488 	else
489 		dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
490 	if (dtsize > maxfilesperproc)
491 		dtsize = maxfilesperproc;
492 	if (dtsize < minfilesperproc)
493 		dtsize = minfilesperproc;
494 
495 	if (new < 0 || new > dtsize)
496 		return (EINVAL);
497 
498 	spin_lock(&fdp->fd_spin);
499 	if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
500 		spin_unlock(&fdp->fd_spin);
501 		return (EBADF);
502 	}
503 	if (type == DUP_FIXED && old == new) {
504 		*res = new;
505 		spin_unlock(&fdp->fd_spin);
506 		return (0);
507 	}
508 	fp = fdp->fd_files[old].fp;
509 	oldflags = fdp->fd_files[old].fileflags;
510 	fhold(fp);	/* MPSAFE - can be called with a spinlock held */
511 
512 	/*
513 	 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
514 	 * if the requested descriptor is beyond the current table size.
515 	 *
516 	 * This can block.  Retry if the source descriptor no longer matches
517 	 * or if our expectation in the expansion case races.
518 	 *
519 	 * If we are not expanding or allocating a new decriptor, then reset
520 	 * the target descriptor to a reserved state so we have a uniform
521 	 * setup for the next code block.
522 	 */
523 	if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
524 		spin_unlock(&fdp->fd_spin);
525 		error = fdalloc(p, new, &newfd);
526 		spin_lock(&fdp->fd_spin);
527 		if (error) {
528 			spin_unlock(&fdp->fd_spin);
529 			fdrop(fp);
530 			return (error);
531 		}
532 		/*
533 		 * Check for ripout
534 		 */
535 		if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
536 			fsetfd_locked(fdp, NULL, newfd);
537 			spin_unlock(&fdp->fd_spin);
538 			fdrop(fp);
539 			goto retry;
540 		}
541 		/*
542 		 * Check for expansion race
543 		 */
544 		if (type != DUP_VARIABLE && new != newfd) {
545 			fsetfd_locked(fdp, NULL, newfd);
546 			spin_unlock(&fdp->fd_spin);
547 			fdrop(fp);
548 			goto retry;
549 		}
550 		/*
551 		 * Check for ripout, newfd reused old (this case probably
552 		 * can't occur).
553 		 */
554 		if (old == newfd) {
555 			fsetfd_locked(fdp, NULL, newfd);
556 			spin_unlock(&fdp->fd_spin);
557 			fdrop(fp);
558 			goto retry;
559 		}
560 		new = newfd;
561 		delfp = NULL;
562 	} else {
563 		if (fdp->fd_files[new].reserved) {
564 			spin_unlock(&fdp->fd_spin);
565 			fdrop(fp);
566 			kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
567 			tsleep(fdp, 0, "fdres", hz);
568 			goto retry;
569 		}
570 
571 		/*
572 		 * If the target descriptor was never allocated we have
573 		 * to allocate it.  If it was we have to clean out the
574 		 * old descriptor.  delfp inherits the ref from the
575 		 * descriptor table.
576 		 */
577 		delfp = fdp->fd_files[new].fp;
578 		fdp->fd_files[new].fp = NULL;
579 		fdp->fd_files[new].reserved = 1;
580 		if (delfp == NULL) {
581 			fdreserve_locked(fdp, new, 1);
582 			if (new > fdp->fd_lastfile)
583 				fdp->fd_lastfile = new;
584 		}
585 
586 	}
587 
588 	/*
589 	 * NOTE: still holding an exclusive spinlock
590 	 */
591 
592 	/*
593 	 * If a descriptor is being overwritten we may hve to tell
594 	 * fdfree() to sleep to ensure that all relevant process
595 	 * leaders can be traversed in closef().
596 	 */
597 	if (delfp != NULL && p->p_fdtol != NULL) {
598 		fdp->fd_holdleaderscount++;
599 		holdleaders = 1;
600 	} else {
601 		holdleaders = 0;
602 	}
603 	KASSERT(delfp == NULL || type == DUP_FIXED,
604 		("dup() picked an open file"));
605 
606 	/*
607 	 * Duplicate the source descriptor, update lastfile.  If the new
608 	 * descriptor was not allocated and we aren't replacing an existing
609 	 * descriptor we have to mark the descriptor as being in use.
610 	 *
611 	 * The fd_files[] array inherits fp's hold reference.
612 	 */
613 	fsetfd_locked(fdp, fp, new);
614 	fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
615 	spin_unlock(&fdp->fd_spin);
616 	fdrop(fp);
617 	*res = new;
618 
619 	/*
620 	 * If we dup'd over a valid file, we now own the reference to it
621 	 * and must dispose of it using closef() semantics (as if a
622 	 * close() were performed on it).
623 	 */
624 	if (delfp) {
625 		if (SLIST_FIRST(&delfp->f_klist))
626 			knote_fdclose(delfp, fdp, new);
627 		closef(delfp, p);
628 		if (holdleaders) {
629 			spin_lock(&fdp->fd_spin);
630 			fdp->fd_holdleaderscount--;
631 			if (fdp->fd_holdleaderscount == 0 &&
632 			    fdp->fd_holdleaderswakeup != 0) {
633 				fdp->fd_holdleaderswakeup = 0;
634 				spin_unlock(&fdp->fd_spin);
635 				wakeup(&fdp->fd_holdleaderscount);
636 			} else {
637 				spin_unlock(&fdp->fd_spin);
638 			}
639 		}
640 	}
641 	return (0);
642 }
643 
644 /*
645  * If sigio is on the list associated with a process or process group,
646  * disable signalling from the device, remove sigio from the list and
647  * free sigio.
648  */
649 void
650 funsetown(struct sigio *sigio)
651 {
652 	if (sigio == NULL)
653 		return;
654 	crit_enter();
655 	*(sigio->sio_myref) = NULL;
656 	crit_exit();
657 	if (sigio->sio_pgid < 0) {
658 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
659 			     sigio, sio_pgsigio);
660 	} else /* if ((*sigiop)->sio_pgid > 0) */ {
661 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
662 			     sigio, sio_pgsigio);
663 	}
664 	crfree(sigio->sio_ucred);
665 	kfree(sigio, M_SIGIO);
666 }
667 
668 /* Free a list of sigio structures. */
669 void
670 funsetownlst(struct sigiolst *sigiolst)
671 {
672 	struct sigio *sigio;
673 
674 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
675 		funsetown(sigio);
676 }
677 
678 /*
679  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
680  *
681  * After permission checking, add a sigio structure to the sigio list for
682  * the process or process group.
683  */
684 int
685 fsetown(pid_t pgid, struct sigio **sigiop)
686 {
687 	struct proc *proc;
688 	struct pgrp *pgrp;
689 	struct sigio *sigio;
690 
691 	if (pgid == 0) {
692 		funsetown(*sigiop);
693 		return (0);
694 	}
695 	if (pgid > 0) {
696 		proc = pfind(pgid);
697 		if (proc == NULL)
698 			return (ESRCH);
699 
700 		/*
701 		 * Policy - Don't allow a process to FSETOWN a process
702 		 * in another session.
703 		 *
704 		 * Remove this test to allow maximum flexibility or
705 		 * restrict FSETOWN to the current process or process
706 		 * group for maximum safety.
707 		 */
708 		if (proc->p_session != curproc->p_session)
709 			return (EPERM);
710 
711 		pgrp = NULL;
712 	} else /* if (pgid < 0) */ {
713 		pgrp = pgfind(-pgid);
714 		if (pgrp == NULL)
715 			return (ESRCH);
716 
717 		/*
718 		 * Policy - Don't allow a process to FSETOWN a process
719 		 * in another session.
720 		 *
721 		 * Remove this test to allow maximum flexibility or
722 		 * restrict FSETOWN to the current process or process
723 		 * group for maximum safety.
724 		 */
725 		if (pgrp->pg_session != curproc->p_session)
726 			return (EPERM);
727 
728 		proc = NULL;
729 	}
730 	funsetown(*sigiop);
731 	sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
732 	if (pgid > 0) {
733 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
734 		sigio->sio_proc = proc;
735 	} else {
736 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
737 		sigio->sio_pgrp = pgrp;
738 	}
739 	sigio->sio_pgid = pgid;
740 	sigio->sio_ucred = crhold(curthread->td_ucred);
741 	/* It would be convenient if p_ruid was in ucred. */
742 	sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
743 	sigio->sio_myref = sigiop;
744 	crit_enter();
745 	*sigiop = sigio;
746 	crit_exit();
747 	return (0);
748 }
749 
750 /*
751  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
752  */
753 pid_t
754 fgetown(struct sigio *sigio)
755 {
756 	return (sigio != NULL ? sigio->sio_pgid : 0);
757 }
758 
759 /*
760  * Close many file descriptors.
761  *
762  * MPSAFE
763  */
764 int
765 sys_closefrom(struct closefrom_args *uap)
766 {
767 	return(kern_closefrom(uap->fd));
768 }
769 
770 /*
771  * Close all file descriptors greater then or equal to fd
772  *
773  * MPSAFE
774  */
775 int
776 kern_closefrom(int fd)
777 {
778 	struct thread *td = curthread;
779 	struct proc *p = td->td_proc;
780 	struct filedesc *fdp;
781 
782 	KKASSERT(p);
783 	fdp = p->p_fd;
784 
785 	if (fd < 0)
786 		return (EINVAL);
787 
788 	/*
789 	 * NOTE: This function will skip unassociated descriptors and
790 	 * reserved descriptors that have not yet been assigned.
791 	 * fd_lastfile can change as a side effect of kern_close().
792 	 */
793 	spin_lock(&fdp->fd_spin);
794 	while (fd <= fdp->fd_lastfile) {
795 		if (fdp->fd_files[fd].fp != NULL) {
796 			spin_unlock(&fdp->fd_spin);
797 			/* ok if this races another close */
798 			if (kern_close(fd) == EINTR)
799 				return (EINTR);
800 			spin_lock(&fdp->fd_spin);
801 		}
802 		++fd;
803 	}
804 	spin_unlock(&fdp->fd_spin);
805 	return (0);
806 }
807 
808 /*
809  * Close a file descriptor.
810  *
811  * MPSAFE
812  */
813 int
814 sys_close(struct close_args *uap)
815 {
816 	return(kern_close(uap->fd));
817 }
818 
819 /*
820  * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
821  */
822 int
823 kern_close(int fd)
824 {
825 	struct thread *td = curthread;
826 	struct proc *p = td->td_proc;
827 	struct filedesc *fdp;
828 	struct file *fp;
829 	int error;
830 	int holdleaders;
831 
832 	KKASSERT(p);
833 	fdp = p->p_fd;
834 
835 	spin_lock(&fdp->fd_spin);
836 	if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
837 		spin_unlock(&fdp->fd_spin);
838 		return (EBADF);
839 	}
840 	holdleaders = 0;
841 	if (p->p_fdtol != NULL) {
842 		/*
843 		 * Ask fdfree() to sleep to ensure that all relevant
844 		 * process leaders can be traversed in closef().
845 		 */
846 		fdp->fd_holdleaderscount++;
847 		holdleaders = 1;
848 	}
849 
850 	/*
851 	 * we now hold the fp reference that used to be owned by the descriptor
852 	 * array.
853 	 */
854 	spin_unlock(&fdp->fd_spin);
855 	if (SLIST_FIRST(&fp->f_klist))
856 		knote_fdclose(fp, fdp, fd);
857 	error = closef(fp, p);
858 	if (holdleaders) {
859 		spin_lock(&fdp->fd_spin);
860 		fdp->fd_holdleaderscount--;
861 		if (fdp->fd_holdleaderscount == 0 &&
862 		    fdp->fd_holdleaderswakeup != 0) {
863 			fdp->fd_holdleaderswakeup = 0;
864 			spin_unlock(&fdp->fd_spin);
865 			wakeup(&fdp->fd_holdleaderscount);
866 		} else {
867 			spin_unlock(&fdp->fd_spin);
868 		}
869 	}
870 	return (error);
871 }
872 
873 /*
874  * shutdown_args(int fd, int how)
875  */
876 int
877 kern_shutdown(int fd, int how)
878 {
879 	struct thread *td = curthread;
880 	struct proc *p = td->td_proc;
881 	struct file *fp;
882 	int error;
883 
884 	KKASSERT(p);
885 
886 	if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
887 		return (EBADF);
888 	error = fo_shutdown(fp, how);
889 	fdrop(fp);
890 
891 	return (error);
892 }
893 
894 /*
895  * MPALMOSTSAFE
896  */
897 int
898 sys_shutdown(struct shutdown_args *uap)
899 {
900 	int error;
901 
902 	get_mplock();
903 	error = kern_shutdown(uap->s, uap->how);
904 	rel_mplock();
905 
906 	return (error);
907 }
908 
909 /*
910  * MPSAFE
911  */
912 int
913 kern_fstat(int fd, struct stat *ub)
914 {
915 	struct thread *td = curthread;
916 	struct proc *p = td->td_proc;
917 	struct file *fp;
918 	int error;
919 
920 	KKASSERT(p);
921 
922 	if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
923 		return (EBADF);
924 	error = fo_stat(fp, ub, td->td_ucred);
925 	fdrop(fp);
926 
927 	return (error);
928 }
929 
930 /*
931  * Return status information about a file descriptor.
932  *
933  * MPSAFE
934  */
935 int
936 sys_fstat(struct fstat_args *uap)
937 {
938 	struct stat st;
939 	int error;
940 
941 	error = kern_fstat(uap->fd, &st);
942 
943 	if (error == 0)
944 		error = copyout(&st, uap->sb, sizeof(st));
945 	return (error);
946 }
947 
948 /*
949  * Return pathconf information about a file descriptor.
950  *
951  * MPALMOSTSAFE
952  */
953 int
954 sys_fpathconf(struct fpathconf_args *uap)
955 {
956 	struct thread *td = curthread;
957 	struct proc *p = td->td_proc;
958 	struct file *fp;
959 	struct vnode *vp;
960 	int error = 0;
961 
962 	if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
963 		return (EBADF);
964 
965 	switch (fp->f_type) {
966 	case DTYPE_PIPE:
967 	case DTYPE_SOCKET:
968 		if (uap->name != _PC_PIPE_BUF) {
969 			error = EINVAL;
970 		} else {
971 			uap->sysmsg_result = PIPE_BUF;
972 			error = 0;
973 		}
974 		break;
975 	case DTYPE_FIFO:
976 	case DTYPE_VNODE:
977 		vp = (struct vnode *)fp->f_data;
978 		get_mplock();
979 		error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
980 		rel_mplock();
981 		break;
982 	default:
983 		error = EOPNOTSUPP;
984 		break;
985 	}
986 	fdrop(fp);
987 	return(error);
988 }
989 
990 static int fdexpand;
991 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand,
992 	   0, "");
993 
994 /*
995  * Grow the file table so it can hold through descriptor (want).
996  *
997  * The fdp's spinlock must be held exclusively on entry and may be held
998  * exclusively on return.  The spinlock may be cycled by the routine.
999  *
1000  * MPSAFE
1001  */
1002 static void
1003 fdgrow_locked(struct filedesc *fdp, int want)
1004 {
1005 	struct fdnode *newfiles;
1006 	struct fdnode *oldfiles;
1007 	int nf, extra;
1008 
1009 	nf = fdp->fd_nfiles;
1010 	do {
1011 		/* nf has to be of the form 2^n - 1 */
1012 		nf = 2 * nf + 1;
1013 	} while (nf <= want);
1014 
1015 	spin_unlock(&fdp->fd_spin);
1016 	newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
1017 	spin_lock(&fdp->fd_spin);
1018 
1019 	/*
1020 	 * We could have raced another extend while we were not holding
1021 	 * the spinlock.
1022 	 */
1023 	if (fdp->fd_nfiles >= nf) {
1024 		spin_unlock(&fdp->fd_spin);
1025 		kfree(newfiles, M_FILEDESC);
1026 		spin_lock(&fdp->fd_spin);
1027 		return;
1028 	}
1029 	/*
1030 	 * Copy the existing ofile and ofileflags arrays
1031 	 * and zero the new portion of each array.
1032 	 */
1033 	extra = nf - fdp->fd_nfiles;
1034 	bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1035 	bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1036 
1037 	oldfiles = fdp->fd_files;
1038 	fdp->fd_files = newfiles;
1039 	fdp->fd_nfiles = nf;
1040 
1041 	if (oldfiles != fdp->fd_builtin_files) {
1042 		spin_unlock(&fdp->fd_spin);
1043 		kfree(oldfiles, M_FILEDESC);
1044 		spin_lock(&fdp->fd_spin);
1045 	}
1046 	fdexpand++;
1047 }
1048 
1049 /*
1050  * Number of nodes in right subtree, including the root.
1051  */
1052 static __inline int
1053 right_subtree_size(int n)
1054 {
1055 	return (n ^ (n | (n + 1)));
1056 }
1057 
1058 /*
1059  * Bigger ancestor.
1060  */
1061 static __inline int
1062 right_ancestor(int n)
1063 {
1064 	return (n | (n + 1));
1065 }
1066 
1067 /*
1068  * Smaller ancestor.
1069  */
1070 static __inline int
1071 left_ancestor(int n)
1072 {
1073 	return ((n & (n + 1)) - 1);
1074 }
1075 
1076 /*
1077  * Traverse the in-place binary tree buttom-up adjusting the allocation
1078  * count so scans can determine where free descriptors are located.
1079  *
1080  * MPSAFE - caller must be holding an exclusive spinlock on fdp
1081  */
1082 static
1083 void
1084 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1085 {
1086 	while (fd >= 0) {
1087 		fdp->fd_files[fd].allocated += incr;
1088 		KKASSERT(fdp->fd_files[fd].allocated >= 0);
1089 		fd = left_ancestor(fd);
1090 	}
1091 }
1092 
1093 /*
1094  * Reserve a file descriptor for the process.  If no error occurs, the
1095  * caller MUST at some point call fsetfd() or assign a file pointer
1096  * or dispose of the reservation.
1097  *
1098  * MPSAFE
1099  */
1100 int
1101 fdalloc(struct proc *p, int want, int *result)
1102 {
1103 	struct filedesc *fdp = p->p_fd;
1104 	struct uidinfo *uip;
1105 	int fd, rsize, rsum, node, lim;
1106 
1107 	/*
1108 	 * Check dtable size limit
1109 	 */
1110 	spin_lock(&p->p_limit->p_spin);
1111 	if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1112 		lim = INT_MAX;
1113 	else
1114 		lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1115 	spin_unlock(&p->p_limit->p_spin);
1116 
1117 	if (lim > maxfilesperproc)
1118 		lim = maxfilesperproc;
1119 	if (lim < minfilesperproc)
1120 		lim = minfilesperproc;
1121 	if (want >= lim)
1122 		return (EMFILE);
1123 
1124 	/*
1125 	 * Check that the user has not run out of descriptors (non-root only).
1126 	 * As a safety measure the dtable is allowed to have at least
1127 	 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1128 	 */
1129 	if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1130 		uip = p->p_ucred->cr_uidinfo;
1131 		if (uip->ui_openfiles > maxfilesperuser) {
1132 			krateprintf(&krate_uidinfo,
1133 				    "Warning: user %d pid %d (%s) ran out of "
1134 				    "file descriptors (%d/%d)\n",
1135 				    p->p_ucred->cr_uid, (int)p->p_pid,
1136 				    p->p_comm,
1137 				    uip->ui_openfiles, maxfilesperuser);
1138 			return(ENFILE);
1139 		}
1140 	}
1141 
1142 	/*
1143 	 * Grow the dtable if necessary
1144 	 */
1145 	spin_lock(&fdp->fd_spin);
1146 	if (want >= fdp->fd_nfiles)
1147 		fdgrow_locked(fdp, want);
1148 
1149 	/*
1150 	 * Search for a free descriptor starting at the higher
1151 	 * of want or fd_freefile.  If that fails, consider
1152 	 * expanding the ofile array.
1153 	 *
1154 	 * NOTE! the 'allocated' field is a cumulative recursive allocation
1155 	 * count.  If we happen to see a value of 0 then we can shortcut
1156 	 * our search.  Otherwise we run through through the tree going
1157 	 * down branches we know have free descriptor(s) until we hit a
1158 	 * leaf node.  The leaf node will be free but will not necessarily
1159 	 * have an allocated field of 0.
1160 	 */
1161 retry:
1162 	/* move up the tree looking for a subtree with a free node */
1163 	for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1164 	     fd = right_ancestor(fd)) {
1165 		if (fdp->fd_files[fd].allocated == 0)
1166 			goto found;
1167 
1168 		rsize = right_subtree_size(fd);
1169 		if (fdp->fd_files[fd].allocated == rsize)
1170 			continue;	/* right subtree full */
1171 
1172 		/*
1173 		 * Free fd is in the right subtree of the tree rooted at fd.
1174 		 * Call that subtree R.  Look for the smallest (leftmost)
1175 		 * subtree of R with an unallocated fd: continue moving
1176 		 * down the left branch until encountering a full left
1177 		 * subtree, then move to the right.
1178 		 */
1179 		for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1180 			node = fd + rsize;
1181 			rsum += fdp->fd_files[node].allocated;
1182 			if (fdp->fd_files[fd].allocated == rsum + rsize) {
1183 				fd = node;	/* move to the right */
1184 				if (fdp->fd_files[node].allocated == 0)
1185 					goto found;
1186 				rsum = 0;
1187 			}
1188 		}
1189 		goto found;
1190 	}
1191 
1192 	/*
1193 	 * No space in current array.  Expand?
1194 	 */
1195 	if (fdp->fd_nfiles >= lim) {
1196 		spin_unlock(&fdp->fd_spin);
1197 		return (EMFILE);
1198 	}
1199 	fdgrow_locked(fdp, want);
1200 	goto retry;
1201 
1202 found:
1203 	KKASSERT(fd < fdp->fd_nfiles);
1204 	if (fd > fdp->fd_lastfile)
1205 		fdp->fd_lastfile = fd;
1206 	if (want <= fdp->fd_freefile)
1207 		fdp->fd_freefile = fd;
1208 	*result = fd;
1209 	KKASSERT(fdp->fd_files[fd].fp == NULL);
1210 	KKASSERT(fdp->fd_files[fd].reserved == 0);
1211 	fdp->fd_files[fd].fileflags = 0;
1212 	fdp->fd_files[fd].reserved = 1;
1213 	fdreserve_locked(fdp, fd, 1);
1214 	spin_unlock(&fdp->fd_spin);
1215 	return (0);
1216 }
1217 
1218 /*
1219  * Check to see whether n user file descriptors
1220  * are available to the process p.
1221  *
1222  * MPSAFE
1223  */
1224 int
1225 fdavail(struct proc *p, int n)
1226 {
1227 	struct filedesc *fdp = p->p_fd;
1228 	struct fdnode *fdnode;
1229 	int i, lim, last;
1230 
1231 	spin_lock(&p->p_limit->p_spin);
1232 	if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1233 		lim = INT_MAX;
1234 	else
1235 		lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1236 	spin_unlock(&p->p_limit->p_spin);
1237 
1238 	if (lim > maxfilesperproc)
1239 		lim = maxfilesperproc;
1240 	if (lim < minfilesperproc)
1241 		lim = minfilesperproc;
1242 
1243 	spin_lock(&fdp->fd_spin);
1244 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1245 		spin_unlock(&fdp->fd_spin);
1246 		return (1);
1247 	}
1248 	last = min(fdp->fd_nfiles, lim);
1249 	fdnode = &fdp->fd_files[fdp->fd_freefile];
1250 	for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1251 		if (fdnode->fp == NULL && --n <= 0) {
1252 			spin_unlock(&fdp->fd_spin);
1253 			return (1);
1254 		}
1255 	}
1256 	spin_unlock(&fdp->fd_spin);
1257 	return (0);
1258 }
1259 
1260 /*
1261  * Revoke open descriptors referencing (f_data, f_type)
1262  *
1263  * Any revoke executed within a prison is only able to
1264  * revoke descriptors for processes within that prison.
1265  *
1266  * Returns 0 on success or an error code.
1267  */
1268 struct fdrevoke_info {
1269 	void *data;
1270 	short type;
1271 	short unused;
1272 	int count;
1273 	int intransit;
1274 	struct ucred *cred;
1275 	struct file *nfp;
1276 };
1277 
1278 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1279 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1280 
1281 int
1282 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1283 {
1284 	struct fdrevoke_info info;
1285 	int error;
1286 
1287 	bzero(&info, sizeof(info));
1288 	info.data = f_data;
1289 	info.type = f_type;
1290 	info.cred = cred;
1291 	error = falloc(NULL, &info.nfp, NULL);
1292 	if (error)
1293 		return (error);
1294 
1295 	/*
1296 	 * Scan the file pointer table once.  dups do not dup file pointers,
1297 	 * only descriptors, so there is no leak.  Set FREVOKED on the fps
1298 	 * being revoked.
1299 	 */
1300 	allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1301 
1302 	/*
1303 	 * If any fps were marked track down the related descriptors
1304 	 * and close them.  Any dup()s at this point will notice
1305 	 * the FREVOKED already set in the fp and do the right thing.
1306 	 *
1307 	 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1308 	 * socket) bumped the intransit counter and will require a
1309 	 * scan.  Races against fps leaving the socket are closed by
1310 	 * the socket code checking for FREVOKED.
1311 	 */
1312 	if (info.count)
1313 		allproc_scan(fdrevoke_proc_callback, &info);
1314 	if (info.intransit)
1315 		unp_revoke_gc(info.nfp);
1316 	fdrop(info.nfp);
1317 	return(0);
1318 }
1319 
1320 /*
1321  * Locate matching file pointers directly.
1322  */
1323 static int
1324 fdrevoke_check_callback(struct file *fp, void *vinfo)
1325 {
1326 	struct fdrevoke_info *info = vinfo;
1327 
1328 	/*
1329 	 * File pointers already flagged for revokation are skipped.
1330 	 */
1331 	if (fp->f_flag & FREVOKED)
1332 		return(0);
1333 
1334 	/*
1335 	 * If revoking from a prison file pointers created outside of
1336 	 * that prison, or file pointers without creds, cannot be revoked.
1337 	 */
1338 	if (info->cred->cr_prison &&
1339 	    (fp->f_cred == NULL ||
1340 	     info->cred->cr_prison != fp->f_cred->cr_prison)) {
1341 		return(0);
1342 	}
1343 
1344 	/*
1345 	 * If the file pointer matches then mark it for revocation.  The
1346 	 * flag is currently only used by unp_revoke_gc().
1347 	 *
1348 	 * info->count is a heuristic and can race in a SMP environment.
1349 	 */
1350 	if (info->data == fp->f_data && info->type == fp->f_type) {
1351 		atomic_set_int(&fp->f_flag, FREVOKED);
1352 		info->count += fp->f_count;
1353 		if (fp->f_msgcount)
1354 			++info->intransit;
1355 	}
1356 	return(0);
1357 }
1358 
1359 /*
1360  * Locate matching file pointers via process descriptor tables.
1361  */
1362 static int
1363 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1364 {
1365 	struct fdrevoke_info *info = vinfo;
1366 	struct filedesc *fdp;
1367 	struct file *fp;
1368 	int n;
1369 
1370 	if (p->p_stat == SIDL || p->p_stat == SZOMB)
1371 		return(0);
1372 	if (info->cred->cr_prison &&
1373 	    info->cred->cr_prison != p->p_ucred->cr_prison) {
1374 		return(0);
1375 	}
1376 
1377 	/*
1378 	 * If the controlling terminal of the process matches the
1379 	 * vnode being revoked we clear the controlling terminal.
1380 	 *
1381 	 * The normal spec_close() may not catch this because it
1382 	 * uses curproc instead of p.
1383 	 */
1384 	if (p->p_session && info->type == DTYPE_VNODE &&
1385 	    info->data == p->p_session->s_ttyvp) {
1386 		p->p_session->s_ttyvp = NULL;
1387 		vrele(info->data);
1388 	}
1389 
1390 	/*
1391 	 * Softref the fdp to prevent it from being destroyed
1392 	 */
1393 	spin_lock(&p->p_spin);
1394 	if ((fdp = p->p_fd) == NULL) {
1395 		spin_unlock(&p->p_spin);
1396 		return(0);
1397 	}
1398 	atomic_add_int(&fdp->fd_softrefs, 1);
1399 	spin_unlock(&p->p_spin);
1400 
1401 	/*
1402 	 * Locate and close any matching file descriptors.
1403 	 */
1404 	spin_lock(&fdp->fd_spin);
1405 	for (n = 0; n < fdp->fd_nfiles; ++n) {
1406 		if ((fp = fdp->fd_files[n].fp) == NULL)
1407 			continue;
1408 		if (fp->f_flag & FREVOKED) {
1409 			fhold(info->nfp);
1410 			fdp->fd_files[n].fp = info->nfp;
1411 			spin_unlock(&fdp->fd_spin);
1412 			knote_fdclose(fp, fdp, n);	/* XXX */
1413 			closef(fp, p);
1414 			spin_lock(&fdp->fd_spin);
1415 			--info->count;
1416 		}
1417 	}
1418 	spin_unlock(&fdp->fd_spin);
1419 	atomic_subtract_int(&fdp->fd_softrefs, 1);
1420 	return(0);
1421 }
1422 
1423 /*
1424  * falloc:
1425  *	Create a new open file structure and reserve a file decriptor
1426  *	for the process that refers to it.
1427  *
1428  *	Root creds are checked using lp, or assumed if lp is NULL.  If
1429  *	resultfd is non-NULL then lp must also be non-NULL.  No file
1430  *	descriptor is reserved (and no process context is needed) if
1431  *	resultfd is NULL.
1432  *
1433  *	A file pointer with a refcount of 1 is returned.  Note that the
1434  *	file pointer is NOT associated with the descriptor.  If falloc
1435  *	returns success, fsetfd() MUST be called to either associate the
1436  *	file pointer or clear the reservation.
1437  *
1438  * MPSAFE
1439  */
1440 int
1441 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1442 {
1443 	static struct timeval lastfail;
1444 	static int curfail;
1445 	struct file *fp;
1446 	struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1447 	int error;
1448 
1449 	fp = NULL;
1450 
1451 	/*
1452 	 * Handle filetable full issues and root overfill.
1453 	 */
1454 	if (nfiles >= maxfiles - maxfilesrootres &&
1455 	    (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1456 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1457 			kprintf("kern.maxfiles limit exceeded by uid %d, "
1458 				"please see tuning(7).\n",
1459 				cred->cr_ruid);
1460 		}
1461 		error = ENFILE;
1462 		goto done;
1463 	}
1464 
1465 	/*
1466 	 * Allocate a new file descriptor.
1467 	 */
1468 	fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1469 	spin_init(&fp->f_spin);
1470 	SLIST_INIT(&fp->f_klist);
1471 	fp->f_count = 1;
1472 	fp->f_ops = &badfileops;
1473 	fp->f_seqcount = 1;
1474 	fsetcred(fp, cred);
1475 	spin_lock(&filehead_spin);
1476 	nfiles++;
1477 	LIST_INSERT_HEAD(&filehead, fp, f_list);
1478 	spin_unlock(&filehead_spin);
1479 	if (resultfd) {
1480 		if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1481 			fdrop(fp);
1482 			fp = NULL;
1483 		}
1484 	} else {
1485 		error = 0;
1486 	}
1487 done:
1488 	*resultfp = fp;
1489 	return (error);
1490 }
1491 
1492 /*
1493  * Check for races against a file descriptor by determining that the
1494  * file pointer is still associated with the specified file descriptor,
1495  * and a close is not currently in progress.
1496  *
1497  * MPSAFE
1498  */
1499 int
1500 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
1501 {
1502 	int error;
1503 
1504 	spin_lock(&fdp->fd_spin);
1505 	if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1506 		error = EBADF;
1507 	else
1508 		error = 0;
1509 	spin_unlock(&fdp->fd_spin);
1510 	return (error);
1511 }
1512 
1513 /*
1514  * Associate a file pointer with a previously reserved file descriptor.
1515  * This function always succeeds.
1516  *
1517  * If fp is NULL, the file descriptor is returned to the pool.
1518  */
1519 
1520 /*
1521  * MPSAFE (exclusive spinlock must be held on call)
1522  */
1523 static void
1524 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1525 {
1526 	KKASSERT((unsigned)fd < fdp->fd_nfiles);
1527 	KKASSERT(fdp->fd_files[fd].reserved != 0);
1528 	if (fp) {
1529 		fhold(fp);
1530 		fdp->fd_files[fd].fp = fp;
1531 		fdp->fd_files[fd].reserved = 0;
1532 	} else {
1533 		fdp->fd_files[fd].reserved = 0;
1534 		fdreserve_locked(fdp, fd, -1);
1535 		fdfixup_locked(fdp, fd);
1536 	}
1537 }
1538 
1539 /*
1540  * MPSAFE
1541  */
1542 void
1543 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1544 {
1545 	spin_lock(&fdp->fd_spin);
1546 	fsetfd_locked(fdp, fp, fd);
1547 	spin_unlock(&fdp->fd_spin);
1548 }
1549 
1550 /*
1551  * MPSAFE (exclusive spinlock must be held on call)
1552  */
1553 static
1554 struct file *
1555 funsetfd_locked(struct filedesc *fdp, int fd)
1556 {
1557 	struct file *fp;
1558 
1559 	if ((unsigned)fd >= fdp->fd_nfiles)
1560 		return (NULL);
1561 	if ((fp = fdp->fd_files[fd].fp) == NULL)
1562 		return (NULL);
1563 	fdp->fd_files[fd].fp = NULL;
1564 	fdp->fd_files[fd].fileflags = 0;
1565 
1566 	fdreserve_locked(fdp, fd, -1);
1567 	fdfixup_locked(fdp, fd);
1568 	return(fp);
1569 }
1570 
1571 /*
1572  * MPSAFE
1573  */
1574 int
1575 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1576 {
1577 	int error;
1578 
1579 	spin_lock(&fdp->fd_spin);
1580 	if (((u_int)fd) >= fdp->fd_nfiles) {
1581 		error = EBADF;
1582 	} else if (fdp->fd_files[fd].fp == NULL) {
1583 		error = EBADF;
1584 	} else {
1585 		*flagsp = fdp->fd_files[fd].fileflags;
1586 		error = 0;
1587 	}
1588 	spin_unlock(&fdp->fd_spin);
1589 	return (error);
1590 }
1591 
1592 /*
1593  * MPSAFE
1594  */
1595 int
1596 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1597 {
1598 	int error;
1599 
1600 	spin_lock(&fdp->fd_spin);
1601 	if (((u_int)fd) >= fdp->fd_nfiles) {
1602 		error = EBADF;
1603 	} else if (fdp->fd_files[fd].fp == NULL) {
1604 		error = EBADF;
1605 	} else {
1606 		fdp->fd_files[fd].fileflags |= add_flags;
1607 		error = 0;
1608 	}
1609 	spin_unlock(&fdp->fd_spin);
1610 	return (error);
1611 }
1612 
1613 /*
1614  * MPSAFE
1615  */
1616 int
1617 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1618 {
1619 	int error;
1620 
1621 	spin_lock(&fdp->fd_spin);
1622 	if (((u_int)fd) >= fdp->fd_nfiles) {
1623 		error = EBADF;
1624 	} else if (fdp->fd_files[fd].fp == NULL) {
1625 		error = EBADF;
1626 	} else {
1627 		fdp->fd_files[fd].fileflags &= ~rem_flags;
1628 		error = 0;
1629 	}
1630 	spin_unlock(&fdp->fd_spin);
1631 	return (error);
1632 }
1633 
1634 /*
1635  * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1636  */
1637 void
1638 fsetcred(struct file *fp, struct ucred *ncr)
1639 {
1640 	struct ucred *ocr;
1641 	struct uidinfo *uip;
1642 
1643 	ocr = fp->f_cred;
1644 	if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1645 		if (ocr) {
1646 			uip = ocr->cr_uidinfo;
1647 			atomic_add_int(&uip->ui_openfiles, -1);
1648 		}
1649 		if (ncr) {
1650 			uip = ncr->cr_uidinfo;
1651 			atomic_add_int(&uip->ui_openfiles, 1);
1652 		}
1653 	}
1654 	if (ncr)
1655 		crhold(ncr);
1656 	fp->f_cred = ncr;
1657 	if (ocr)
1658 		crfree(ocr);
1659 }
1660 
1661 /*
1662  * Free a file descriptor.
1663  */
1664 static
1665 void
1666 ffree(struct file *fp)
1667 {
1668 	KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1669 	spin_lock(&filehead_spin);
1670 	LIST_REMOVE(fp, f_list);
1671 	nfiles--;
1672 	spin_unlock(&filehead_spin);
1673 	fsetcred(fp, NULL);
1674 	if (fp->f_nchandle.ncp)
1675 	    cache_drop(&fp->f_nchandle);
1676 	kfree(fp, M_FILE);
1677 }
1678 
1679 /*
1680  * called from init_main, initialize filedesc0 for proc0.
1681  */
1682 void
1683 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1684 {
1685 	p0->p_fd = fdp0;
1686 	p0->p_fdtol = NULL;
1687 	fdp0->fd_refcnt = 1;
1688 	fdp0->fd_cmask = cmask;
1689 	fdp0->fd_files = fdp0->fd_builtin_files;
1690 	fdp0->fd_nfiles = NDFILE;
1691 	fdp0->fd_lastfile = -1;
1692 	spin_init(&fdp0->fd_spin);
1693 }
1694 
1695 /*
1696  * Build a new filedesc structure.
1697  *
1698  * NOT MPSAFE (vref)
1699  */
1700 struct filedesc *
1701 fdinit(struct proc *p)
1702 {
1703 	struct filedesc *newfdp;
1704 	struct filedesc *fdp = p->p_fd;
1705 
1706 	newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1707 	spin_lock(&fdp->fd_spin);
1708 	if (fdp->fd_cdir) {
1709 		newfdp->fd_cdir = fdp->fd_cdir;
1710 		vref(newfdp->fd_cdir);
1711 		cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1712 	}
1713 
1714 	/*
1715 	 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1716 	 * proc0, but should unconditionally exist in other processes.
1717 	 */
1718 	if (fdp->fd_rdir) {
1719 		newfdp->fd_rdir = fdp->fd_rdir;
1720 		vref(newfdp->fd_rdir);
1721 		cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1722 	}
1723 	if (fdp->fd_jdir) {
1724 		newfdp->fd_jdir = fdp->fd_jdir;
1725 		vref(newfdp->fd_jdir);
1726 		cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1727 	}
1728 	spin_unlock(&fdp->fd_spin);
1729 
1730 	/* Create the file descriptor table. */
1731 	newfdp->fd_refcnt = 1;
1732 	newfdp->fd_cmask = cmask;
1733 	newfdp->fd_files = newfdp->fd_builtin_files;
1734 	newfdp->fd_nfiles = NDFILE;
1735 	newfdp->fd_lastfile = -1;
1736 	spin_init(&newfdp->fd_spin);
1737 
1738 	return (newfdp);
1739 }
1740 
1741 /*
1742  * Share a filedesc structure.
1743  *
1744  * MPSAFE
1745  */
1746 struct filedesc *
1747 fdshare(struct proc *p)
1748 {
1749 	struct filedesc *fdp;
1750 
1751 	fdp = p->p_fd;
1752 	spin_lock(&fdp->fd_spin);
1753 	fdp->fd_refcnt++;
1754 	spin_unlock(&fdp->fd_spin);
1755 	return (fdp);
1756 }
1757 
1758 /*
1759  * Copy a filedesc structure.
1760  *
1761  * MPSAFE
1762  */
1763 struct filedesc *
1764 fdcopy(struct proc *p)
1765 {
1766 	struct filedesc *fdp = p->p_fd;
1767 	struct filedesc *newfdp;
1768 	struct fdnode *fdnode;
1769 	int i;
1770 	int ni;
1771 
1772 	/*
1773 	 * Certain daemons might not have file descriptors.
1774 	 */
1775 	if (fdp == NULL)
1776 		return (NULL);
1777 
1778 	/*
1779 	 * Allocate the new filedesc and fd_files[] array.  This can race
1780 	 * with operations by other threads on the fdp so we have to be
1781 	 * careful.
1782 	 */
1783 	newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1784 again:
1785 	spin_lock(&fdp->fd_spin);
1786 	if (fdp->fd_lastfile < NDFILE) {
1787 		newfdp->fd_files = newfdp->fd_builtin_files;
1788 		i = NDFILE;
1789 	} else {
1790 		/*
1791 		 * We have to allocate (N^2-1) entries for our in-place
1792 		 * binary tree.  Allow the table to shrink.
1793 		 */
1794 		i = fdp->fd_nfiles;
1795 		ni = (i - 1) / 2;
1796 		while (ni > fdp->fd_lastfile && ni > NDFILE) {
1797 			i = ni;
1798 			ni = (i - 1) / 2;
1799 		}
1800 		spin_unlock(&fdp->fd_spin);
1801 		newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1802 					  M_FILEDESC, M_WAITOK | M_ZERO);
1803 
1804 		/*
1805 		 * Check for race, retry
1806 		 */
1807 		spin_lock(&fdp->fd_spin);
1808 		if (i <= fdp->fd_lastfile) {
1809 			spin_unlock(&fdp->fd_spin);
1810 			kfree(newfdp->fd_files, M_FILEDESC);
1811 			goto again;
1812 		}
1813 	}
1814 
1815 	/*
1816 	 * Dup the remaining fields. vref() and cache_hold() can be
1817 	 * safely called while holding the read spinlock on fdp.
1818 	 *
1819 	 * The read spinlock on fdp is still being held.
1820 	 *
1821 	 * NOTE: vref and cache_hold calls for the case where the vnode
1822 	 * or cache entry already has at least one ref may be called
1823 	 * while holding spin locks.
1824 	 */
1825 	if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1826 		vref(newfdp->fd_cdir);
1827 		cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1828 	}
1829 	/*
1830 	 * We must check for fd_rdir here, at least for now because
1831 	 * the init process is created before we have access to the
1832 	 * rootvode to take a reference to it.
1833 	 */
1834 	if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1835 		vref(newfdp->fd_rdir);
1836 		cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1837 	}
1838 	if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1839 		vref(newfdp->fd_jdir);
1840 		cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1841 	}
1842 	newfdp->fd_refcnt = 1;
1843 	newfdp->fd_nfiles = i;
1844 	newfdp->fd_lastfile = fdp->fd_lastfile;
1845 	newfdp->fd_freefile = fdp->fd_freefile;
1846 	newfdp->fd_cmask = fdp->fd_cmask;
1847 	spin_init(&newfdp->fd_spin);
1848 
1849 	/*
1850 	 * Copy the descriptor table through (i).  This also copies the
1851 	 * allocation state.   Then go through and ref the file pointers
1852 	 * and clean up any KQ descriptors.
1853 	 *
1854 	 * kq descriptors cannot be copied.  Since we haven't ref'd the
1855 	 * copied files yet we can ignore the return value from funsetfd().
1856 	 *
1857 	 * The read spinlock on fdp is still being held.
1858 	 */
1859 	bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1860 	for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1861 		fdnode = &newfdp->fd_files[i];
1862 		if (fdnode->reserved) {
1863 			fdreserve_locked(newfdp, i, -1);
1864 			fdnode->reserved = 0;
1865 			fdfixup_locked(newfdp, i);
1866 		} else if (fdnode->fp) {
1867 			if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1868 				(void)funsetfd_locked(newfdp, i);
1869 			} else {
1870 				fhold(fdnode->fp);
1871 			}
1872 		}
1873 	}
1874 	spin_unlock(&fdp->fd_spin);
1875 	return (newfdp);
1876 }
1877 
1878 /*
1879  * Release a filedesc structure.
1880  *
1881  * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1882  */
1883 void
1884 fdfree(struct proc *p, struct filedesc *repl)
1885 {
1886 	struct filedesc *fdp;
1887 	struct fdnode *fdnode;
1888 	int i;
1889 	struct filedesc_to_leader *fdtol;
1890 	struct file *fp;
1891 	struct vnode *vp;
1892 	struct flock lf;
1893 
1894 	/*
1895 	 * Certain daemons might not have file descriptors.
1896 	 */
1897 	fdp = p->p_fd;
1898 	if (fdp == NULL) {
1899 		p->p_fd = repl;
1900 		return;
1901 	}
1902 
1903 	/*
1904 	 * Severe messing around to follow.
1905 	 */
1906 	spin_lock(&fdp->fd_spin);
1907 
1908 	/* Check for special need to clear POSIX style locks */
1909 	fdtol = p->p_fdtol;
1910 	if (fdtol != NULL) {
1911 		KASSERT(fdtol->fdl_refcount > 0,
1912 			("filedesc_to_refcount botch: fdl_refcount=%d",
1913 			 fdtol->fdl_refcount));
1914 		if (fdtol->fdl_refcount == 1 &&
1915 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1916 			for (i = 0; i <= fdp->fd_lastfile; ++i) {
1917 				fdnode = &fdp->fd_files[i];
1918 				if (fdnode->fp == NULL ||
1919 				    fdnode->fp->f_type != DTYPE_VNODE) {
1920 					continue;
1921 				}
1922 				fp = fdnode->fp;
1923 				fhold(fp);
1924 				spin_unlock(&fdp->fd_spin);
1925 
1926 				lf.l_whence = SEEK_SET;
1927 				lf.l_start = 0;
1928 				lf.l_len = 0;
1929 				lf.l_type = F_UNLCK;
1930 				vp = (struct vnode *)fp->f_data;
1931 				(void) VOP_ADVLOCK(vp,
1932 						   (caddr_t)p->p_leader,
1933 						   F_UNLCK,
1934 						   &lf,
1935 						   F_POSIX);
1936 				fdrop(fp);
1937 				spin_lock(&fdp->fd_spin);
1938 			}
1939 		}
1940 	retry:
1941 		if (fdtol->fdl_refcount == 1) {
1942 			if (fdp->fd_holdleaderscount > 0 &&
1943 			    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1944 				/*
1945 				 * close() or do_dup() has cleared a reference
1946 				 * in a shared file descriptor table.
1947 				 */
1948 				fdp->fd_holdleaderswakeup = 1;
1949 				ssleep(&fdp->fd_holdleaderscount,
1950 				       &fdp->fd_spin, 0, "fdlhold", 0);
1951 				goto retry;
1952 			}
1953 			if (fdtol->fdl_holdcount > 0) {
1954 				/*
1955 				 * Ensure that fdtol->fdl_leader
1956 				 * remains valid in closef().
1957 				 */
1958 				fdtol->fdl_wakeup = 1;
1959 				ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1960 				goto retry;
1961 			}
1962 		}
1963 		fdtol->fdl_refcount--;
1964 		if (fdtol->fdl_refcount == 0 &&
1965 		    fdtol->fdl_holdcount == 0) {
1966 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1967 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1968 		} else {
1969 			fdtol = NULL;
1970 		}
1971 		p->p_fdtol = NULL;
1972 		if (fdtol != NULL) {
1973 			spin_unlock(&fdp->fd_spin);
1974 			kfree(fdtol, M_FILEDESC_TO_LEADER);
1975 			spin_lock(&fdp->fd_spin);
1976 		}
1977 	}
1978 	if (--fdp->fd_refcnt > 0) {
1979 		spin_unlock(&fdp->fd_spin);
1980 		spin_lock(&p->p_spin);
1981 		p->p_fd = repl;
1982 		spin_unlock(&p->p_spin);
1983 		return;
1984 	}
1985 
1986 	/*
1987 	 * Even though we are the last reference to the structure allproc
1988 	 * scans may still reference the structure.  Maintain proper
1989 	 * locks until we can replace p->p_fd.
1990 	 *
1991 	 * Also note that kqueue's closef still needs to reference the
1992 	 * fdp via p->p_fd, so we have to close the descriptors before
1993 	 * we replace p->p_fd.
1994 	 */
1995 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1996 		if (fdp->fd_files[i].fp) {
1997 			fp = funsetfd_locked(fdp, i);
1998 			if (fp) {
1999 				spin_unlock(&fdp->fd_spin);
2000 				if (SLIST_FIRST(&fp->f_klist))
2001 					knote_fdclose(fp, fdp, i);
2002 				closef(fp, p);
2003 				spin_lock(&fdp->fd_spin);
2004 			}
2005 		}
2006 	}
2007 	spin_unlock(&fdp->fd_spin);
2008 
2009 	/*
2010 	 * Interlock against an allproc scan operations (typically frevoke).
2011 	 */
2012 	spin_lock(&p->p_spin);
2013 	p->p_fd = repl;
2014 	spin_unlock(&p->p_spin);
2015 
2016 	/*
2017 	 * Wait for any softrefs to go away.  This race rarely occurs so
2018 	 * we can use a non-critical-path style poll/sleep loop.  The
2019 	 * race only occurs against allproc scans.
2020 	 *
2021 	 * No new softrefs can occur with the fdp disconnected from the
2022 	 * process.
2023 	 */
2024 	if (fdp->fd_softrefs) {
2025 		kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2026 		while (fdp->fd_softrefs)
2027 			tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2028 	}
2029 
2030 	if (fdp->fd_files != fdp->fd_builtin_files)
2031 		kfree(fdp->fd_files, M_FILEDESC);
2032 	if (fdp->fd_cdir) {
2033 		cache_drop(&fdp->fd_ncdir);
2034 		vrele(fdp->fd_cdir);
2035 	}
2036 	if (fdp->fd_rdir) {
2037 		cache_drop(&fdp->fd_nrdir);
2038 		vrele(fdp->fd_rdir);
2039 	}
2040 	if (fdp->fd_jdir) {
2041 		cache_drop(&fdp->fd_njdir);
2042 		vrele(fdp->fd_jdir);
2043 	}
2044 	kfree(fdp, M_FILEDESC);
2045 }
2046 
2047 /*
2048  * Retrieve and reference the file pointer associated with a descriptor.
2049  *
2050  * MPSAFE
2051  */
2052 struct file *
2053 holdfp(struct filedesc *fdp, int fd, int flag)
2054 {
2055 	struct file* fp;
2056 
2057 	spin_lock(&fdp->fd_spin);
2058 	if (((u_int)fd) >= fdp->fd_nfiles) {
2059 		fp = NULL;
2060 		goto done;
2061 	}
2062 	if ((fp = fdp->fd_files[fd].fp) == NULL)
2063 		goto done;
2064 	if ((fp->f_flag & flag) == 0 && flag != -1) {
2065 		fp = NULL;
2066 		goto done;
2067 	}
2068 	fhold(fp);
2069 done:
2070 	spin_unlock(&fdp->fd_spin);
2071 	return (fp);
2072 }
2073 
2074 /*
2075  * holdsock() - load the struct file pointer associated
2076  * with a socket into *fpp.  If an error occurs, non-zero
2077  * will be returned and *fpp will be set to NULL.
2078  *
2079  * MPSAFE
2080  */
2081 int
2082 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2083 {
2084 	struct file *fp;
2085 	int error;
2086 
2087 	spin_lock(&fdp->fd_spin);
2088 	if ((unsigned)fd >= fdp->fd_nfiles) {
2089 		error = EBADF;
2090 		fp = NULL;
2091 		goto done;
2092 	}
2093 	if ((fp = fdp->fd_files[fd].fp) == NULL) {
2094 		error = EBADF;
2095 		goto done;
2096 	}
2097 	if (fp->f_type != DTYPE_SOCKET) {
2098 		error = ENOTSOCK;
2099 		goto done;
2100 	}
2101 	fhold(fp);
2102 	error = 0;
2103 done:
2104 	spin_unlock(&fdp->fd_spin);
2105 	*fpp = fp;
2106 	return (error);
2107 }
2108 
2109 /*
2110  * Convert a user file descriptor to a held file pointer.
2111  *
2112  * MPSAFE
2113  */
2114 int
2115 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2116 {
2117 	struct file *fp;
2118 	int error;
2119 
2120 	spin_lock(&fdp->fd_spin);
2121 	if ((unsigned)fd >= fdp->fd_nfiles) {
2122 		error = EBADF;
2123 		fp = NULL;
2124 		goto done;
2125 	}
2126 	if ((fp = fdp->fd_files[fd].fp) == NULL) {
2127 		error = EBADF;
2128 		goto done;
2129 	}
2130 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2131 		fp = NULL;
2132 		error = EINVAL;
2133 		goto done;
2134 	}
2135 	fhold(fp);
2136 	error = 0;
2137 done:
2138 	spin_unlock(&fdp->fd_spin);
2139 	*fpp = fp;
2140 	return (error);
2141 }
2142 
2143 /*
2144  * For setugid programs, we don't want to people to use that setugidness
2145  * to generate error messages which write to a file which otherwise would
2146  * otherwise be off-limits to the process.
2147  *
2148  * This is a gross hack to plug the hole.  A better solution would involve
2149  * a special vop or other form of generalized access control mechanism.  We
2150  * go ahead and just reject all procfs file systems accesses as dangerous.
2151  *
2152  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2153  * sufficient.  We also don't for check setugidness since we know we are.
2154  */
2155 static int
2156 is_unsafe(struct file *fp)
2157 {
2158 	if (fp->f_type == DTYPE_VNODE &&
2159 	    ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2160 		return (1);
2161 	return (0);
2162 }
2163 
2164 /*
2165  * Make this setguid thing safe, if at all possible.
2166  *
2167  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2168  */
2169 void
2170 setugidsafety(struct proc *p)
2171 {
2172 	struct filedesc *fdp = p->p_fd;
2173 	int i;
2174 
2175 	/* Certain daemons might not have file descriptors. */
2176 	if (fdp == NULL)
2177 		return;
2178 
2179 	/*
2180 	 * note: fdp->fd_files may be reallocated out from under us while
2181 	 * we are blocked in a close.  Be careful!
2182 	 */
2183 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2184 		if (i > 2)
2185 			break;
2186 		if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2187 			struct file *fp;
2188 
2189 			/*
2190 			 * NULL-out descriptor prior to close to avoid
2191 			 * a race while close blocks.
2192 			 */
2193 			if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2194 				knote_fdclose(fp, fdp, i);
2195 				closef(fp, p);
2196 			}
2197 		}
2198 	}
2199 }
2200 
2201 /*
2202  * Close any files on exec?
2203  *
2204  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2205  */
2206 void
2207 fdcloseexec(struct proc *p)
2208 {
2209 	struct filedesc *fdp = p->p_fd;
2210 	int i;
2211 
2212 	/* Certain daemons might not have file descriptors. */
2213 	if (fdp == NULL)
2214 		return;
2215 
2216 	/*
2217 	 * We cannot cache fd_files since operations may block and rip
2218 	 * them out from under us.
2219 	 */
2220 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2221 		if (fdp->fd_files[i].fp != NULL &&
2222 		    (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2223 			struct file *fp;
2224 
2225 			/*
2226 			 * NULL-out descriptor prior to close to avoid
2227 			 * a race while close blocks.
2228 			 */
2229 			if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2230 				knote_fdclose(fp, fdp, i);
2231 				closef(fp, p);
2232 			}
2233 		}
2234 	}
2235 }
2236 
2237 /*
2238  * It is unsafe for set[ug]id processes to be started with file
2239  * descriptors 0..2 closed, as these descriptors are given implicit
2240  * significance in the Standard C library.  fdcheckstd() will create a
2241  * descriptor referencing /dev/null for each of stdin, stdout, and
2242  * stderr that is not already open.
2243  *
2244  * NOT MPSAFE - calls falloc, vn_open, etc
2245  */
2246 int
2247 fdcheckstd(struct lwp *lp)
2248 {
2249 	struct nlookupdata nd;
2250 	struct filedesc *fdp;
2251 	struct file *fp;
2252 	int retval;
2253 	int i, error, flags, devnull;
2254 
2255 	fdp = lp->lwp_proc->p_fd;
2256 	if (fdp == NULL)
2257 		return (0);
2258 	devnull = -1;
2259 	error = 0;
2260 	for (i = 0; i < 3; i++) {
2261 		if (fdp->fd_files[i].fp != NULL)
2262 			continue;
2263 		if (devnull < 0) {
2264 			if ((error = falloc(lp, &fp, &devnull)) != 0)
2265 				break;
2266 
2267 			error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2268 						NLC_FOLLOW|NLC_LOCKVP);
2269 			flags = FREAD | FWRITE;
2270 			if (error == 0)
2271 				error = vn_open(&nd, fp, flags, 0);
2272 			if (error == 0)
2273 				fsetfd(fdp, fp, devnull);
2274 			else
2275 				fsetfd(fdp, NULL, devnull);
2276 			fdrop(fp);
2277 			nlookup_done(&nd);
2278 			if (error)
2279 				break;
2280 			KKASSERT(i == devnull);
2281 		} else {
2282 			error = kern_dup(DUP_FIXED, devnull, i, &retval);
2283 			if (error != 0)
2284 				break;
2285 		}
2286 	}
2287 	return (error);
2288 }
2289 
2290 /*
2291  * Internal form of close.
2292  * Decrement reference count on file structure.
2293  * Note: td and/or p may be NULL when closing a file
2294  * that was being passed in a message.
2295  *
2296  * MPALMOSTSAFE - acquires mplock for VOP operations
2297  */
2298 int
2299 closef(struct file *fp, struct proc *p)
2300 {
2301 	struct vnode *vp;
2302 	struct flock lf;
2303 	struct filedesc_to_leader *fdtol;
2304 
2305 	if (fp == NULL)
2306 		return (0);
2307 
2308 	/*
2309 	 * POSIX record locking dictates that any close releases ALL
2310 	 * locks owned by this process.  This is handled by setting
2311 	 * a flag in the unlock to free ONLY locks obeying POSIX
2312 	 * semantics, and not to free BSD-style file locks.
2313 	 * If the descriptor was in a message, POSIX-style locks
2314 	 * aren't passed with the descriptor.
2315 	 */
2316 	if (p != NULL && fp->f_type == DTYPE_VNODE &&
2317 	    (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2318 	) {
2319 		get_mplock();
2320 		if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2321 			lf.l_whence = SEEK_SET;
2322 			lf.l_start = 0;
2323 			lf.l_len = 0;
2324 			lf.l_type = F_UNLCK;
2325 			vp = (struct vnode *)fp->f_data;
2326 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2327 					   &lf, F_POSIX);
2328 		}
2329 		fdtol = p->p_fdtol;
2330 		if (fdtol != NULL) {
2331 			/*
2332 			 * Handle special case where file descriptor table
2333 			 * is shared between multiple process leaders.
2334 			 */
2335 			for (fdtol = fdtol->fdl_next;
2336 			     fdtol != p->p_fdtol;
2337 			     fdtol = fdtol->fdl_next) {
2338 				if ((fdtol->fdl_leader->p_flag &
2339 				     P_ADVLOCK) == 0)
2340 					continue;
2341 				fdtol->fdl_holdcount++;
2342 				lf.l_whence = SEEK_SET;
2343 				lf.l_start = 0;
2344 				lf.l_len = 0;
2345 				lf.l_type = F_UNLCK;
2346 				vp = (struct vnode *)fp->f_data;
2347 				(void) VOP_ADVLOCK(vp,
2348 						   (caddr_t)fdtol->fdl_leader,
2349 						   F_UNLCK, &lf, F_POSIX);
2350 				fdtol->fdl_holdcount--;
2351 				if (fdtol->fdl_holdcount == 0 &&
2352 				    fdtol->fdl_wakeup != 0) {
2353 					fdtol->fdl_wakeup = 0;
2354 					wakeup(fdtol);
2355 				}
2356 			}
2357 		}
2358 		rel_mplock();
2359 	}
2360 	return (fdrop(fp));
2361 }
2362 
2363 /*
2364  * MPSAFE
2365  *
2366  * fhold() can only be called if f_count is already at least 1 (i.e. the
2367  * caller of fhold() already has a reference to the file pointer in some
2368  * manner or other).
2369  *
2370  * f_count is not spin-locked.  Instead, atomic ops are used for
2371  * incrementing, decrementing, and handling the 1->0 transition.
2372  */
2373 void
2374 fhold(struct file *fp)
2375 {
2376 	atomic_add_int(&fp->f_count, 1);
2377 }
2378 
2379 /*
2380  * fdrop() - drop a reference to a descriptor
2381  *
2382  * MPALMOSTSAFE - acquires mplock for final close sequence
2383  */
2384 int
2385 fdrop(struct file *fp)
2386 {
2387 	struct flock lf;
2388 	struct vnode *vp;
2389 	int error;
2390 
2391 	/*
2392 	 * A combined fetch and subtract is needed to properly detect
2393 	 * 1->0 transitions, otherwise two cpus dropping from a ref
2394 	 * count of 2 might both try to run the 1->0 code.
2395 	 */
2396 	if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2397 		return (0);
2398 
2399 	KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
2400 	get_mplock();
2401 
2402 	/*
2403 	 * The last reference has gone away, we own the fp structure free
2404 	 * and clear.
2405 	 */
2406 	if (fp->f_count < 0)
2407 		panic("fdrop: count < 0");
2408 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2409 	    (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2410 	) {
2411 		lf.l_whence = SEEK_SET;
2412 		lf.l_start = 0;
2413 		lf.l_len = 0;
2414 		lf.l_type = F_UNLCK;
2415 		vp = (struct vnode *)fp->f_data;
2416 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2417 	}
2418 	if (fp->f_ops != &badfileops)
2419 		error = fo_close(fp);
2420 	else
2421 		error = 0;
2422 	ffree(fp);
2423 	rel_mplock();
2424 	return (error);
2425 }
2426 
2427 /*
2428  * Apply an advisory lock on a file descriptor.
2429  *
2430  * Just attempt to get a record lock of the requested type on
2431  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2432  *
2433  * MPALMOSTSAFE
2434  */
2435 int
2436 sys_flock(struct flock_args *uap)
2437 {
2438 	struct proc *p = curproc;
2439 	struct file *fp;
2440 	struct vnode *vp;
2441 	struct flock lf;
2442 	int error;
2443 
2444 	if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2445 		return (EBADF);
2446 	get_mplock();
2447 	if (fp->f_type != DTYPE_VNODE) {
2448 		error = EOPNOTSUPP;
2449 		goto done;
2450 	}
2451 	vp = (struct vnode *)fp->f_data;
2452 	lf.l_whence = SEEK_SET;
2453 	lf.l_start = 0;
2454 	lf.l_len = 0;
2455 	if (uap->how & LOCK_UN) {
2456 		lf.l_type = F_UNLCK;
2457 		fp->f_flag &= ~FHASLOCK;
2458 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2459 		goto done;
2460 	}
2461 	if (uap->how & LOCK_EX)
2462 		lf.l_type = F_WRLCK;
2463 	else if (uap->how & LOCK_SH)
2464 		lf.l_type = F_RDLCK;
2465 	else {
2466 		error = EBADF;
2467 		goto done;
2468 	}
2469 	fp->f_flag |= FHASLOCK;
2470 	if (uap->how & LOCK_NB)
2471 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2472 	else
2473 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2474 done:
2475 	rel_mplock();
2476 	fdrop(fp);
2477 	return (error);
2478 }
2479 
2480 /*
2481  * File Descriptor pseudo-device driver (/dev/fd/).
2482  *
2483  * Opening minor device N dup()s the file (if any) connected to file
2484  * descriptor N belonging to the calling process.  Note that this driver
2485  * consists of only the ``open()'' routine, because all subsequent
2486  * references to this file will be direct to the other driver.
2487  */
2488 static int
2489 fdopen(struct dev_open_args *ap)
2490 {
2491 	thread_t td = curthread;
2492 
2493 	KKASSERT(td->td_lwp != NULL);
2494 
2495 	/*
2496 	 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2497 	 * the file descriptor being sought for duplication. The error
2498 	 * return ensures that the vnode for this device will be released
2499 	 * by vn_open. Open will detect this special error and take the
2500 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2501 	 * will simply report the error.
2502 	 */
2503 	td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2504 	return (ENODEV);
2505 }
2506 
2507 /*
2508  * The caller has reserved the file descriptor dfd for us.  On success we
2509  * must fsetfd() it.  On failure the caller will clean it up.
2510  *
2511  * MPSAFE
2512  */
2513 int
2514 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2515 {
2516 	struct file *wfp;
2517 	struct file *xfp;
2518 	int werror;
2519 
2520 	if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2521 		return (EBADF);
2522 
2523 	/*
2524 	 * Close a revoke/dup race.  Duping a descriptor marked as revoked
2525 	 * will dup a dummy descriptor instead of the real one.
2526 	 */
2527 	if (wfp->f_flag & FREVOKED) {
2528 		kprintf("Warning: attempt to dup() a revoked descriptor\n");
2529 		fdrop(wfp);
2530 		wfp = NULL;
2531 		werror = falloc(NULL, &wfp, NULL);
2532 		if (werror)
2533 			return (werror);
2534 	}
2535 
2536 	/*
2537 	 * There are two cases of interest here.
2538 	 *
2539 	 * For ENODEV simply dup sfd to file descriptor dfd and return.
2540 	 *
2541 	 * For ENXIO steal away the file structure from sfd and store it
2542 	 * dfd.  sfd is effectively closed by this operation.
2543 	 *
2544 	 * Any other error code is just returned.
2545 	 */
2546 	switch (error) {
2547 	case ENODEV:
2548 		/*
2549 		 * Check that the mode the file is being opened for is a
2550 		 * subset of the mode of the existing descriptor.
2551 		 */
2552 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2553 			error = EACCES;
2554 			break;
2555 		}
2556 		spin_lock(&fdp->fd_spin);
2557 		fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2558 		fsetfd_locked(fdp, wfp, dfd);
2559 		spin_unlock(&fdp->fd_spin);
2560 		error = 0;
2561 		break;
2562 	case ENXIO:
2563 		/*
2564 		 * Steal away the file pointer from dfd, and stuff it into indx.
2565 		 */
2566 		spin_lock(&fdp->fd_spin);
2567 		fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2568 		fsetfd(fdp, wfp, dfd);
2569 		if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
2570 			spin_unlock(&fdp->fd_spin);
2571 			fdrop(xfp);
2572 		} else {
2573 			spin_unlock(&fdp->fd_spin);
2574 		}
2575 		error = 0;
2576 		break;
2577 	default:
2578 		break;
2579 	}
2580 	fdrop(wfp);
2581 	return (error);
2582 }
2583 
2584 /*
2585  * NOT MPSAFE - I think these refer to a common file descriptor table
2586  * and we need to spinlock that to link fdtol in.
2587  */
2588 struct filedesc_to_leader *
2589 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2590 			 struct proc *leader)
2591 {
2592 	struct filedesc_to_leader *fdtol;
2593 
2594 	fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2595 			M_FILEDESC_TO_LEADER, M_WAITOK);
2596 	fdtol->fdl_refcount = 1;
2597 	fdtol->fdl_holdcount = 0;
2598 	fdtol->fdl_wakeup = 0;
2599 	fdtol->fdl_leader = leader;
2600 	if (old != NULL) {
2601 		fdtol->fdl_next = old->fdl_next;
2602 		fdtol->fdl_prev = old;
2603 		old->fdl_next = fdtol;
2604 		fdtol->fdl_next->fdl_prev = fdtol;
2605 	} else {
2606 		fdtol->fdl_next = fdtol;
2607 		fdtol->fdl_prev = fdtol;
2608 	}
2609 	return fdtol;
2610 }
2611 
2612 /*
2613  * Scan all file pointers in the system.  The callback is made with
2614  * the master list spinlock held exclusively.
2615  *
2616  * MPSAFE
2617  */
2618 void
2619 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2620 {
2621 	struct file *fp;
2622 	int res;
2623 
2624 	spin_lock(&filehead_spin);
2625 	LIST_FOREACH(fp, &filehead, f_list) {
2626 		res = callback(fp, data);
2627 		if (res < 0)
2628 			break;
2629 	}
2630 	spin_unlock(&filehead_spin);
2631 }
2632 
2633 /*
2634  * Get file structures.
2635  *
2636  * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2637  */
2638 
2639 struct sysctl_kern_file_info {
2640 	int count;
2641 	int error;
2642 	struct sysctl_req *req;
2643 };
2644 
2645 static int sysctl_kern_file_callback(struct proc *p, void *data);
2646 
2647 static int
2648 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2649 {
2650 	struct sysctl_kern_file_info info;
2651 
2652 	/*
2653 	 * Note: because the number of file descriptors is calculated
2654 	 * in different ways for sizing vs returning the data,
2655 	 * there is information leakage from the first loop.  However,
2656 	 * it is of a similar order of magnitude to the leakage from
2657 	 * global system statistics such as kern.openfiles.
2658 	 *
2659 	 * When just doing a count, note that we cannot just count
2660 	 * the elements and add f_count via the filehead list because
2661 	 * threaded processes share their descriptor table and f_count might
2662 	 * still be '1' in that case.
2663 	 *
2664 	 * Since the SYSCTL op can block, we must hold the process to
2665 	 * prevent it being ripped out from under us either in the
2666 	 * file descriptor loop or in the greater LIST_FOREACH.  The
2667 	 * process may be in varying states of disrepair.  If the process
2668 	 * is in SZOMB we may have caught it just as it is being removed
2669 	 * from the allproc list, we must skip it in that case to maintain
2670 	 * an unbroken chain through the allproc list.
2671 	 */
2672 	info.count = 0;
2673 	info.error = 0;
2674 	info.req = req;
2675 	allproc_scan(sysctl_kern_file_callback, &info);
2676 
2677 	/*
2678 	 * When just calculating the size, overestimate a bit to try to
2679 	 * prevent system activity from causing the buffer-fill call
2680 	 * to fail later on.
2681 	 */
2682 	if (req->oldptr == NULL) {
2683 		info.count = (info.count + 16) + (info.count / 10);
2684 		info.error = SYSCTL_OUT(req, NULL,
2685 					info.count * sizeof(struct kinfo_file));
2686 	}
2687 	return (info.error);
2688 }
2689 
2690 static int
2691 sysctl_kern_file_callback(struct proc *p, void *data)
2692 {
2693 	struct sysctl_kern_file_info *info = data;
2694 	struct kinfo_file kf;
2695 	struct filedesc *fdp;
2696 	struct file *fp;
2697 	uid_t uid;
2698 	int n;
2699 
2700 	if (p->p_stat == SIDL || p->p_stat == SZOMB)
2701 		return(0);
2702 	if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2703 		return(0);
2704 
2705 	/*
2706 	 * Softref the fdp to prevent it from being destroyed
2707 	 */
2708 	spin_lock(&p->p_spin);
2709 	if ((fdp = p->p_fd) == NULL) {
2710 		spin_unlock(&p->p_spin);
2711 		return(0);
2712 	}
2713 	atomic_add_int(&fdp->fd_softrefs, 1);
2714 	spin_unlock(&p->p_spin);
2715 
2716 	/*
2717 	 * The fdp's own spinlock prevents the contents from being
2718 	 * modified.
2719 	 */
2720 	spin_lock(&fdp->fd_spin);
2721 	for (n = 0; n < fdp->fd_nfiles; ++n) {
2722 		if ((fp = fdp->fd_files[n].fp) == NULL)
2723 			continue;
2724 		if (info->req->oldptr == NULL) {
2725 			++info->count;
2726 		} else {
2727 			uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2728 			kcore_make_file(&kf, fp, p->p_pid, uid, n);
2729 			spin_unlock(&fdp->fd_spin);
2730 			info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2731 			spin_lock(&fdp->fd_spin);
2732 			if (info->error)
2733 				break;
2734 		}
2735 	}
2736 	spin_unlock(&fdp->fd_spin);
2737 	atomic_subtract_int(&fdp->fd_softrefs, 1);
2738 	if (info->error)
2739 		return(-1);
2740 	return(0);
2741 }
2742 
2743 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2744     0, 0, sysctl_kern_file, "S,file", "Entire file table");
2745 
2746 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2747     &minfilesperproc, 0, "Minimum files allowed open per process");
2748 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2749     &maxfilesperproc, 0, "Maximum files allowed open per process");
2750 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2751     &maxfilesperuser, 0, "Maximum files allowed open per user");
2752 
2753 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2754     &maxfiles, 0, "Maximum number of files");
2755 
2756 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2757     &maxfilesrootres, 0, "Descriptors reserved for root use");
2758 
2759 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2760 	&nfiles, 0, "System-wide number of open files");
2761 
2762 static void
2763 fildesc_drvinit(void *unused)
2764 {
2765 	int fd;
2766 
2767 	for (fd = 0; fd < NUMFDESC; fd++) {
2768 		make_dev(&fildesc_ops, fd,
2769 			 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2770 	}
2771 
2772 	make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2773 	make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2774 	make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2775 }
2776 
2777 /*
2778  * MPSAFE
2779  */
2780 struct fileops badfileops = {
2781 	.fo_read = badfo_readwrite,
2782 	.fo_write = badfo_readwrite,
2783 	.fo_ioctl = badfo_ioctl,
2784 	.fo_poll = badfo_poll,
2785 	.fo_kqfilter = badfo_kqfilter,
2786 	.fo_stat = badfo_stat,
2787 	.fo_close = badfo_close,
2788 	.fo_shutdown = badfo_shutdown
2789 };
2790 
2791 int
2792 badfo_readwrite(
2793 	struct file *fp,
2794 	struct uio *uio,
2795 	struct ucred *cred,
2796 	int flags
2797 ) {
2798 	return (EBADF);
2799 }
2800 
2801 int
2802 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2803 	    struct ucred *cred, struct sysmsg *msgv)
2804 {
2805 	return (EBADF);
2806 }
2807 
2808 int
2809 badfo_poll(struct file *fp, int events, struct ucred *cred)
2810 {
2811 	return (0);
2812 }
2813 
2814 /*
2815  * Must return an error to prevent registration, typically
2816  * due to a revoked descriptor (file_filtops assigned).
2817  */
2818 int
2819 badfo_kqfilter(struct file *fp, struct knote *kn)
2820 {
2821 	return (EOPNOTSUPP);
2822 }
2823 
2824 /*
2825  * MPSAFE
2826  */
2827 int
2828 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2829 {
2830 	return (EBADF);
2831 }
2832 
2833 /*
2834  * MPSAFE
2835  */
2836 int
2837 badfo_close(struct file *fp)
2838 {
2839 	return (EBADF);
2840 }
2841 
2842 /*
2843  * MPSAFE
2844  */
2845 int
2846 badfo_shutdown(struct file *fp, int how)
2847 {
2848 	return (EBADF);
2849 }
2850 
2851 /*
2852  * MPSAFE
2853  */
2854 int
2855 nofo_shutdown(struct file *fp, int how)
2856 {
2857 	return (EOPNOTSUPP);
2858 }
2859 
2860 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2861 					fildesc_drvinit,NULL)
2862