xref: /dflybsd-src/sys/kern/kern_descrip.c (revision 308af577f531e89dce6dff0ecea68ab4474f4d24)
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  * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1324  */
1325 static int
1326 fdrevoke_check_callback(struct file *fp, void *vinfo)
1327 {
1328 	struct fdrevoke_info *info = vinfo;
1329 
1330 	/*
1331 	 * File pointers already flagged for revokation are skipped.
1332 	 */
1333 	if (fp->f_flag & FREVOKED)
1334 		return(0);
1335 
1336 	/*
1337 	 * If revoking from a prison file pointers created outside of
1338 	 * that prison, or file pointers without creds, cannot be revoked.
1339 	 */
1340 	if (info->cred->cr_prison &&
1341 	    (fp->f_cred == NULL ||
1342 	     info->cred->cr_prison != fp->f_cred->cr_prison)) {
1343 		return(0);
1344 	}
1345 
1346 	/*
1347 	 * If the file pointer matches then mark it for revocation.  The
1348 	 * flag is currently only used by unp_revoke_gc().
1349 	 *
1350 	 * info->count is a heuristic and can race in a SMP environment.
1351 	 */
1352 	if (info->data == fp->f_data && info->type == fp->f_type) {
1353 		atomic_set_int(&fp->f_flag, FREVOKED);
1354 		info->count += fp->f_count;
1355 		if (fp->f_msgcount)
1356 			++info->intransit;
1357 	}
1358 	return(0);
1359 }
1360 
1361 /*
1362  * Locate matching file pointers via process descriptor tables.
1363  */
1364 static int
1365 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1366 {
1367 	struct fdrevoke_info *info = vinfo;
1368 	struct filedesc *fdp;
1369 	struct file *fp;
1370 	int n;
1371 
1372 	if (p->p_stat == SIDL || p->p_stat == SZOMB)
1373 		return(0);
1374 	if (info->cred->cr_prison &&
1375 	    info->cred->cr_prison != p->p_ucred->cr_prison) {
1376 		return(0);
1377 	}
1378 
1379 	/*
1380 	 * If the controlling terminal of the process matches the
1381 	 * vnode being revoked we clear the controlling terminal.
1382 	 *
1383 	 * The normal spec_close() may not catch this because it
1384 	 * uses curproc instead of p.
1385 	 */
1386 	if (p->p_session && info->type == DTYPE_VNODE &&
1387 	    info->data == p->p_session->s_ttyvp) {
1388 		p->p_session->s_ttyvp = NULL;
1389 		vrele(info->data);
1390 	}
1391 
1392 	/*
1393 	 * Softref the fdp to prevent it from being destroyed
1394 	 */
1395 	spin_lock(&p->p_spin);
1396 	if ((fdp = p->p_fd) == NULL) {
1397 		spin_unlock(&p->p_spin);
1398 		return(0);
1399 	}
1400 	atomic_add_int(&fdp->fd_softrefs, 1);
1401 	spin_unlock(&p->p_spin);
1402 
1403 	/*
1404 	 * Locate and close any matching file descriptors.
1405 	 */
1406 	spin_lock(&fdp->fd_spin);
1407 	for (n = 0; n < fdp->fd_nfiles; ++n) {
1408 		if ((fp = fdp->fd_files[n].fp) == NULL)
1409 			continue;
1410 		if (fp->f_flag & FREVOKED) {
1411 			fhold(info->nfp);
1412 			fdp->fd_files[n].fp = info->nfp;
1413 			spin_unlock(&fdp->fd_spin);
1414 			knote_fdclose(fp, fdp, n);	/* XXX */
1415 			closef(fp, p);
1416 			spin_lock(&fdp->fd_spin);
1417 			--info->count;
1418 		}
1419 	}
1420 	spin_unlock(&fdp->fd_spin);
1421 	atomic_subtract_int(&fdp->fd_softrefs, 1);
1422 	return(0);
1423 }
1424 
1425 /*
1426  * falloc:
1427  *	Create a new open file structure and reserve a file decriptor
1428  *	for the process that refers to it.
1429  *
1430  *	Root creds are checked using lp, or assumed if lp is NULL.  If
1431  *	resultfd is non-NULL then lp must also be non-NULL.  No file
1432  *	descriptor is reserved (and no process context is needed) if
1433  *	resultfd is NULL.
1434  *
1435  *	A file pointer with a refcount of 1 is returned.  Note that the
1436  *	file pointer is NOT associated with the descriptor.  If falloc
1437  *	returns success, fsetfd() MUST be called to either associate the
1438  *	file pointer or clear the reservation.
1439  *
1440  * MPSAFE
1441  */
1442 int
1443 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1444 {
1445 	static struct timeval lastfail;
1446 	static int curfail;
1447 	struct file *fp;
1448 	struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1449 	int error;
1450 
1451 	fp = NULL;
1452 
1453 	/*
1454 	 * Handle filetable full issues and root overfill.
1455 	 */
1456 	if (nfiles >= maxfiles - maxfilesrootres &&
1457 	    (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1458 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1459 			kprintf("kern.maxfiles limit exceeded by uid %d, "
1460 				"please see tuning(7).\n",
1461 				cred->cr_ruid);
1462 		}
1463 		error = ENFILE;
1464 		goto done;
1465 	}
1466 
1467 	/*
1468 	 * Allocate a new file descriptor.
1469 	 */
1470 	fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1471 	spin_init(&fp->f_spin);
1472 	SLIST_INIT(&fp->f_klist);
1473 	fp->f_count = 1;
1474 	fp->f_ops = &badfileops;
1475 	fp->f_seqcount = 1;
1476 	fsetcred(fp, cred);
1477 	spin_lock(&filehead_spin);
1478 	nfiles++;
1479 	LIST_INSERT_HEAD(&filehead, fp, f_list);
1480 	spin_unlock(&filehead_spin);
1481 	if (resultfd) {
1482 		if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1483 			fdrop(fp);
1484 			fp = NULL;
1485 		}
1486 	} else {
1487 		error = 0;
1488 	}
1489 done:
1490 	*resultfp = fp;
1491 	return (error);
1492 }
1493 
1494 /*
1495  * Check for races against a file descriptor by determining that the
1496  * file pointer is still associated with the specified file descriptor,
1497  * and a close is not currently in progress.
1498  *
1499  * MPSAFE
1500  */
1501 int
1502 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
1503 {
1504 	int error;
1505 
1506 	spin_lock(&fdp->fd_spin);
1507 	if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1508 		error = EBADF;
1509 	else
1510 		error = 0;
1511 	spin_unlock(&fdp->fd_spin);
1512 	return (error);
1513 }
1514 
1515 /*
1516  * Associate a file pointer with a previously reserved file descriptor.
1517  * This function always succeeds.
1518  *
1519  * If fp is NULL, the file descriptor is returned to the pool.
1520  */
1521 
1522 /*
1523  * MPSAFE (exclusive spinlock must be held on call)
1524  */
1525 static void
1526 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1527 {
1528 	KKASSERT((unsigned)fd < fdp->fd_nfiles);
1529 	KKASSERT(fdp->fd_files[fd].reserved != 0);
1530 	if (fp) {
1531 		fhold(fp);
1532 		fdp->fd_files[fd].fp = fp;
1533 		fdp->fd_files[fd].reserved = 0;
1534 	} else {
1535 		fdp->fd_files[fd].reserved = 0;
1536 		fdreserve_locked(fdp, fd, -1);
1537 		fdfixup_locked(fdp, fd);
1538 	}
1539 }
1540 
1541 /*
1542  * MPSAFE
1543  */
1544 void
1545 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1546 {
1547 	spin_lock(&fdp->fd_spin);
1548 	fsetfd_locked(fdp, fp, fd);
1549 	spin_unlock(&fdp->fd_spin);
1550 }
1551 
1552 /*
1553  * MPSAFE (exclusive spinlock must be held on call)
1554  */
1555 static
1556 struct file *
1557 funsetfd_locked(struct filedesc *fdp, int fd)
1558 {
1559 	struct file *fp;
1560 
1561 	if ((unsigned)fd >= fdp->fd_nfiles)
1562 		return (NULL);
1563 	if ((fp = fdp->fd_files[fd].fp) == NULL)
1564 		return (NULL);
1565 	fdp->fd_files[fd].fp = NULL;
1566 	fdp->fd_files[fd].fileflags = 0;
1567 
1568 	fdreserve_locked(fdp, fd, -1);
1569 	fdfixup_locked(fdp, fd);
1570 	return(fp);
1571 }
1572 
1573 /*
1574  * MPSAFE
1575  */
1576 int
1577 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1578 {
1579 	int error;
1580 
1581 	spin_lock(&fdp->fd_spin);
1582 	if (((u_int)fd) >= fdp->fd_nfiles) {
1583 		error = EBADF;
1584 	} else if (fdp->fd_files[fd].fp == NULL) {
1585 		error = EBADF;
1586 	} else {
1587 		*flagsp = fdp->fd_files[fd].fileflags;
1588 		error = 0;
1589 	}
1590 	spin_unlock(&fdp->fd_spin);
1591 	return (error);
1592 }
1593 
1594 /*
1595  * MPSAFE
1596  */
1597 int
1598 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1599 {
1600 	int error;
1601 
1602 	spin_lock(&fdp->fd_spin);
1603 	if (((u_int)fd) >= fdp->fd_nfiles) {
1604 		error = EBADF;
1605 	} else if (fdp->fd_files[fd].fp == NULL) {
1606 		error = EBADF;
1607 	} else {
1608 		fdp->fd_files[fd].fileflags |= add_flags;
1609 		error = 0;
1610 	}
1611 	spin_unlock(&fdp->fd_spin);
1612 	return (error);
1613 }
1614 
1615 /*
1616  * MPSAFE
1617  */
1618 int
1619 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1620 {
1621 	int error;
1622 
1623 	spin_lock(&fdp->fd_spin);
1624 	if (((u_int)fd) >= fdp->fd_nfiles) {
1625 		error = EBADF;
1626 	} else if (fdp->fd_files[fd].fp == NULL) {
1627 		error = EBADF;
1628 	} else {
1629 		fdp->fd_files[fd].fileflags &= ~rem_flags;
1630 		error = 0;
1631 	}
1632 	spin_unlock(&fdp->fd_spin);
1633 	return (error);
1634 }
1635 
1636 /*
1637  * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1638  */
1639 void
1640 fsetcred(struct file *fp, struct ucred *ncr)
1641 {
1642 	struct ucred *ocr;
1643 	struct uidinfo *uip;
1644 
1645 	ocr = fp->f_cred;
1646 	if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1647 		if (ocr) {
1648 			uip = ocr->cr_uidinfo;
1649 			atomic_add_int(&uip->ui_openfiles, -1);
1650 		}
1651 		if (ncr) {
1652 			uip = ncr->cr_uidinfo;
1653 			atomic_add_int(&uip->ui_openfiles, 1);
1654 		}
1655 	}
1656 	if (ncr)
1657 		crhold(ncr);
1658 	fp->f_cred = ncr;
1659 	if (ocr)
1660 		crfree(ocr);
1661 }
1662 
1663 /*
1664  * Free a file descriptor.
1665  */
1666 static
1667 void
1668 ffree(struct file *fp)
1669 {
1670 	KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1671 	spin_lock(&filehead_spin);
1672 	LIST_REMOVE(fp, f_list);
1673 	nfiles--;
1674 	spin_unlock(&filehead_spin);
1675 	fsetcred(fp, NULL);
1676 	if (fp->f_nchandle.ncp)
1677 	    cache_drop(&fp->f_nchandle);
1678 	kfree(fp, M_FILE);
1679 }
1680 
1681 /*
1682  * called from init_main, initialize filedesc0 for proc0.
1683  */
1684 void
1685 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1686 {
1687 	p0->p_fd = fdp0;
1688 	p0->p_fdtol = NULL;
1689 	fdp0->fd_refcnt = 1;
1690 	fdp0->fd_cmask = cmask;
1691 	fdp0->fd_files = fdp0->fd_builtin_files;
1692 	fdp0->fd_nfiles = NDFILE;
1693 	fdp0->fd_lastfile = -1;
1694 	spin_init(&fdp0->fd_spin);
1695 }
1696 
1697 /*
1698  * Build a new filedesc structure.
1699  *
1700  * NOT MPSAFE (vref)
1701  */
1702 struct filedesc *
1703 fdinit(struct proc *p)
1704 {
1705 	struct filedesc *newfdp;
1706 	struct filedesc *fdp = p->p_fd;
1707 
1708 	newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1709 	spin_lock(&fdp->fd_spin);
1710 	if (fdp->fd_cdir) {
1711 		newfdp->fd_cdir = fdp->fd_cdir;
1712 		vref(newfdp->fd_cdir);
1713 		cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1714 	}
1715 
1716 	/*
1717 	 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1718 	 * proc0, but should unconditionally exist in other processes.
1719 	 */
1720 	if (fdp->fd_rdir) {
1721 		newfdp->fd_rdir = fdp->fd_rdir;
1722 		vref(newfdp->fd_rdir);
1723 		cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1724 	}
1725 	if (fdp->fd_jdir) {
1726 		newfdp->fd_jdir = fdp->fd_jdir;
1727 		vref(newfdp->fd_jdir);
1728 		cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1729 	}
1730 	spin_unlock(&fdp->fd_spin);
1731 
1732 	/* Create the file descriptor table. */
1733 	newfdp->fd_refcnt = 1;
1734 	newfdp->fd_cmask = cmask;
1735 	newfdp->fd_files = newfdp->fd_builtin_files;
1736 	newfdp->fd_nfiles = NDFILE;
1737 	newfdp->fd_lastfile = -1;
1738 	spin_init(&newfdp->fd_spin);
1739 
1740 	return (newfdp);
1741 }
1742 
1743 /*
1744  * Share a filedesc structure.
1745  *
1746  * MPSAFE
1747  */
1748 struct filedesc *
1749 fdshare(struct proc *p)
1750 {
1751 	struct filedesc *fdp;
1752 
1753 	fdp = p->p_fd;
1754 	spin_lock(&fdp->fd_spin);
1755 	fdp->fd_refcnt++;
1756 	spin_unlock(&fdp->fd_spin);
1757 	return (fdp);
1758 }
1759 
1760 /*
1761  * Copy a filedesc structure.
1762  *
1763  * MPSAFE
1764  */
1765 struct filedesc *
1766 fdcopy(struct proc *p)
1767 {
1768 	struct filedesc *fdp = p->p_fd;
1769 	struct filedesc *newfdp;
1770 	struct fdnode *fdnode;
1771 	int i;
1772 	int ni;
1773 
1774 	/*
1775 	 * Certain daemons might not have file descriptors.
1776 	 */
1777 	if (fdp == NULL)
1778 		return (NULL);
1779 
1780 	/*
1781 	 * Allocate the new filedesc and fd_files[] array.  This can race
1782 	 * with operations by other threads on the fdp so we have to be
1783 	 * careful.
1784 	 */
1785 	newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
1786 again:
1787 	spin_lock(&fdp->fd_spin);
1788 	if (fdp->fd_lastfile < NDFILE) {
1789 		newfdp->fd_files = newfdp->fd_builtin_files;
1790 		i = NDFILE;
1791 	} else {
1792 		/*
1793 		 * We have to allocate (N^2-1) entries for our in-place
1794 		 * binary tree.  Allow the table to shrink.
1795 		 */
1796 		i = fdp->fd_nfiles;
1797 		ni = (i - 1) / 2;
1798 		while (ni > fdp->fd_lastfile && ni > NDFILE) {
1799 			i = ni;
1800 			ni = (i - 1) / 2;
1801 		}
1802 		spin_unlock(&fdp->fd_spin);
1803 		newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1804 					  M_FILEDESC, M_WAITOK | M_ZERO);
1805 
1806 		/*
1807 		 * Check for race, retry
1808 		 */
1809 		spin_lock(&fdp->fd_spin);
1810 		if (i <= fdp->fd_lastfile) {
1811 			spin_unlock(&fdp->fd_spin);
1812 			kfree(newfdp->fd_files, M_FILEDESC);
1813 			goto again;
1814 		}
1815 	}
1816 
1817 	/*
1818 	 * Dup the remaining fields. vref() and cache_hold() can be
1819 	 * safely called while holding the read spinlock on fdp.
1820 	 *
1821 	 * The read spinlock on fdp is still being held.
1822 	 *
1823 	 * NOTE: vref and cache_hold calls for the case where the vnode
1824 	 * or cache entry already has at least one ref may be called
1825 	 * while holding spin locks.
1826 	 */
1827 	if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1828 		vref(newfdp->fd_cdir);
1829 		cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1830 	}
1831 	/*
1832 	 * We must check for fd_rdir here, at least for now because
1833 	 * the init process is created before we have access to the
1834 	 * rootvode to take a reference to it.
1835 	 */
1836 	if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1837 		vref(newfdp->fd_rdir);
1838 		cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1839 	}
1840 	if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1841 		vref(newfdp->fd_jdir);
1842 		cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1843 	}
1844 	newfdp->fd_refcnt = 1;
1845 	newfdp->fd_nfiles = i;
1846 	newfdp->fd_lastfile = fdp->fd_lastfile;
1847 	newfdp->fd_freefile = fdp->fd_freefile;
1848 	newfdp->fd_cmask = fdp->fd_cmask;
1849 	spin_init(&newfdp->fd_spin);
1850 
1851 	/*
1852 	 * Copy the descriptor table through (i).  This also copies the
1853 	 * allocation state.   Then go through and ref the file pointers
1854 	 * and clean up any KQ descriptors.
1855 	 *
1856 	 * kq descriptors cannot be copied.  Since we haven't ref'd the
1857 	 * copied files yet we can ignore the return value from funsetfd().
1858 	 *
1859 	 * The read spinlock on fdp is still being held.
1860 	 */
1861 	bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1862 	for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1863 		fdnode = &newfdp->fd_files[i];
1864 		if (fdnode->reserved) {
1865 			fdreserve_locked(newfdp, i, -1);
1866 			fdnode->reserved = 0;
1867 			fdfixup_locked(newfdp, i);
1868 		} else if (fdnode->fp) {
1869 			if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1870 				(void)funsetfd_locked(newfdp, i);
1871 			} else {
1872 				fhold(fdnode->fp);
1873 			}
1874 		}
1875 	}
1876 	spin_unlock(&fdp->fd_spin);
1877 	return (newfdp);
1878 }
1879 
1880 /*
1881  * Release a filedesc structure.
1882  *
1883  * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1884  */
1885 void
1886 fdfree(struct proc *p, struct filedesc *repl)
1887 {
1888 	struct filedesc *fdp;
1889 	struct fdnode *fdnode;
1890 	int i;
1891 	struct filedesc_to_leader *fdtol;
1892 	struct file *fp;
1893 	struct vnode *vp;
1894 	struct flock lf;
1895 
1896 	/*
1897 	 * Certain daemons might not have file descriptors.
1898 	 */
1899 	fdp = p->p_fd;
1900 	if (fdp == NULL) {
1901 		p->p_fd = repl;
1902 		return;
1903 	}
1904 
1905 	/*
1906 	 * Severe messing around to follow.
1907 	 */
1908 	spin_lock(&fdp->fd_spin);
1909 
1910 	/* Check for special need to clear POSIX style locks */
1911 	fdtol = p->p_fdtol;
1912 	if (fdtol != NULL) {
1913 		KASSERT(fdtol->fdl_refcount > 0,
1914 			("filedesc_to_refcount botch: fdl_refcount=%d",
1915 			 fdtol->fdl_refcount));
1916 		if (fdtol->fdl_refcount == 1 &&
1917 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1918 			for (i = 0; i <= fdp->fd_lastfile; ++i) {
1919 				fdnode = &fdp->fd_files[i];
1920 				if (fdnode->fp == NULL ||
1921 				    fdnode->fp->f_type != DTYPE_VNODE) {
1922 					continue;
1923 				}
1924 				fp = fdnode->fp;
1925 				fhold(fp);
1926 				spin_unlock(&fdp->fd_spin);
1927 
1928 				lf.l_whence = SEEK_SET;
1929 				lf.l_start = 0;
1930 				lf.l_len = 0;
1931 				lf.l_type = F_UNLCK;
1932 				vp = (struct vnode *)fp->f_data;
1933 				(void) VOP_ADVLOCK(vp,
1934 						   (caddr_t)p->p_leader,
1935 						   F_UNLCK,
1936 						   &lf,
1937 						   F_POSIX);
1938 				fdrop(fp);
1939 				spin_lock(&fdp->fd_spin);
1940 			}
1941 		}
1942 	retry:
1943 		if (fdtol->fdl_refcount == 1) {
1944 			if (fdp->fd_holdleaderscount > 0 &&
1945 			    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1946 				/*
1947 				 * close() or do_dup() has cleared a reference
1948 				 * in a shared file descriptor table.
1949 				 */
1950 				fdp->fd_holdleaderswakeup = 1;
1951 				ssleep(&fdp->fd_holdleaderscount,
1952 				       &fdp->fd_spin, 0, "fdlhold", 0);
1953 				goto retry;
1954 			}
1955 			if (fdtol->fdl_holdcount > 0) {
1956 				/*
1957 				 * Ensure that fdtol->fdl_leader
1958 				 * remains valid in closef().
1959 				 */
1960 				fdtol->fdl_wakeup = 1;
1961 				ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
1962 				goto retry;
1963 			}
1964 		}
1965 		fdtol->fdl_refcount--;
1966 		if (fdtol->fdl_refcount == 0 &&
1967 		    fdtol->fdl_holdcount == 0) {
1968 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1969 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1970 		} else {
1971 			fdtol = NULL;
1972 		}
1973 		p->p_fdtol = NULL;
1974 		if (fdtol != NULL) {
1975 			spin_unlock(&fdp->fd_spin);
1976 			kfree(fdtol, M_FILEDESC_TO_LEADER);
1977 			spin_lock(&fdp->fd_spin);
1978 		}
1979 	}
1980 	if (--fdp->fd_refcnt > 0) {
1981 		spin_unlock(&fdp->fd_spin);
1982 		spin_lock(&p->p_spin);
1983 		p->p_fd = repl;
1984 		spin_unlock(&p->p_spin);
1985 		return;
1986 	}
1987 
1988 	/*
1989 	 * Even though we are the last reference to the structure allproc
1990 	 * scans may still reference the structure.  Maintain proper
1991 	 * locks until we can replace p->p_fd.
1992 	 *
1993 	 * Also note that kqueue's closef still needs to reference the
1994 	 * fdp via p->p_fd, so we have to close the descriptors before
1995 	 * we replace p->p_fd.
1996 	 */
1997 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1998 		if (fdp->fd_files[i].fp) {
1999 			fp = funsetfd_locked(fdp, i);
2000 			if (fp) {
2001 				spin_unlock(&fdp->fd_spin);
2002 				if (SLIST_FIRST(&fp->f_klist))
2003 					knote_fdclose(fp, fdp, i);
2004 				closef(fp, p);
2005 				spin_lock(&fdp->fd_spin);
2006 			}
2007 		}
2008 	}
2009 	spin_unlock(&fdp->fd_spin);
2010 
2011 	/*
2012 	 * Interlock against an allproc scan operations (typically frevoke).
2013 	 */
2014 	spin_lock(&p->p_spin);
2015 	p->p_fd = repl;
2016 	spin_unlock(&p->p_spin);
2017 
2018 	/*
2019 	 * Wait for any softrefs to go away.  This race rarely occurs so
2020 	 * we can use a non-critical-path style poll/sleep loop.  The
2021 	 * race only occurs against allproc scans.
2022 	 *
2023 	 * No new softrefs can occur with the fdp disconnected from the
2024 	 * process.
2025 	 */
2026 	if (fdp->fd_softrefs) {
2027 		kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2028 		while (fdp->fd_softrefs)
2029 			tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2030 	}
2031 
2032 	if (fdp->fd_files != fdp->fd_builtin_files)
2033 		kfree(fdp->fd_files, M_FILEDESC);
2034 	if (fdp->fd_cdir) {
2035 		cache_drop(&fdp->fd_ncdir);
2036 		vrele(fdp->fd_cdir);
2037 	}
2038 	if (fdp->fd_rdir) {
2039 		cache_drop(&fdp->fd_nrdir);
2040 		vrele(fdp->fd_rdir);
2041 	}
2042 	if (fdp->fd_jdir) {
2043 		cache_drop(&fdp->fd_njdir);
2044 		vrele(fdp->fd_jdir);
2045 	}
2046 	kfree(fdp, M_FILEDESC);
2047 }
2048 
2049 /*
2050  * Retrieve and reference the file pointer associated with a descriptor.
2051  *
2052  * MPSAFE
2053  */
2054 struct file *
2055 holdfp(struct filedesc *fdp, int fd, int flag)
2056 {
2057 	struct file* fp;
2058 
2059 	spin_lock(&fdp->fd_spin);
2060 	if (((u_int)fd) >= fdp->fd_nfiles) {
2061 		fp = NULL;
2062 		goto done;
2063 	}
2064 	if ((fp = fdp->fd_files[fd].fp) == NULL)
2065 		goto done;
2066 	if ((fp->f_flag & flag) == 0 && flag != -1) {
2067 		fp = NULL;
2068 		goto done;
2069 	}
2070 	fhold(fp);
2071 done:
2072 	spin_unlock(&fdp->fd_spin);
2073 	return (fp);
2074 }
2075 
2076 /*
2077  * holdsock() - load the struct file pointer associated
2078  * with a socket into *fpp.  If an error occurs, non-zero
2079  * will be returned and *fpp will be set to NULL.
2080  *
2081  * MPSAFE
2082  */
2083 int
2084 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2085 {
2086 	struct file *fp;
2087 	int error;
2088 
2089 	spin_lock(&fdp->fd_spin);
2090 	if ((unsigned)fd >= fdp->fd_nfiles) {
2091 		error = EBADF;
2092 		fp = NULL;
2093 		goto done;
2094 	}
2095 	if ((fp = fdp->fd_files[fd].fp) == NULL) {
2096 		error = EBADF;
2097 		goto done;
2098 	}
2099 	if (fp->f_type != DTYPE_SOCKET) {
2100 		error = ENOTSOCK;
2101 		goto done;
2102 	}
2103 	fhold(fp);
2104 	error = 0;
2105 done:
2106 	spin_unlock(&fdp->fd_spin);
2107 	*fpp = fp;
2108 	return (error);
2109 }
2110 
2111 /*
2112  * Convert a user file descriptor to a held file pointer.
2113  *
2114  * MPSAFE
2115  */
2116 int
2117 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2118 {
2119 	struct file *fp;
2120 	int error;
2121 
2122 	spin_lock(&fdp->fd_spin);
2123 	if ((unsigned)fd >= fdp->fd_nfiles) {
2124 		error = EBADF;
2125 		fp = NULL;
2126 		goto done;
2127 	}
2128 	if ((fp = fdp->fd_files[fd].fp) == NULL) {
2129 		error = EBADF;
2130 		goto done;
2131 	}
2132 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2133 		fp = NULL;
2134 		error = EINVAL;
2135 		goto done;
2136 	}
2137 	fhold(fp);
2138 	error = 0;
2139 done:
2140 	spin_unlock(&fdp->fd_spin);
2141 	*fpp = fp;
2142 	return (error);
2143 }
2144 
2145 /*
2146  * For setugid programs, we don't want to people to use that setugidness
2147  * to generate error messages which write to a file which otherwise would
2148  * otherwise be off-limits to the process.
2149  *
2150  * This is a gross hack to plug the hole.  A better solution would involve
2151  * a special vop or other form of generalized access control mechanism.  We
2152  * go ahead and just reject all procfs file systems accesses as dangerous.
2153  *
2154  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2155  * sufficient.  We also don't for check setugidness since we know we are.
2156  */
2157 static int
2158 is_unsafe(struct file *fp)
2159 {
2160 	if (fp->f_type == DTYPE_VNODE &&
2161 	    ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2162 		return (1);
2163 	return (0);
2164 }
2165 
2166 /*
2167  * Make this setguid thing safe, if at all possible.
2168  *
2169  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2170  */
2171 void
2172 setugidsafety(struct proc *p)
2173 {
2174 	struct filedesc *fdp = p->p_fd;
2175 	int i;
2176 
2177 	/* Certain daemons might not have file descriptors. */
2178 	if (fdp == NULL)
2179 		return;
2180 
2181 	/*
2182 	 * note: fdp->fd_files may be reallocated out from under us while
2183 	 * we are blocked in a close.  Be careful!
2184 	 */
2185 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2186 		if (i > 2)
2187 			break;
2188 		if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2189 			struct file *fp;
2190 
2191 			/*
2192 			 * NULL-out descriptor prior to close to avoid
2193 			 * a race while close blocks.
2194 			 */
2195 			if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2196 				knote_fdclose(fp, fdp, i);
2197 				closef(fp, p);
2198 			}
2199 		}
2200 	}
2201 }
2202 
2203 /*
2204  * Close any files on exec?
2205  *
2206  * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2207  */
2208 void
2209 fdcloseexec(struct proc *p)
2210 {
2211 	struct filedesc *fdp = p->p_fd;
2212 	int i;
2213 
2214 	/* Certain daemons might not have file descriptors. */
2215 	if (fdp == NULL)
2216 		return;
2217 
2218 	/*
2219 	 * We cannot cache fd_files since operations may block and rip
2220 	 * them out from under us.
2221 	 */
2222 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2223 		if (fdp->fd_files[i].fp != NULL &&
2224 		    (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2225 			struct file *fp;
2226 
2227 			/*
2228 			 * NULL-out descriptor prior to close to avoid
2229 			 * a race while close blocks.
2230 			 */
2231 			if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2232 				knote_fdclose(fp, fdp, i);
2233 				closef(fp, p);
2234 			}
2235 		}
2236 	}
2237 }
2238 
2239 /*
2240  * It is unsafe for set[ug]id processes to be started with file
2241  * descriptors 0..2 closed, as these descriptors are given implicit
2242  * significance in the Standard C library.  fdcheckstd() will create a
2243  * descriptor referencing /dev/null for each of stdin, stdout, and
2244  * stderr that is not already open.
2245  *
2246  * NOT MPSAFE - calls falloc, vn_open, etc
2247  */
2248 int
2249 fdcheckstd(struct lwp *lp)
2250 {
2251 	struct nlookupdata nd;
2252 	struct filedesc *fdp;
2253 	struct file *fp;
2254 	int retval;
2255 	int i, error, flags, devnull;
2256 
2257 	fdp = lp->lwp_proc->p_fd;
2258 	if (fdp == NULL)
2259 		return (0);
2260 	devnull = -1;
2261 	error = 0;
2262 	for (i = 0; i < 3; i++) {
2263 		if (fdp->fd_files[i].fp != NULL)
2264 			continue;
2265 		if (devnull < 0) {
2266 			if ((error = falloc(lp, &fp, &devnull)) != 0)
2267 				break;
2268 
2269 			error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2270 						NLC_FOLLOW|NLC_LOCKVP);
2271 			flags = FREAD | FWRITE;
2272 			if (error == 0)
2273 				error = vn_open(&nd, fp, flags, 0);
2274 			if (error == 0)
2275 				fsetfd(fdp, fp, devnull);
2276 			else
2277 				fsetfd(fdp, NULL, devnull);
2278 			fdrop(fp);
2279 			nlookup_done(&nd);
2280 			if (error)
2281 				break;
2282 			KKASSERT(i == devnull);
2283 		} else {
2284 			error = kern_dup(DUP_FIXED, devnull, i, &retval);
2285 			if (error != 0)
2286 				break;
2287 		}
2288 	}
2289 	return (error);
2290 }
2291 
2292 /*
2293  * Internal form of close.
2294  * Decrement reference count on file structure.
2295  * Note: td and/or p may be NULL when closing a file
2296  * that was being passed in a message.
2297  *
2298  * MPALMOSTSAFE - acquires mplock for VOP operations
2299  */
2300 int
2301 closef(struct file *fp, struct proc *p)
2302 {
2303 	struct vnode *vp;
2304 	struct flock lf;
2305 	struct filedesc_to_leader *fdtol;
2306 
2307 	if (fp == NULL)
2308 		return (0);
2309 
2310 	/*
2311 	 * POSIX record locking dictates that any close releases ALL
2312 	 * locks owned by this process.  This is handled by setting
2313 	 * a flag in the unlock to free ONLY locks obeying POSIX
2314 	 * semantics, and not to free BSD-style file locks.
2315 	 * If the descriptor was in a message, POSIX-style locks
2316 	 * aren't passed with the descriptor.
2317 	 */
2318 	if (p != NULL && fp->f_type == DTYPE_VNODE &&
2319 	    (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2320 	) {
2321 		get_mplock();
2322 		if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2323 			lf.l_whence = SEEK_SET;
2324 			lf.l_start = 0;
2325 			lf.l_len = 0;
2326 			lf.l_type = F_UNLCK;
2327 			vp = (struct vnode *)fp->f_data;
2328 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2329 					   &lf, F_POSIX);
2330 		}
2331 		fdtol = p->p_fdtol;
2332 		if (fdtol != NULL) {
2333 			/*
2334 			 * Handle special case where file descriptor table
2335 			 * is shared between multiple process leaders.
2336 			 */
2337 			for (fdtol = fdtol->fdl_next;
2338 			     fdtol != p->p_fdtol;
2339 			     fdtol = fdtol->fdl_next) {
2340 				if ((fdtol->fdl_leader->p_flag &
2341 				     P_ADVLOCK) == 0)
2342 					continue;
2343 				fdtol->fdl_holdcount++;
2344 				lf.l_whence = SEEK_SET;
2345 				lf.l_start = 0;
2346 				lf.l_len = 0;
2347 				lf.l_type = F_UNLCK;
2348 				vp = (struct vnode *)fp->f_data;
2349 				(void) VOP_ADVLOCK(vp,
2350 						   (caddr_t)fdtol->fdl_leader,
2351 						   F_UNLCK, &lf, F_POSIX);
2352 				fdtol->fdl_holdcount--;
2353 				if (fdtol->fdl_holdcount == 0 &&
2354 				    fdtol->fdl_wakeup != 0) {
2355 					fdtol->fdl_wakeup = 0;
2356 					wakeup(fdtol);
2357 				}
2358 			}
2359 		}
2360 		rel_mplock();
2361 	}
2362 	return (fdrop(fp));
2363 }
2364 
2365 /*
2366  * MPSAFE
2367  *
2368  * fhold() can only be called if f_count is already at least 1 (i.e. the
2369  * caller of fhold() already has a reference to the file pointer in some
2370  * manner or other).
2371  *
2372  * f_count is not spin-locked.  Instead, atomic ops are used for
2373  * incrementing, decrementing, and handling the 1->0 transition.
2374  */
2375 void
2376 fhold(struct file *fp)
2377 {
2378 	atomic_add_int(&fp->f_count, 1);
2379 }
2380 
2381 /*
2382  * fdrop() - drop a reference to a descriptor
2383  *
2384  * MPALMOSTSAFE - acquires mplock for final close sequence
2385  */
2386 int
2387 fdrop(struct file *fp)
2388 {
2389 	struct flock lf;
2390 	struct vnode *vp;
2391 	int error;
2392 
2393 	/*
2394 	 * A combined fetch and subtract is needed to properly detect
2395 	 * 1->0 transitions, otherwise two cpus dropping from a ref
2396 	 * count of 2 might both try to run the 1->0 code.
2397 	 */
2398 	if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2399 		return (0);
2400 
2401 	KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
2402 	get_mplock();
2403 
2404 	/*
2405 	 * The last reference has gone away, we own the fp structure free
2406 	 * and clear.
2407 	 */
2408 	if (fp->f_count < 0)
2409 		panic("fdrop: count < 0");
2410 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2411 	    (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2412 	) {
2413 		lf.l_whence = SEEK_SET;
2414 		lf.l_start = 0;
2415 		lf.l_len = 0;
2416 		lf.l_type = F_UNLCK;
2417 		vp = (struct vnode *)fp->f_data;
2418 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2419 	}
2420 	if (fp->f_ops != &badfileops)
2421 		error = fo_close(fp);
2422 	else
2423 		error = 0;
2424 	ffree(fp);
2425 	rel_mplock();
2426 	return (error);
2427 }
2428 
2429 /*
2430  * Apply an advisory lock on a file descriptor.
2431  *
2432  * Just attempt to get a record lock of the requested type on
2433  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2434  *
2435  * MPALMOSTSAFE
2436  */
2437 int
2438 sys_flock(struct flock_args *uap)
2439 {
2440 	struct proc *p = curproc;
2441 	struct file *fp;
2442 	struct vnode *vp;
2443 	struct flock lf;
2444 	int error;
2445 
2446 	if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2447 		return (EBADF);
2448 	get_mplock();
2449 	if (fp->f_type != DTYPE_VNODE) {
2450 		error = EOPNOTSUPP;
2451 		goto done;
2452 	}
2453 	vp = (struct vnode *)fp->f_data;
2454 	lf.l_whence = SEEK_SET;
2455 	lf.l_start = 0;
2456 	lf.l_len = 0;
2457 	if (uap->how & LOCK_UN) {
2458 		lf.l_type = F_UNLCK;
2459 		fp->f_flag &= ~FHASLOCK;
2460 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2461 		goto done;
2462 	}
2463 	if (uap->how & LOCK_EX)
2464 		lf.l_type = F_WRLCK;
2465 	else if (uap->how & LOCK_SH)
2466 		lf.l_type = F_RDLCK;
2467 	else {
2468 		error = EBADF;
2469 		goto done;
2470 	}
2471 	fp->f_flag |= FHASLOCK;
2472 	if (uap->how & LOCK_NB)
2473 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2474 	else
2475 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2476 done:
2477 	rel_mplock();
2478 	fdrop(fp);
2479 	return (error);
2480 }
2481 
2482 /*
2483  * File Descriptor pseudo-device driver (/dev/fd/).
2484  *
2485  * Opening minor device N dup()s the file (if any) connected to file
2486  * descriptor N belonging to the calling process.  Note that this driver
2487  * consists of only the ``open()'' routine, because all subsequent
2488  * references to this file will be direct to the other driver.
2489  */
2490 static int
2491 fdopen(struct dev_open_args *ap)
2492 {
2493 	thread_t td = curthread;
2494 
2495 	KKASSERT(td->td_lwp != NULL);
2496 
2497 	/*
2498 	 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2499 	 * the file descriptor being sought for duplication. The error
2500 	 * return ensures that the vnode for this device will be released
2501 	 * by vn_open. Open will detect this special error and take the
2502 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2503 	 * will simply report the error.
2504 	 */
2505 	td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2506 	return (ENODEV);
2507 }
2508 
2509 /*
2510  * The caller has reserved the file descriptor dfd for us.  On success we
2511  * must fsetfd() it.  On failure the caller will clean it up.
2512  *
2513  * MPSAFE
2514  */
2515 int
2516 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2517 {
2518 	struct file *wfp;
2519 	struct file *xfp;
2520 	int werror;
2521 
2522 	if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2523 		return (EBADF);
2524 
2525 	/*
2526 	 * Close a revoke/dup race.  Duping a descriptor marked as revoked
2527 	 * will dup a dummy descriptor instead of the real one.
2528 	 */
2529 	if (wfp->f_flag & FREVOKED) {
2530 		kprintf("Warning: attempt to dup() a revoked descriptor\n");
2531 		fdrop(wfp);
2532 		wfp = NULL;
2533 		werror = falloc(NULL, &wfp, NULL);
2534 		if (werror)
2535 			return (werror);
2536 	}
2537 
2538 	/*
2539 	 * There are two cases of interest here.
2540 	 *
2541 	 * For ENODEV simply dup sfd to file descriptor dfd and return.
2542 	 *
2543 	 * For ENXIO steal away the file structure from sfd and store it
2544 	 * dfd.  sfd is effectively closed by this operation.
2545 	 *
2546 	 * Any other error code is just returned.
2547 	 */
2548 	switch (error) {
2549 	case ENODEV:
2550 		/*
2551 		 * Check that the mode the file is being opened for is a
2552 		 * subset of the mode of the existing descriptor.
2553 		 */
2554 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2555 			error = EACCES;
2556 			break;
2557 		}
2558 		spin_lock(&fdp->fd_spin);
2559 		fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2560 		fsetfd_locked(fdp, wfp, dfd);
2561 		spin_unlock(&fdp->fd_spin);
2562 		error = 0;
2563 		break;
2564 	case ENXIO:
2565 		/*
2566 		 * Steal away the file pointer from dfd, and stuff it into indx.
2567 		 */
2568 		spin_lock(&fdp->fd_spin);
2569 		fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2570 		fsetfd(fdp, wfp, dfd);
2571 		if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
2572 			spin_unlock(&fdp->fd_spin);
2573 			fdrop(xfp);
2574 		} else {
2575 			spin_unlock(&fdp->fd_spin);
2576 		}
2577 		error = 0;
2578 		break;
2579 	default:
2580 		break;
2581 	}
2582 	fdrop(wfp);
2583 	return (error);
2584 }
2585 
2586 /*
2587  * NOT MPSAFE - I think these refer to a common file descriptor table
2588  * and we need to spinlock that to link fdtol in.
2589  */
2590 struct filedesc_to_leader *
2591 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2592 			 struct proc *leader)
2593 {
2594 	struct filedesc_to_leader *fdtol;
2595 
2596 	fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2597 			M_FILEDESC_TO_LEADER, M_WAITOK);
2598 	fdtol->fdl_refcount = 1;
2599 	fdtol->fdl_holdcount = 0;
2600 	fdtol->fdl_wakeup = 0;
2601 	fdtol->fdl_leader = leader;
2602 	if (old != NULL) {
2603 		fdtol->fdl_next = old->fdl_next;
2604 		fdtol->fdl_prev = old;
2605 		old->fdl_next = fdtol;
2606 		fdtol->fdl_next->fdl_prev = fdtol;
2607 	} else {
2608 		fdtol->fdl_next = fdtol;
2609 		fdtol->fdl_prev = fdtol;
2610 	}
2611 	return fdtol;
2612 }
2613 
2614 /*
2615  * Scan all file pointers in the system.  The callback is made with
2616  * the master list spinlock held exclusively.
2617  *
2618  * MPSAFE
2619  */
2620 void
2621 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2622 {
2623 	struct file *fp;
2624 	int res;
2625 
2626 	spin_lock(&filehead_spin);
2627 	LIST_FOREACH(fp, &filehead, f_list) {
2628 		res = callback(fp, data);
2629 		if (res < 0)
2630 			break;
2631 	}
2632 	spin_unlock(&filehead_spin);
2633 }
2634 
2635 /*
2636  * Get file structures.
2637  *
2638  * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2639  */
2640 
2641 struct sysctl_kern_file_info {
2642 	int count;
2643 	int error;
2644 	struct sysctl_req *req;
2645 };
2646 
2647 static int sysctl_kern_file_callback(struct proc *p, void *data);
2648 
2649 static int
2650 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2651 {
2652 	struct sysctl_kern_file_info info;
2653 
2654 	/*
2655 	 * Note: because the number of file descriptors is calculated
2656 	 * in different ways for sizing vs returning the data,
2657 	 * there is information leakage from the first loop.  However,
2658 	 * it is of a similar order of magnitude to the leakage from
2659 	 * global system statistics such as kern.openfiles.
2660 	 *
2661 	 * When just doing a count, note that we cannot just count
2662 	 * the elements and add f_count via the filehead list because
2663 	 * threaded processes share their descriptor table and f_count might
2664 	 * still be '1' in that case.
2665 	 *
2666 	 * Since the SYSCTL op can block, we must hold the process to
2667 	 * prevent it being ripped out from under us either in the
2668 	 * file descriptor loop or in the greater LIST_FOREACH.  The
2669 	 * process may be in varying states of disrepair.  If the process
2670 	 * is in SZOMB we may have caught it just as it is being removed
2671 	 * from the allproc list, we must skip it in that case to maintain
2672 	 * an unbroken chain through the allproc list.
2673 	 */
2674 	info.count = 0;
2675 	info.error = 0;
2676 	info.req = req;
2677 	allproc_scan(sysctl_kern_file_callback, &info);
2678 
2679 	/*
2680 	 * When just calculating the size, overestimate a bit to try to
2681 	 * prevent system activity from causing the buffer-fill call
2682 	 * to fail later on.
2683 	 */
2684 	if (req->oldptr == NULL) {
2685 		info.count = (info.count + 16) + (info.count / 10);
2686 		info.error = SYSCTL_OUT(req, NULL,
2687 					info.count * sizeof(struct kinfo_file));
2688 	}
2689 	return (info.error);
2690 }
2691 
2692 static int
2693 sysctl_kern_file_callback(struct proc *p, void *data)
2694 {
2695 	struct sysctl_kern_file_info *info = data;
2696 	struct kinfo_file kf;
2697 	struct filedesc *fdp;
2698 	struct file *fp;
2699 	uid_t uid;
2700 	int n;
2701 
2702 	if (p->p_stat == SIDL || p->p_stat == SZOMB)
2703 		return(0);
2704 	if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2705 		return(0);
2706 
2707 	/*
2708 	 * Softref the fdp to prevent it from being destroyed
2709 	 */
2710 	spin_lock(&p->p_spin);
2711 	if ((fdp = p->p_fd) == NULL) {
2712 		spin_unlock(&p->p_spin);
2713 		return(0);
2714 	}
2715 	atomic_add_int(&fdp->fd_softrefs, 1);
2716 	spin_unlock(&p->p_spin);
2717 
2718 	/*
2719 	 * The fdp's own spinlock prevents the contents from being
2720 	 * modified.
2721 	 */
2722 	spin_lock(&fdp->fd_spin);
2723 	for (n = 0; n < fdp->fd_nfiles; ++n) {
2724 		if ((fp = fdp->fd_files[n].fp) == NULL)
2725 			continue;
2726 		if (info->req->oldptr == NULL) {
2727 			++info->count;
2728 		} else {
2729 			uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2730 			kcore_make_file(&kf, fp, p->p_pid, uid, n);
2731 			spin_unlock(&fdp->fd_spin);
2732 			info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2733 			spin_lock(&fdp->fd_spin);
2734 			if (info->error)
2735 				break;
2736 		}
2737 	}
2738 	spin_unlock(&fdp->fd_spin);
2739 	atomic_subtract_int(&fdp->fd_softrefs, 1);
2740 	if (info->error)
2741 		return(-1);
2742 	return(0);
2743 }
2744 
2745 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2746     0, 0, sysctl_kern_file, "S,file", "Entire file table");
2747 
2748 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2749     &minfilesperproc, 0, "Minimum files allowed open per process");
2750 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2751     &maxfilesperproc, 0, "Maximum files allowed open per process");
2752 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2753     &maxfilesperuser, 0, "Maximum files allowed open per user");
2754 
2755 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2756     &maxfiles, 0, "Maximum number of files");
2757 
2758 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2759     &maxfilesrootres, 0, "Descriptors reserved for root use");
2760 
2761 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2762 	&nfiles, 0, "System-wide number of open files");
2763 
2764 static void
2765 fildesc_drvinit(void *unused)
2766 {
2767 	int fd;
2768 
2769 	for (fd = 0; fd < NUMFDESC; fd++) {
2770 		make_dev(&fildesc_ops, fd,
2771 			 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2772 	}
2773 
2774 	make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2775 	make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2776 	make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2777 }
2778 
2779 /*
2780  * MPSAFE
2781  */
2782 struct fileops badfileops = {
2783 	.fo_read = badfo_readwrite,
2784 	.fo_write = badfo_readwrite,
2785 	.fo_ioctl = badfo_ioctl,
2786 	.fo_poll = badfo_poll,
2787 	.fo_kqfilter = badfo_kqfilter,
2788 	.fo_stat = badfo_stat,
2789 	.fo_close = badfo_close,
2790 	.fo_shutdown = badfo_shutdown
2791 };
2792 
2793 int
2794 badfo_readwrite(
2795 	struct file *fp,
2796 	struct uio *uio,
2797 	struct ucred *cred,
2798 	int flags
2799 ) {
2800 	return (EBADF);
2801 }
2802 
2803 int
2804 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2805 	    struct ucred *cred, struct sysmsg *msgv)
2806 {
2807 	return (EBADF);
2808 }
2809 
2810 int
2811 badfo_poll(struct file *fp, int events, struct ucred *cred)
2812 {
2813 	return (0);
2814 }
2815 
2816 /*
2817  * Must return an error to prevent registration, typically
2818  * due to a revoked descriptor (file_filtops assigned).
2819  */
2820 int
2821 badfo_kqfilter(struct file *fp, struct knote *kn)
2822 {
2823 	return (EOPNOTSUPP);
2824 }
2825 
2826 /*
2827  * MPSAFE
2828  */
2829 int
2830 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2831 {
2832 	return (EBADF);
2833 }
2834 
2835 /*
2836  * MPSAFE
2837  */
2838 int
2839 badfo_close(struct file *fp)
2840 {
2841 	return (EBADF);
2842 }
2843 
2844 /*
2845  * MPSAFE
2846  */
2847 int
2848 badfo_shutdown(struct file *fp, int how)
2849 {
2850 	return (EBADF);
2851 }
2852 
2853 /*
2854  * MPSAFE
2855  */
2856 int
2857 nofo_shutdown(struct file *fp, int how)
2858 {
2859 	return (EOPNOTSUPP);
2860 }
2861 
2862 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2863 					fildesc_drvinit,NULL)
2864