xref: /freebsd/sys/kern/kern_descrip.c (revision 7660b554bc59a07be0431c17e0e33815818baa69)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_compat.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysproto.h>
50 #include <sys/conf.h>
51 #include <sys/filedesc.h>
52 #include <sys/lock.h>
53 #include <sys/kernel.h>
54 #include <sys/limits.h>
55 #include <sys/malloc.h>
56 #include <sys/mutex.h>
57 #include <sys/sysctl.h>
58 #include <sys/vnode.h>
59 #include <sys/mount.h>
60 #include <sys/proc.h>
61 #include <sys/namei.h>
62 #include <sys/file.h>
63 #include <sys/stat.h>
64 #include <sys/filio.h>
65 #include <sys/fcntl.h>
66 #include <sys/unistd.h>
67 #include <sys/resourcevar.h>
68 #include <sys/event.h>
69 #include <sys/sx.h>
70 #include <sys/socketvar.h>
71 #include <sys/signalvar.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_extern.h>
75 #include <vm/uma.h>
76 
77 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
78 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
79 		     "file desc to leader structures");
80 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
81 
82 static uma_zone_t file_zone;
83 
84 static	 d_open_t  fdopen;
85 #define	NUMFDESC 64
86 
87 #define	CDEV_MAJOR 22
88 static struct cdevsw fildesc_cdevsw = {
89 	.d_open =	fdopen,
90 	.d_name =	"FD",
91 	.d_maj =	CDEV_MAJOR,
92 };
93 
94 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
95 enum dup_type { DUP_VARIABLE, DUP_FIXED };
96 
97 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
98     register_t *retval);
99 
100 /*
101  * Descriptor management.
102  */
103 struct filelist filehead;	/* head of list of open files */
104 int nfiles;			/* actual number of open files */
105 extern int cmask;
106 struct sx filelist_lock;	/* sx to protect filelist */
107 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
108 
109 /*
110  * System calls on descriptors.
111  */
112 #ifndef _SYS_SYSPROTO_H_
113 struct getdtablesize_args {
114 	int	dummy;
115 };
116 #endif
117 /*
118  * MPSAFE
119  */
120 /* ARGSUSED */
121 int
122 getdtablesize(td, uap)
123 	struct thread *td;
124 	struct getdtablesize_args *uap;
125 {
126 	struct proc *p = td->td_proc;
127 
128 	mtx_lock(&Giant);
129 	td->td_retval[0] =
130 	    min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
131 	mtx_unlock(&Giant);
132 	return (0);
133 }
134 
135 /*
136  * Duplicate a file descriptor to a particular value.
137  *
138  * note: keep in mind that a potential race condition exists when closing
139  * descriptors from a shared descriptor table (via rfork).
140  */
141 #ifndef _SYS_SYSPROTO_H_
142 struct dup2_args {
143 	u_int	from;
144 	u_int	to;
145 };
146 #endif
147 /*
148  * MPSAFE
149  */
150 /* ARGSUSED */
151 int
152 dup2(td, uap)
153 	struct thread *td;
154 	struct dup2_args *uap;
155 {
156 
157 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
158 		    td->td_retval));
159 }
160 
161 /*
162  * Duplicate a file descriptor.
163  */
164 #ifndef _SYS_SYSPROTO_H_
165 struct dup_args {
166 	u_int	fd;
167 };
168 #endif
169 /*
170  * MPSAFE
171  */
172 /* ARGSUSED */
173 int
174 dup(td, uap)
175 	struct thread *td;
176 	struct dup_args *uap;
177 {
178 
179 	return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
180 }
181 
182 /*
183  * The file control system call.
184  */
185 #ifndef _SYS_SYSPROTO_H_
186 struct fcntl_args {
187 	int	fd;
188 	int	cmd;
189 	long	arg;
190 };
191 #endif
192 /*
193  * MPSAFE
194  */
195 /* ARGSUSED */
196 int
197 fcntl(td, uap)
198 	struct thread *td;
199 	struct fcntl_args *uap;
200 {
201 	struct flock fl;
202 	intptr_t arg;
203 	int error;
204 
205 	error = 0;
206 	switch (uap->cmd) {
207 	case F_GETLK:
208 	case F_SETLK:
209 	case F_SETLKW:
210 		error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
211 		arg = (intptr_t)&fl;
212 		break;
213 	default:
214 		arg = uap->arg;
215 		break;
216 	}
217 	if (error)
218 		return (error);
219 	error = kern_fcntl(td, uap->fd, uap->cmd, arg);
220 	if (error)
221 		return (error);
222 	if (uap->cmd == F_GETLK)
223 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
224 	return (error);
225 }
226 
227 int
228 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
229 {
230 	struct filedesc *fdp;
231 	struct flock *flp;
232 	struct file *fp;
233 	struct proc *p;
234 	char *pop;
235 	struct vnode *vp;
236 	u_int newmin;
237 	int error, flg, tmp;
238 
239 	error = 0;
240 	flg = F_POSIX;
241 	p = td->td_proc;
242 	fdp = p->p_fd;
243 	mtx_lock(&Giant);
244 	FILEDESC_LOCK(fdp);
245 	if ((unsigned)fd >= fdp->fd_nfiles ||
246 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
247 		FILEDESC_UNLOCK(fdp);
248 		error = EBADF;
249 		goto done2;
250 	}
251 	pop = &fdp->fd_ofileflags[fd];
252 
253 	switch (cmd) {
254 	case F_DUPFD:
255 		FILEDESC_UNLOCK(fdp);
256 		newmin = arg;
257 		if (newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
258 		    newmin >= maxfilesperproc) {
259 			error = EINVAL;
260 			break;
261 		}
262 		error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
263 		break;
264 
265 	case F_GETFD:
266 		td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
267 		FILEDESC_UNLOCK(fdp);
268 		break;
269 
270 	case F_SETFD:
271 		*pop = (*pop &~ UF_EXCLOSE) |
272 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
273 		FILEDESC_UNLOCK(fdp);
274 		break;
275 
276 	case F_GETFL:
277 		FILE_LOCK(fp);
278 		FILEDESC_UNLOCK(fdp);
279 		td->td_retval[0] = OFLAGS(fp->f_flag);
280 		FILE_UNLOCK(fp);
281 		break;
282 
283 	case F_SETFL:
284 		FILE_LOCK(fp);
285 		FILEDESC_UNLOCK(fdp);
286 		fhold_locked(fp);
287 		fp->f_flag &= ~FCNTLFLAGS;
288 		fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
289 		FILE_UNLOCK(fp);
290 		tmp = fp->f_flag & FNONBLOCK;
291 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
292 		if (error) {
293 			fdrop(fp, td);
294 			break;
295 		}
296 		tmp = fp->f_flag & FASYNC;
297 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
298 		if (error == 0) {
299 			fdrop(fp, td);
300 			break;
301 		}
302 		FILE_LOCK(fp);
303 		fp->f_flag &= ~FNONBLOCK;
304 		FILE_UNLOCK(fp);
305 		tmp = 0;
306 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
307 		fdrop(fp, td);
308 		break;
309 
310 	case F_GETOWN:
311 		fhold(fp);
312 		FILEDESC_UNLOCK(fdp);
313 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
314 		if (error == 0)
315 			td->td_retval[0] = tmp;
316 		fdrop(fp, td);
317 		break;
318 
319 	case F_SETOWN:
320 		fhold(fp);
321 		FILEDESC_UNLOCK(fdp);
322 		tmp = arg;
323 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
324 		fdrop(fp, td);
325 		break;
326 
327 	case F_SETLKW:
328 		flg |= F_WAIT;
329 		/* FALLTHROUGH F_SETLK */
330 
331 	case F_SETLK:
332 		if (fp->f_type != DTYPE_VNODE) {
333 			FILEDESC_UNLOCK(fdp);
334 			error = EBADF;
335 			break;
336 		}
337 
338 		flp = (struct flock *)arg;
339 		if (flp->l_whence == SEEK_CUR) {
340 			if (fp->f_offset < 0 ||
341 			    (flp->l_start > 0 &&
342 			     fp->f_offset > OFF_MAX - flp->l_start)) {
343 				FILEDESC_UNLOCK(fdp);
344 				error = EOVERFLOW;
345 				break;
346 			}
347 			flp->l_start += fp->f_offset;
348 		}
349 
350 		/*
351 		 * VOP_ADVLOCK() may block.
352 		 */
353 		fhold(fp);
354 		FILEDESC_UNLOCK(fdp);
355 		vp = fp->f_vnode;
356 
357 		switch (flp->l_type) {
358 		case F_RDLCK:
359 			if ((fp->f_flag & FREAD) == 0) {
360 				error = EBADF;
361 				break;
362 			}
363 			PROC_LOCK(p->p_leader);
364 			p->p_leader->p_flag |= P_ADVLOCK;
365 			PROC_UNLOCK(p->p_leader);
366 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
367 			    flp, flg);
368 			break;
369 		case F_WRLCK:
370 			if ((fp->f_flag & FWRITE) == 0) {
371 				error = EBADF;
372 				break;
373 			}
374 			PROC_LOCK(p->p_leader);
375 			p->p_leader->p_flag |= P_ADVLOCK;
376 			PROC_UNLOCK(p->p_leader);
377 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
378 			    flp, flg);
379 			break;
380 		case F_UNLCK:
381 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
382 			    flp, F_POSIX);
383 			break;
384 		default:
385 			error = EINVAL;
386 			break;
387 		}
388 		/* Check for race with close */
389 		FILEDESC_LOCK(fdp);
390 		if ((unsigned) fd >= fdp->fd_nfiles ||
391 		    fp != fdp->fd_ofiles[fd]) {
392 			FILEDESC_UNLOCK(fdp);
393 			flp->l_whence = SEEK_SET;
394 			flp->l_start = 0;
395 			flp->l_len = 0;
396 			flp->l_type = F_UNLCK;
397 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
398 					   F_UNLCK, flp, F_POSIX);
399 		} else
400 			FILEDESC_UNLOCK(fdp);
401 		fdrop(fp, td);
402 		break;
403 
404 	case F_GETLK:
405 		if (fp->f_type != DTYPE_VNODE) {
406 			FILEDESC_UNLOCK(fdp);
407 			error = EBADF;
408 			break;
409 		}
410 		flp = (struct flock *)arg;
411 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
412 		    flp->l_type != F_UNLCK) {
413 			FILEDESC_UNLOCK(fdp);
414 			error = EINVAL;
415 			break;
416 		}
417 		if (flp->l_whence == SEEK_CUR) {
418 			if ((flp->l_start > 0 &&
419 			    fp->f_offset > OFF_MAX - flp->l_start) ||
420 			    (flp->l_start < 0 &&
421 			     fp->f_offset < OFF_MIN - flp->l_start)) {
422 				FILEDESC_UNLOCK(fdp);
423 				error = EOVERFLOW;
424 				break;
425 			}
426 			flp->l_start += fp->f_offset;
427 		}
428 		/*
429 		 * VOP_ADVLOCK() may block.
430 		 */
431 		fhold(fp);
432 		FILEDESC_UNLOCK(fdp);
433 		vp = fp->f_vnode;
434 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
435 		    F_POSIX);
436 		fdrop(fp, td);
437 		break;
438 	default:
439 		FILEDESC_UNLOCK(fdp);
440 		error = EINVAL;
441 		break;
442 	}
443 done2:
444 	mtx_unlock(&Giant);
445 	return (error);
446 }
447 
448 /*
449  * Common code for dup, dup2, and fcntl(F_DUPFD).
450  */
451 static int
452 do_dup(td, type, old, new, retval)
453 	enum dup_type type;
454 	int old, new;
455 	register_t *retval;
456 	struct thread *td;
457 {
458 	struct filedesc *fdp;
459 	struct proc *p;
460 	struct file *fp;
461 	struct file *delfp;
462 	int error, newfd;
463 	int holdleaders;
464 
465 	p = td->td_proc;
466 	fdp = p->p_fd;
467 
468 	/*
469 	 * Verify we have a valid descriptor to dup from and possibly to
470 	 * dup to.
471 	 */
472 	if (old < 0 || new < 0 || new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
473 	    new >= maxfilesperproc)
474 		return (EBADF);
475 	FILEDESC_LOCK(fdp);
476 	if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
477 		FILEDESC_UNLOCK(fdp);
478 		return (EBADF);
479 	}
480 	if (type == DUP_FIXED && old == new) {
481 		*retval = new;
482 		FILEDESC_UNLOCK(fdp);
483 		return (0);
484 	}
485 	fp = fdp->fd_ofiles[old];
486 	fhold(fp);
487 
488 	/*
489 	 * Expand the table for the new descriptor if needed.  This may
490 	 * block and drop and reacquire the filedesc lock.
491 	 */
492 	if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
493 		error = fdalloc(td, new, &newfd);
494 		if (error) {
495 			FILEDESC_UNLOCK(fdp);
496 			fdrop(fp, td);
497 			return (error);
498 		}
499 	}
500 	if (type == DUP_VARIABLE)
501 		new = newfd;
502 
503 	/*
504 	 * If the old file changed out from under us then treat it as a
505 	 * bad file descriptor.  Userland should do its own locking to
506 	 * avoid this case.
507 	 */
508 	if (fdp->fd_ofiles[old] != fp) {
509 		if (fdp->fd_ofiles[new] == NULL) {
510 			if (new < fdp->fd_freefile)
511 				fdp->fd_freefile = new;
512 			while (fdp->fd_lastfile > 0 &&
513 			    fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
514 				fdp->fd_lastfile--;
515 		}
516 		FILEDESC_UNLOCK(fdp);
517 		fdrop(fp, td);
518 		return (EBADF);
519 	}
520 	KASSERT(old != new, ("new fd is same as old"));
521 
522 	/*
523 	 * Save info on the descriptor being overwritten.  We have
524 	 * to do the unmap now, but we cannot close it without
525 	 * introducing an ownership race for the slot.
526 	 */
527 	delfp = fdp->fd_ofiles[new];
528 	if (delfp != NULL && p->p_fdtol != NULL) {
529 		/*
530 		 * Ask fdfree() to sleep to ensure that all relevant
531 		 * process leaders can be traversed in closef().
532 		 */
533 		fdp->fd_holdleaderscount++;
534 		holdleaders = 1;
535 	} else
536 		holdleaders = 0;
537 	KASSERT(delfp == NULL || type == DUP_FIXED,
538 	    ("dup() picked an open file"));
539 #if 0
540 	if (delfp && (fdp->fd_ofileflags[new] & UF_MAPPED))
541 		(void) munmapfd(td, new);
542 #endif
543 
544 	/*
545 	 * Duplicate the source descriptor, update lastfile
546 	 */
547 	fdp->fd_ofiles[new] = fp;
548  	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
549 	if (new > fdp->fd_lastfile)
550 		fdp->fd_lastfile = new;
551 	FILEDESC_UNLOCK(fdp);
552 	*retval = new;
553 
554 	/*
555 	 * If we dup'd over a valid file, we now own the reference to it
556 	 * and must dispose of it using closef() semantics (as if a
557 	 * close() were performed on it).
558 	 */
559 	if (delfp) {
560 		mtx_lock(&Giant);
561 		(void) closef(delfp, td);
562 		mtx_unlock(&Giant);
563 		if (holdleaders) {
564 			FILEDESC_LOCK(fdp);
565 			fdp->fd_holdleaderscount--;
566 			if (fdp->fd_holdleaderscount == 0 &&
567 			    fdp->fd_holdleaderswakeup != 0) {
568 				fdp->fd_holdleaderswakeup = 0;
569 				wakeup(&fdp->fd_holdleaderscount);
570 			}
571 			FILEDESC_UNLOCK(fdp);
572 		}
573 	}
574 	return (0);
575 }
576 
577 /*
578  * If sigio is on the list associated with a process or process group,
579  * disable signalling from the device, remove sigio from the list and
580  * free sigio.
581  */
582 void
583 funsetown(sigiop)
584 	struct sigio **sigiop;
585 {
586 	struct sigio *sigio;
587 
588 	SIGIO_LOCK();
589 	sigio = *sigiop;
590 	if (sigio == NULL) {
591 		SIGIO_UNLOCK();
592 		return;
593 	}
594 	*(sigio->sio_myref) = NULL;
595 	if ((sigio)->sio_pgid < 0) {
596 		struct pgrp *pg = (sigio)->sio_pgrp;
597 		PGRP_LOCK(pg);
598 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
599 			     sigio, sio_pgsigio);
600 		PGRP_UNLOCK(pg);
601 	} else {
602 		struct proc *p = (sigio)->sio_proc;
603 		PROC_LOCK(p);
604 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
605 			     sigio, sio_pgsigio);
606 		PROC_UNLOCK(p);
607 	}
608 	SIGIO_UNLOCK();
609 	crfree(sigio->sio_ucred);
610 	FREE(sigio, M_SIGIO);
611 }
612 
613 /*
614  * Free a list of sigio structures.
615  * We only need to lock the SIGIO_LOCK because we have made ourselves
616  * inaccessable to callers of fsetown and therefore do not need to lock
617  * the proc or pgrp struct for the list manipulation.
618  */
619 void
620 funsetownlst(sigiolst)
621 	struct sigiolst *sigiolst;
622 {
623 	struct proc *p;
624 	struct pgrp *pg;
625 	struct sigio *sigio;
626 
627 	sigio = SLIST_FIRST(sigiolst);
628 	if (sigio == NULL)
629 		return;
630 	p = NULL;
631 	pg = NULL;
632 
633 	/*
634 	 * Every entry of the list should belong
635 	 * to a single proc or pgrp.
636 	 */
637 	if (sigio->sio_pgid < 0) {
638 		pg = sigio->sio_pgrp;
639 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
640 	} else /* if (sigio->sio_pgid > 0) */ {
641 		p = sigio->sio_proc;
642 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
643 	}
644 
645 	SIGIO_LOCK();
646 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
647 		*(sigio->sio_myref) = NULL;
648 		if (pg != NULL) {
649 			KASSERT(sigio->sio_pgid < 0,
650 			    ("Proc sigio in pgrp sigio list"));
651 			KASSERT(sigio->sio_pgrp == pg,
652 			    ("Bogus pgrp in sigio list"));
653 			PGRP_LOCK(pg);
654 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
655 			    sio_pgsigio);
656 			PGRP_UNLOCK(pg);
657 		} else /* if (p != NULL) */ {
658 			KASSERT(sigio->sio_pgid > 0,
659 			    ("Pgrp sigio in proc sigio list"));
660 			KASSERT(sigio->sio_proc == p,
661 			    ("Bogus proc in sigio list"));
662 			PROC_LOCK(p);
663 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
664 			    sio_pgsigio);
665 			PROC_UNLOCK(p);
666 		}
667 		SIGIO_UNLOCK();
668 		crfree(sigio->sio_ucred);
669 		FREE(sigio, M_SIGIO);
670 		SIGIO_LOCK();
671 	}
672 	SIGIO_UNLOCK();
673 }
674 
675 /*
676  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
677  *
678  * After permission checking, add a sigio structure to the sigio list for
679  * the process or process group.
680  */
681 int
682 fsetown(pgid, sigiop)
683 	pid_t pgid;
684 	struct sigio **sigiop;
685 {
686 	struct proc *proc;
687 	struct pgrp *pgrp;
688 	struct sigio *sigio;
689 	int ret;
690 
691 	if (pgid == 0) {
692 		funsetown(sigiop);
693 		return (0);
694 	}
695 
696 	ret = 0;
697 
698 	/* Allocate and fill in the new sigio out of locks. */
699 	MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
700 	sigio->sio_pgid = pgid;
701 	sigio->sio_ucred = crhold(curthread->td_ucred);
702 	sigio->sio_myref = sigiop;
703 
704 	sx_slock(&proctree_lock);
705 	if (pgid > 0) {
706 		proc = pfind(pgid);
707 		if (proc == NULL) {
708 			ret = ESRCH;
709 			goto fail;
710 		}
711 
712 		/*
713 		 * Policy - Don't allow a process to FSETOWN a process
714 		 * in another session.
715 		 *
716 		 * Remove this test to allow maximum flexibility or
717 		 * restrict FSETOWN to the current process or process
718 		 * group for maximum safety.
719 		 */
720 		PROC_UNLOCK(proc);
721 		if (proc->p_session != curthread->td_proc->p_session) {
722 			ret = EPERM;
723 			goto fail;
724 		}
725 
726 		pgrp = NULL;
727 	} else /* if (pgid < 0) */ {
728 		pgrp = pgfind(-pgid);
729 		if (pgrp == NULL) {
730 			ret = ESRCH;
731 			goto fail;
732 		}
733 		PGRP_UNLOCK(pgrp);
734 
735 		/*
736 		 * Policy - Don't allow a process to FSETOWN a process
737 		 * in another session.
738 		 *
739 		 * Remove this test to allow maximum flexibility or
740 		 * restrict FSETOWN to the current process or process
741 		 * group for maximum safety.
742 		 */
743 		if (pgrp->pg_session != curthread->td_proc->p_session) {
744 			ret = EPERM;
745 			goto fail;
746 		}
747 
748 		proc = NULL;
749 	}
750 	funsetown(sigiop);
751 	if (pgid > 0) {
752 		PROC_LOCK(proc);
753 		/*
754 		 * Since funsetownlst() is called without the proctree
755 		 * locked, we need to check for P_WEXIT.
756 		 * XXX: is ESRCH correct?
757 		 */
758 		if ((proc->p_flag & P_WEXIT) != 0) {
759 			PROC_UNLOCK(proc);
760 			ret = ESRCH;
761 			goto fail;
762 		}
763 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
764 		sigio->sio_proc = proc;
765 		PROC_UNLOCK(proc);
766 	} else {
767 		PGRP_LOCK(pgrp);
768 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
769 		sigio->sio_pgrp = pgrp;
770 		PGRP_UNLOCK(pgrp);
771 	}
772 	sx_sunlock(&proctree_lock);
773 	SIGIO_LOCK();
774 	*sigiop = sigio;
775 	SIGIO_UNLOCK();
776 	return (0);
777 
778 fail:
779 	sx_sunlock(&proctree_lock);
780 	crfree(sigio->sio_ucred);
781 	FREE(sigio, M_SIGIO);
782 	return (ret);
783 }
784 
785 /*
786  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
787  */
788 pid_t
789 fgetown(sigiop)
790 	struct sigio **sigiop;
791 {
792 	pid_t pgid;
793 
794 	SIGIO_LOCK();
795 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
796 	SIGIO_UNLOCK();
797 	return (pgid);
798 }
799 
800 /*
801  * Close a file descriptor.
802  */
803 #ifndef _SYS_SYSPROTO_H_
804 struct close_args {
805         int     fd;
806 };
807 #endif
808 /*
809  * MPSAFE
810  */
811 /* ARGSUSED */
812 int
813 close(td, uap)
814 	struct thread *td;
815 	struct close_args *uap;
816 {
817 	struct filedesc *fdp;
818 	struct file *fp;
819 	int fd, error;
820 	int holdleaders;
821 
822 	fd = uap->fd;
823 	error = 0;
824 	holdleaders = 0;
825 	fdp = td->td_proc->p_fd;
826 	mtx_lock(&Giant);
827 	FILEDESC_LOCK(fdp);
828 	if ((unsigned)fd >= fdp->fd_nfiles ||
829 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
830 		FILEDESC_UNLOCK(fdp);
831 		error = EBADF;
832 		goto done2;
833 	}
834 #if 0
835 	if (fdp->fd_ofileflags[fd] & UF_MAPPED)
836 		(void) munmapfd(td, fd);
837 #endif
838 	fdp->fd_ofiles[fd] = NULL;
839 	fdp->fd_ofileflags[fd] = 0;
840 	if (td->td_proc->p_fdtol != NULL) {
841 		/*
842 		 * Ask fdfree() to sleep to ensure that all relevant
843 		 * process leaders can be traversed in closef().
844 		 */
845 		fdp->fd_holdleaderscount++;
846 		holdleaders = 1;
847 	}
848 
849 	/*
850 	 * we now hold the fp reference that used to be owned by the descriptor
851 	 * array.
852 	 */
853 	while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
854 		fdp->fd_lastfile--;
855 	if (fd < fdp->fd_freefile)
856 		fdp->fd_freefile = fd;
857 	if (fd < fdp->fd_knlistsize) {
858 		FILEDESC_UNLOCK(fdp);
859 		knote_fdclose(td, fd);
860 	} else
861 		FILEDESC_UNLOCK(fdp);
862 
863 	error = closef(fp, td);
864 done2:
865 	mtx_unlock(&Giant);
866 	if (holdleaders) {
867 		FILEDESC_LOCK(fdp);
868 		fdp->fd_holdleaderscount--;
869 		if (fdp->fd_holdleaderscount == 0 &&
870 		    fdp->fd_holdleaderswakeup != 0) {
871 			fdp->fd_holdleaderswakeup = 0;
872 			wakeup(&fdp->fd_holdleaderscount);
873 		}
874 		FILEDESC_UNLOCK(fdp);
875 	}
876 	return (error);
877 }
878 
879 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
880 /*
881  * Return status information about a file descriptor.
882  */
883 #ifndef _SYS_SYSPROTO_H_
884 struct ofstat_args {
885 	int	fd;
886 	struct	ostat *sb;
887 };
888 #endif
889 /*
890  * MPSAFE
891  */
892 /* ARGSUSED */
893 int
894 ofstat(td, uap)
895 	struct thread *td;
896 	struct ofstat_args *uap;
897 {
898 	struct file *fp;
899 	struct stat ub;
900 	struct ostat oub;
901 	int error;
902 
903 	if ((error = fget(td, uap->fd, &fp)) != 0)
904 		goto done2;
905 	mtx_lock(&Giant);
906 	error = fo_stat(fp, &ub, td->td_ucred, td);
907 	mtx_unlock(&Giant);
908 	if (error == 0) {
909 		cvtstat(&ub, &oub);
910 		error = copyout(&oub, uap->sb, sizeof(oub));
911 	}
912 	fdrop(fp, td);
913 done2:
914 	return (error);
915 }
916 #endif /* COMPAT_43 || COMPAT_SUNOS */
917 
918 /*
919  * Return status information about a file descriptor.
920  */
921 #ifndef _SYS_SYSPROTO_H_
922 struct fstat_args {
923 	int	fd;
924 	struct	stat *sb;
925 };
926 #endif
927 /*
928  * MPSAFE
929  */
930 /* ARGSUSED */
931 int
932 fstat(td, uap)
933 	struct thread *td;
934 	struct fstat_args *uap;
935 {
936 	struct file *fp;
937 	struct stat ub;
938 	int error;
939 
940 	if ((error = fget(td, uap->fd, &fp)) != 0)
941 		goto done2;
942 	mtx_lock(&Giant);
943 	error = fo_stat(fp, &ub, td->td_ucred, td);
944 	mtx_unlock(&Giant);
945 	if (error == 0)
946 		error = copyout(&ub, uap->sb, sizeof(ub));
947 	fdrop(fp, td);
948 done2:
949 	return (error);
950 }
951 
952 /*
953  * Return status information about a file descriptor.
954  */
955 #ifndef _SYS_SYSPROTO_H_
956 struct nfstat_args {
957 	int	fd;
958 	struct	nstat *sb;
959 };
960 #endif
961 /*
962  * MPSAFE
963  */
964 /* ARGSUSED */
965 int
966 nfstat(td, uap)
967 	struct thread *td;
968 	struct nfstat_args *uap;
969 {
970 	struct file *fp;
971 	struct stat ub;
972 	struct nstat nub;
973 	int error;
974 
975 	if ((error = fget(td, uap->fd, &fp)) != 0)
976 		goto done2;
977 	mtx_lock(&Giant);
978 	error = fo_stat(fp, &ub, td->td_ucred, td);
979 	mtx_unlock(&Giant);
980 	if (error == 0) {
981 		cvtnstat(&ub, &nub);
982 		error = copyout(&nub, uap->sb, sizeof(nub));
983 	}
984 	fdrop(fp, td);
985 done2:
986 	return (error);
987 }
988 
989 /*
990  * Return pathconf information about a file descriptor.
991  */
992 #ifndef _SYS_SYSPROTO_H_
993 struct fpathconf_args {
994 	int	fd;
995 	int	name;
996 };
997 #endif
998 /*
999  * MPSAFE
1000  */
1001 /* ARGSUSED */
1002 int
1003 fpathconf(td, uap)
1004 	struct thread *td;
1005 	struct fpathconf_args *uap;
1006 {
1007 	struct file *fp;
1008 	struct vnode *vp;
1009 	int error;
1010 
1011 	if ((error = fget(td, uap->fd, &fp)) != 0)
1012 		return (error);
1013 
1014 	/* If asynchronous I/O is available, it works for all descriptors. */
1015 	if (uap->name == _PC_ASYNC_IO) {
1016 		td->td_retval[0] = async_io_version;
1017 		goto out;
1018 	}
1019 	vp = fp->f_vnode;
1020 	if (vp != NULL) {
1021 		mtx_lock(&Giant);
1022 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1023 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1024 		VOP_UNLOCK(vp, 0, td);
1025 		mtx_unlock(&Giant);
1026 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1027 		if (uap->name != _PC_PIPE_BUF) {
1028 			error = EINVAL;
1029 		} else {
1030 			td->td_retval[0] = PIPE_BUF;
1031 		error = 0;
1032 		}
1033 	} else {
1034 		error = EOPNOTSUPP;
1035 	}
1036 out:
1037 	fdrop(fp, td);
1038 	return (error);
1039 }
1040 
1041 /*
1042  * Allocate a file descriptor for the process.
1043  */
1044 static int fdexpand;
1045 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
1046 
1047 int
1048 fdalloc(td, want, result)
1049 	struct thread *td;
1050 	int want;
1051 	int *result;
1052 {
1053 	struct proc *p = td->td_proc;
1054 	struct filedesc *fdp = td->td_proc->p_fd;
1055 	int i;
1056 	int lim, last, nfiles;
1057 	struct file **newofile, **oldofile;
1058 	char *newofileflags;
1059 
1060 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1061 
1062 	/*
1063 	 * Search for a free descriptor starting at the higher
1064 	 * of want or fd_freefile.  If that fails, consider
1065 	 * expanding the ofile array.
1066 	 */
1067 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1068 	for (;;) {
1069 		last = min(fdp->fd_nfiles, lim);
1070 		i = max(want, fdp->fd_freefile);
1071 		for (; i < last; i++) {
1072 			if (fdp->fd_ofiles[i] == NULL) {
1073 				fdp->fd_ofileflags[i] = 0;
1074 				if (i > fdp->fd_lastfile)
1075 					fdp->fd_lastfile = i;
1076 				if (want <= fdp->fd_freefile)
1077 					fdp->fd_freefile = i;
1078 				*result = i;
1079 				return (0);
1080 			}
1081 		}
1082 
1083 		/*
1084 		 * No space in current array.  Expand?
1085 		 */
1086 		if (i >= lim)
1087 			return (EMFILE);
1088 		if (fdp->fd_nfiles < NDEXTENT)
1089 			nfiles = NDEXTENT;
1090 		else
1091 			nfiles = 2 * fdp->fd_nfiles;
1092 		while (nfiles < want)
1093 			nfiles <<= 1;
1094 		FILEDESC_UNLOCK(fdp);
1095 		newofile = malloc(nfiles * OFILESIZE, M_FILEDESC, M_WAITOK);
1096 
1097 		/*
1098 		 * Deal with file-table extend race that might have
1099 		 * occurred while filedesc was unlocked.
1100 		 */
1101 		FILEDESC_LOCK(fdp);
1102 		if (fdp->fd_nfiles >= nfiles) {
1103 			FILEDESC_UNLOCK(fdp);
1104 			free(newofile, M_FILEDESC);
1105 			FILEDESC_LOCK(fdp);
1106 			continue;
1107 		}
1108 		newofileflags = (char *) &newofile[nfiles];
1109 		/*
1110 		 * Copy the existing ofile and ofileflags arrays
1111 		 * and zero the new portion of each array.
1112 		 */
1113 		i = fdp->fd_nfiles * sizeof(struct file *);
1114 		bcopy(fdp->fd_ofiles, newofile,	i);
1115 		bzero((char *)newofile + i,
1116 		    nfiles * sizeof(struct file *) - i);
1117 		i = fdp->fd_nfiles * sizeof(char);
1118 		bcopy(fdp->fd_ofileflags, newofileflags, i);
1119 		bzero(newofileflags + i, nfiles * sizeof(char) - i);
1120 		if (fdp->fd_nfiles > NDFILE)
1121 			oldofile = fdp->fd_ofiles;
1122 		else
1123 			oldofile = NULL;
1124 		fdp->fd_ofiles = newofile;
1125 		fdp->fd_ofileflags = newofileflags;
1126 		fdp->fd_nfiles = nfiles;
1127 		fdexpand++;
1128 		if (oldofile != NULL) {
1129 			FILEDESC_UNLOCK(fdp);
1130 			free(oldofile, M_FILEDESC);
1131 			FILEDESC_LOCK(fdp);
1132 		}
1133 	}
1134 }
1135 
1136 /*
1137  * Check to see whether n user file descriptors
1138  * are available to the process p.
1139  */
1140 int
1141 fdavail(td, n)
1142 	struct thread *td;
1143 	int n;
1144 {
1145 	struct proc *p = td->td_proc;
1146 	struct filedesc *fdp = td->td_proc->p_fd;
1147 	struct file **fpp;
1148 	int i, lim, last;
1149 
1150 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1151 
1152 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
1153 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1154 		return (1);
1155 	last = min(fdp->fd_nfiles, lim);
1156 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1157 	for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1158 		if (*fpp == NULL && --n <= 0)
1159 			return (1);
1160 	}
1161 	return (0);
1162 }
1163 
1164 /*
1165  * Create a new open file structure and allocate
1166  * a file decriptor for the process that refers to it.
1167  */
1168 int
1169 falloc(td, resultfp, resultfd)
1170 	struct thread *td;
1171 	struct file **resultfp;
1172 	int *resultfd;
1173 {
1174 	struct proc *p = td->td_proc;
1175 	struct file *fp, *fq;
1176 	int error, i;
1177 	int maxuserfiles = maxfiles - (maxfiles / 20);
1178 	static struct timeval lastfail;
1179 	static int curfail;
1180 
1181 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1182 	sx_xlock(&filelist_lock);
1183 	if ((nfiles >= maxuserfiles && td->td_ucred->cr_ruid != 0)
1184 	   || nfiles >= maxfiles) {
1185 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1186 			printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1187 				td->td_ucred->cr_ruid);
1188 		}
1189 		sx_xunlock(&filelist_lock);
1190 		uma_zfree(file_zone, fp);
1191 		return (ENFILE);
1192 	}
1193 	nfiles++;
1194 
1195 	/*
1196 	 * If the process has file descriptor zero open, add the new file
1197 	 * descriptor to the list of open files at that point, otherwise
1198 	 * put it at the front of the list of open files.
1199 	 */
1200 	fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1201 	fp->f_count = 1;
1202 	fp->f_cred = crhold(td->td_ucred);
1203 	fp->f_ops = &badfileops;
1204 	FILEDESC_LOCK(p->p_fd);
1205 	if ((fq = p->p_fd->fd_ofiles[0])) {
1206 		LIST_INSERT_AFTER(fq, fp, f_list);
1207 	} else {
1208 		LIST_INSERT_HEAD(&filehead, fp, f_list);
1209 	}
1210 	sx_xunlock(&filelist_lock);
1211 	if ((error = fdalloc(td, 0, &i))) {
1212 		FILEDESC_UNLOCK(p->p_fd);
1213 		fdrop(fp, td);
1214 		return (error);
1215 	}
1216 	p->p_fd->fd_ofiles[i] = fp;
1217 	FILEDESC_UNLOCK(p->p_fd);
1218 	if (resultfp)
1219 		*resultfp = fp;
1220 	if (resultfd)
1221 		*resultfd = i;
1222 	return (0);
1223 }
1224 
1225 /*
1226  * Free a file descriptor.
1227  */
1228 void
1229 ffree(fp)
1230 	struct file *fp;
1231 {
1232 
1233 	KASSERT(fp->f_count == 0, ("ffree: fp_fcount not 0!"));
1234 	sx_xlock(&filelist_lock);
1235 	LIST_REMOVE(fp, f_list);
1236 	nfiles--;
1237 	sx_xunlock(&filelist_lock);
1238 	crfree(fp->f_cred);
1239 	uma_zfree(file_zone, fp);
1240 }
1241 
1242 /*
1243  * Build a new filedesc structure from another.
1244  * Copy the current, root, and jail root vnode references.
1245  */
1246 struct filedesc *
1247 fdinit(fdp)
1248 	struct filedesc *fdp;
1249 {
1250 	struct filedesc0 *newfdp;
1251 
1252 	MALLOC(newfdp, struct filedesc0 *, sizeof(struct filedesc0),
1253 	    M_FILEDESC, M_WAITOK | M_ZERO);
1254 	mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1255 	newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1256 	if (newfdp->fd_fd.fd_cdir)
1257 		VREF(newfdp->fd_fd.fd_cdir);
1258 	newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1259 	if (newfdp->fd_fd.fd_rdir)
1260 		VREF(newfdp->fd_fd.fd_rdir);
1261 	newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1262 	if (newfdp->fd_fd.fd_jdir)
1263 		VREF(newfdp->fd_fd.fd_jdir);
1264 
1265 	/* Create the file descriptor table. */
1266 	newfdp->fd_fd.fd_refcnt = 1;
1267 	newfdp->fd_fd.fd_cmask = cmask;
1268 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1269 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1270 	newfdp->fd_fd.fd_nfiles = NDFILE;
1271 	newfdp->fd_fd.fd_knlistsize = -1;
1272 	return (&newfdp->fd_fd);
1273 }
1274 
1275 /*
1276  * Share a filedesc structure.
1277  */
1278 struct filedesc *
1279 fdshare(fdp)
1280 	struct filedesc *fdp;
1281 {
1282 	FILEDESC_LOCK(fdp);
1283 	fdp->fd_refcnt++;
1284 	FILEDESC_UNLOCK(fdp);
1285 	return (fdp);
1286 }
1287 
1288 /*
1289  * Copy a filedesc structure.
1290  * A NULL pointer in returns a NULL reference, this is to ease callers,
1291  * not catch errors.
1292  */
1293 struct filedesc *
1294 fdcopy(fdp)
1295 	struct filedesc *fdp;
1296 {
1297 	struct filedesc *newfdp;
1298 	struct file **fpp;
1299 	int i, j;
1300 
1301 	/* Certain daemons might not have file descriptors. */
1302 	if (fdp == NULL)
1303 		return (NULL);
1304 
1305 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1306 
1307 	FILEDESC_UNLOCK(fdp);
1308 	MALLOC(newfdp, struct filedesc *, sizeof(struct filedesc0),
1309 	    M_FILEDESC, M_WAITOK);
1310 	FILEDESC_LOCK(fdp);
1311 	bcopy(fdp, newfdp, sizeof(struct filedesc));
1312 	FILEDESC_UNLOCK(fdp);
1313 	bzero(&newfdp->fd_mtx, sizeof(newfdp->fd_mtx));
1314 	mtx_init(&newfdp->fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1315 	if (newfdp->fd_cdir)
1316 		VREF(newfdp->fd_cdir);
1317 	if (newfdp->fd_rdir)
1318 		VREF(newfdp->fd_rdir);
1319 	if (newfdp->fd_jdir)
1320 		VREF(newfdp->fd_jdir);
1321 	newfdp->fd_refcnt = 1;
1322 
1323 	/*
1324 	 * If the number of open files fits in the internal arrays
1325 	 * of the open file structure, use them, otherwise allocate
1326 	 * additional memory for the number of descriptors currently
1327 	 * in use.
1328 	 */
1329 	FILEDESC_LOCK(fdp);
1330 	newfdp->fd_lastfile = fdp->fd_lastfile;
1331 	newfdp->fd_nfiles = fdp->fd_nfiles;
1332 	if (newfdp->fd_lastfile < NDFILE) {
1333 		newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
1334 		newfdp->fd_ofileflags =
1335 		    ((struct filedesc0 *) newfdp)->fd_dfileflags;
1336 		i = NDFILE;
1337 	} else {
1338 		/*
1339 		 * Compute the smallest multiple of NDEXTENT needed
1340 		 * for the file descriptors currently in use,
1341 		 * allowing the table to shrink.
1342 		 */
1343 retry:
1344 		i = newfdp->fd_nfiles;
1345 		while (i > 2 * NDEXTENT && i > newfdp->fd_lastfile * 2)
1346 			i /= 2;
1347 		FILEDESC_UNLOCK(fdp);
1348 		MALLOC(newfdp->fd_ofiles, struct file **, i * OFILESIZE,
1349 		    M_FILEDESC, M_WAITOK);
1350 		FILEDESC_LOCK(fdp);
1351 		newfdp->fd_lastfile = fdp->fd_lastfile;
1352 		newfdp->fd_nfiles = fdp->fd_nfiles;
1353 		j = newfdp->fd_nfiles;
1354 		while (j > 2 * NDEXTENT && j > newfdp->fd_lastfile * 2)
1355 			j /= 2;
1356 		if (i != j) {
1357 			/*
1358 			 * The size of the original table has changed.
1359 			 * Go over once again.
1360 			 */
1361 			FILEDESC_UNLOCK(fdp);
1362 			FREE(newfdp->fd_ofiles, M_FILEDESC);
1363 			FILEDESC_LOCK(fdp);
1364 			newfdp->fd_lastfile = fdp->fd_lastfile;
1365 			newfdp->fd_nfiles = fdp->fd_nfiles;
1366 			goto retry;
1367 		}
1368 		newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
1369 	}
1370 	newfdp->fd_nfiles = i;
1371 	bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **));
1372 	bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char));
1373 
1374 	/*
1375 	 * kq descriptors cannot be copied.
1376 	 */
1377 	if (newfdp->fd_knlistsize != -1) {
1378 		fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile];
1379 		for (i = newfdp->fd_lastfile; i >= 0; i--, fpp--) {
1380 			if (*fpp != NULL && (*fpp)->f_type == DTYPE_KQUEUE) {
1381 				*fpp = NULL;
1382 				if (i < newfdp->fd_freefile)
1383 					newfdp->fd_freefile = i;
1384 			}
1385 			if (*fpp == NULL && i == newfdp->fd_lastfile && i > 0)
1386 				newfdp->fd_lastfile--;
1387 		}
1388 		newfdp->fd_knlist = NULL;
1389 		newfdp->fd_knlistsize = -1;
1390 		newfdp->fd_knhash = NULL;
1391 		newfdp->fd_knhashmask = 0;
1392 	}
1393 
1394 	fpp = newfdp->fd_ofiles;
1395 	for (i = newfdp->fd_lastfile; i-- >= 0; fpp++) {
1396 		if (*fpp != NULL)
1397 			fhold(*fpp);
1398 	}
1399 	return (newfdp);
1400 }
1401 
1402 /* A mutex to protect the association between a proc and filedesc. */
1403 struct mtx	fdesc_mtx;
1404 MTX_SYSINIT(fdesc, &fdesc_mtx, "fdesc", MTX_DEF);
1405 
1406 /*
1407  * Release a filedesc structure.
1408  */
1409 void
1410 fdfree(td)
1411 	struct thread *td;
1412 {
1413 	struct filedesc *fdp;
1414 	struct file **fpp;
1415 	int i;
1416 	struct filedesc_to_leader *fdtol;
1417 	struct file *fp;
1418 	struct vnode *vp;
1419 	struct flock lf;
1420 
1421 	/* Certain daemons might not have file descriptors. */
1422 	fdp = td->td_proc->p_fd;
1423 	if (fdp == NULL)
1424 		return;
1425 
1426 	/* Check for special need to clear POSIX style locks */
1427 	fdtol = td->td_proc->p_fdtol;
1428 	if (fdtol != NULL) {
1429 		FILEDESC_LOCK(fdp);
1430 		KASSERT(fdtol->fdl_refcount > 0,
1431 			("filedesc_to_refcount botch: fdl_refcount=%d",
1432 			 fdtol->fdl_refcount));
1433 		if (fdtol->fdl_refcount == 1 &&
1434 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1435 			i = 0;
1436 			fpp = fdp->fd_ofiles;
1437 			for (i = 0, fpp = fdp->fd_ofiles;
1438 			     i < fdp->fd_lastfile;
1439 			     i++, fpp++) {
1440 				if (*fpp == NULL ||
1441 				    (*fpp)->f_type != DTYPE_VNODE)
1442 					continue;
1443 				fp = *fpp;
1444 				fhold(fp);
1445 				FILEDESC_UNLOCK(fdp);
1446 				lf.l_whence = SEEK_SET;
1447 				lf.l_start = 0;
1448 				lf.l_len = 0;
1449 				lf.l_type = F_UNLCK;
1450 				vp = fp->f_vnode;
1451 				(void) VOP_ADVLOCK(vp,
1452 						   (caddr_t)td->td_proc->
1453 						   p_leader,
1454 						   F_UNLCK,
1455 						   &lf,
1456 						   F_POSIX);
1457 				FILEDESC_LOCK(fdp);
1458 				fdrop(fp, td);
1459 				fpp = fdp->fd_ofiles + i;
1460 			}
1461 		}
1462 	retry:
1463 		if (fdtol->fdl_refcount == 1) {
1464 			if (fdp->fd_holdleaderscount > 0 &&
1465 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1466 				/*
1467 				 * close() or do_dup() has cleared a reference
1468 				 * in a shared file descriptor table.
1469 				 */
1470 				fdp->fd_holdleaderswakeup = 1;
1471 				msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1472 				       PLOCK, "fdlhold", 0);
1473 				goto retry;
1474 			}
1475 			if (fdtol->fdl_holdcount > 0) {
1476 				/*
1477 				 * Ensure that fdtol->fdl_leader
1478 				 * remains valid in closef().
1479 				 */
1480 				fdtol->fdl_wakeup = 1;
1481 				msleep(fdtol, &fdp->fd_mtx,
1482 				       PLOCK, "fdlhold", 0);
1483 				goto retry;
1484 			}
1485 		}
1486 		fdtol->fdl_refcount--;
1487 		if (fdtol->fdl_refcount == 0 &&
1488 		    fdtol->fdl_holdcount == 0) {
1489 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1490 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1491 		} else
1492 			fdtol = NULL;
1493 		td->td_proc->p_fdtol = NULL;
1494 		FILEDESC_UNLOCK(fdp);
1495 		if (fdtol != NULL)
1496 			FREE(fdtol, M_FILEDESC_TO_LEADER);
1497 	}
1498 	FILEDESC_LOCK(fdp);
1499 	if (--fdp->fd_refcnt > 0) {
1500 		FILEDESC_UNLOCK(fdp);
1501 		return;
1502 	}
1503 
1504 	/*
1505 	 * We are the last reference to the structure, so we can
1506 	 * safely assume it will not change out from under us.
1507 	 */
1508 	FILEDESC_UNLOCK(fdp);
1509 	fpp = fdp->fd_ofiles;
1510 	for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1511 		if (*fpp)
1512 			(void) closef(*fpp, td);
1513 	}
1514 
1515 	/* XXX This should happen earlier. */
1516 	mtx_lock(&fdesc_mtx);
1517 	td->td_proc->p_fd = NULL;
1518 	mtx_unlock(&fdesc_mtx);
1519 
1520 	if (fdp->fd_nfiles > NDFILE)
1521 		FREE(fdp->fd_ofiles, M_FILEDESC);
1522 	if (fdp->fd_cdir)
1523 		vrele(fdp->fd_cdir);
1524 	if (fdp->fd_rdir)
1525 		vrele(fdp->fd_rdir);
1526 	if (fdp->fd_jdir)
1527 		vrele(fdp->fd_jdir);
1528 	if (fdp->fd_knlist)
1529 		FREE(fdp->fd_knlist, M_KQUEUE);
1530 	if (fdp->fd_knhash)
1531 		FREE(fdp->fd_knhash, M_KQUEUE);
1532 	mtx_destroy(&fdp->fd_mtx);
1533 	FREE(fdp, M_FILEDESC);
1534 }
1535 
1536 /*
1537  * For setugid programs, we don't want to people to use that setugidness
1538  * to generate error messages which write to a file which otherwise would
1539  * otherwise be off-limits to the process.  We check for filesystems where
1540  * the vnode can change out from under us after execve (like [lin]procfs).
1541  *
1542  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1543  * sufficient.  We also don't for check setugidness since we know we are.
1544  */
1545 static int
1546 is_unsafe(struct file *fp)
1547 {
1548 	if (fp->f_type == DTYPE_VNODE) {
1549 		struct vnode *vp = fp->f_vnode;
1550 
1551 		if ((vp->v_vflag & VV_PROCDEP) != 0)
1552 			return (1);
1553 	}
1554 	return (0);
1555 }
1556 
1557 /*
1558  * Make this setguid thing safe, if at all possible.
1559  */
1560 void
1561 setugidsafety(td)
1562 	struct thread *td;
1563 {
1564 	struct filedesc *fdp;
1565 	int i;
1566 
1567 	/* Certain daemons might not have file descriptors. */
1568 	fdp = td->td_proc->p_fd;
1569 	if (fdp == NULL)
1570 		return;
1571 
1572 	/*
1573 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
1574 	 * we are blocked in a close.  Be careful!
1575 	 */
1576 	FILEDESC_LOCK(fdp);
1577 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1578 		if (i > 2)
1579 			break;
1580 		if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1581 			struct file *fp;
1582 
1583 #if 0
1584 			if ((fdp->fd_ofileflags[i] & UF_MAPPED) != 0)
1585 				(void) munmapfd(td, i);
1586 #endif
1587 			if (i < fdp->fd_knlistsize) {
1588 				FILEDESC_UNLOCK(fdp);
1589 				knote_fdclose(td, i);
1590 				FILEDESC_LOCK(fdp);
1591 			}
1592 			/*
1593 			 * NULL-out descriptor prior to close to avoid
1594 			 * a race while close blocks.
1595 			 */
1596 			fp = fdp->fd_ofiles[i];
1597 			fdp->fd_ofiles[i] = NULL;
1598 			fdp->fd_ofileflags[i] = 0;
1599 			if (i < fdp->fd_freefile)
1600 				fdp->fd_freefile = i;
1601 			FILEDESC_UNLOCK(fdp);
1602 			(void) closef(fp, td);
1603 			FILEDESC_LOCK(fdp);
1604 		}
1605 	}
1606 	while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
1607 		fdp->fd_lastfile--;
1608 	FILEDESC_UNLOCK(fdp);
1609 }
1610 
1611 /*
1612  * Close any files on exec?
1613  */
1614 void
1615 fdcloseexec(td)
1616 	struct thread *td;
1617 {
1618 	struct filedesc *fdp;
1619 	int i;
1620 
1621 	/* Certain daemons might not have file descriptors. */
1622 	fdp = td->td_proc->p_fd;
1623 	if (fdp == NULL)
1624 		return;
1625 
1626 	FILEDESC_LOCK(fdp);
1627 
1628 	/*
1629 	 * We cannot cache fd_ofiles or fd_ofileflags since operations
1630 	 * may block and rip them out from under us.
1631 	 */
1632 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1633 		if (fdp->fd_ofiles[i] != NULL &&
1634 		    (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1635 			struct file *fp;
1636 
1637 #if 0
1638 			if (fdp->fd_ofileflags[i] & UF_MAPPED)
1639 				(void) munmapfd(td, i);
1640 #endif
1641 			if (i < fdp->fd_knlistsize) {
1642 				FILEDESC_UNLOCK(fdp);
1643 				knote_fdclose(td, i);
1644 				FILEDESC_LOCK(fdp);
1645 			}
1646 			/*
1647 			 * NULL-out descriptor prior to close to avoid
1648 			 * a race while close blocks.
1649 			 */
1650 			fp = fdp->fd_ofiles[i];
1651 			fdp->fd_ofiles[i] = NULL;
1652 			fdp->fd_ofileflags[i] = 0;
1653 			if (i < fdp->fd_freefile)
1654 				fdp->fd_freefile = i;
1655 			FILEDESC_UNLOCK(fdp);
1656 			(void) closef(fp, td);
1657 			FILEDESC_LOCK(fdp);
1658 		}
1659 	}
1660 	while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
1661 		fdp->fd_lastfile--;
1662 	FILEDESC_UNLOCK(fdp);
1663 }
1664 
1665 /*
1666  * It is unsafe for set[ug]id processes to be started with file
1667  * descriptors 0..2 closed, as these descriptors are given implicit
1668  * significance in the Standard C library.  fdcheckstd() will create a
1669  * descriptor referencing /dev/null for each of stdin, stdout, and
1670  * stderr that is not already open.
1671  */
1672 int
1673 fdcheckstd(td)
1674 	struct thread *td;
1675 {
1676 	struct nameidata nd;
1677 	struct filedesc *fdp;
1678 	struct file *fp;
1679 	register_t retval;
1680 	int fd, i, error, flags, devnull;
1681 
1682 	fdp = td->td_proc->p_fd;
1683 	if (fdp == NULL)
1684 		return (0);
1685 	devnull = -1;
1686 	error = 0;
1687 	for (i = 0; i < 3; i++) {
1688 		if (fdp->fd_ofiles[i] != NULL)
1689 			continue;
1690 		if (devnull < 0) {
1691 			error = falloc(td, &fp, &fd);
1692 			if (error != 0)
1693 				break;
1694 			KASSERT(fd == i, ("oof, we didn't get our fd"));
1695 			NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null",
1696 			    td);
1697 			flags = FREAD | FWRITE;
1698 			error = vn_open(&nd, &flags, 0, -1);
1699 			if (error != 0) {
1700 				FILEDESC_LOCK(fdp);
1701 				fdp->fd_ofiles[fd] = NULL;
1702 				FILEDESC_UNLOCK(fdp);
1703 				fdrop(fp, td);
1704 				break;
1705 			}
1706 			NDFREE(&nd, NDF_ONLY_PNBUF);
1707 			fp->f_vnode = nd.ni_vp;
1708 			fp->f_data = nd.ni_vp;
1709 			fp->f_flag = flags;
1710 			fp->f_ops = &vnops;
1711 			fp->f_type = DTYPE_VNODE;
1712 			VOP_UNLOCK(nd.ni_vp, 0, td);
1713 			devnull = fd;
1714 		} else {
1715 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1716 			if (error != 0)
1717 				break;
1718 		}
1719 	}
1720 	return (error);
1721 }
1722 
1723 /*
1724  * Internal form of close.
1725  * Decrement reference count on file structure.
1726  * Note: td may be NULL when closing a file
1727  * that was being passed in a message.
1728  */
1729 int
1730 closef(fp, td)
1731 	struct file *fp;
1732 	struct thread *td;
1733 {
1734 	struct vnode *vp;
1735 	struct flock lf;
1736 	struct filedesc_to_leader *fdtol;
1737 	struct filedesc *fdp;
1738 
1739 	if (fp == NULL)
1740 		return (0);
1741 	/*
1742 	 * POSIX record locking dictates that any close releases ALL
1743 	 * locks owned by this process.  This is handled by setting
1744 	 * a flag in the unlock to free ONLY locks obeying POSIX
1745 	 * semantics, and not to free BSD-style file locks.
1746 	 * If the descriptor was in a message, POSIX-style locks
1747 	 * aren't passed with the descriptor.
1748 	 */
1749 	if (td != NULL &&
1750 	    fp->f_type == DTYPE_VNODE) {
1751 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1752 			lf.l_whence = SEEK_SET;
1753 			lf.l_start = 0;
1754 			lf.l_len = 0;
1755 			lf.l_type = F_UNLCK;
1756 			vp = fp->f_vnode;
1757 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1758 					   F_UNLCK, &lf, F_POSIX);
1759 		}
1760 		fdtol = td->td_proc->p_fdtol;
1761 		if (fdtol != NULL) {
1762 			/*
1763 			 * Handle special case where file descriptor table
1764 			 * is shared between multiple process leaders.
1765 			 */
1766 			fdp = td->td_proc->p_fd;
1767 			FILEDESC_LOCK(fdp);
1768 			for (fdtol = fdtol->fdl_next;
1769 			     fdtol != td->td_proc->p_fdtol;
1770 			     fdtol = fdtol->fdl_next) {
1771 				if ((fdtol->fdl_leader->p_flag &
1772 				     P_ADVLOCK) == 0)
1773 					continue;
1774 				fdtol->fdl_holdcount++;
1775 				FILEDESC_UNLOCK(fdp);
1776 				lf.l_whence = SEEK_SET;
1777 				lf.l_start = 0;
1778 				lf.l_len = 0;
1779 				lf.l_type = F_UNLCK;
1780 				vp = fp->f_vnode;
1781 				(void) VOP_ADVLOCK(vp,
1782 						   (caddr_t)fdtol->fdl_leader,
1783 						   F_UNLCK, &lf, F_POSIX);
1784 				FILEDESC_LOCK(fdp);
1785 				fdtol->fdl_holdcount--;
1786 				if (fdtol->fdl_holdcount == 0 &&
1787 				    fdtol->fdl_wakeup != 0) {
1788 					fdtol->fdl_wakeup = 0;
1789 					wakeup(fdtol);
1790 				}
1791 			}
1792 			FILEDESC_UNLOCK(fdp);
1793 		}
1794 	}
1795 	return (fdrop(fp, td));
1796 }
1797 
1798 /*
1799  * Drop reference on struct file passed in, may call closef if the
1800  * reference hits zero.
1801  */
1802 int
1803 fdrop(fp, td)
1804 	struct file *fp;
1805 	struct thread *td;
1806 {
1807 
1808 	FILE_LOCK(fp);
1809 	return (fdrop_locked(fp, td));
1810 }
1811 
1812 /*
1813  * Extract the file pointer associated with the specified descriptor for
1814  * the current user process.
1815  *
1816  * If the descriptor doesn't exist, EBADF is returned.
1817  *
1818  * If the descriptor exists but doesn't match 'flags' then
1819  * return EBADF for read attempts and EINVAL for write attempts.
1820  *
1821  * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1822  * It should be droped with fdrop().
1823  * If it is not set, then the refcount will not be bumped however the
1824  * thread's filedesc struct will be returned locked (for fgetsock).
1825  *
1826  * If an error occured the non-zero error is returned and *fpp is set to NULL.
1827  * Otherwise *fpp is set and zero is returned.
1828  */
1829 static __inline int
1830 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1831 {
1832 	struct filedesc *fdp;
1833 	struct file *fp;
1834 
1835 	*fpp = NULL;
1836 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1837 		return (EBADF);
1838 	FILEDESC_LOCK(fdp);
1839 	if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
1840 		FILEDESC_UNLOCK(fdp);
1841 		return (EBADF);
1842 	}
1843 
1844 	/*
1845 	 * Note: FREAD failures returns EBADF to maintain backwards
1846 	 * compatibility with what routines returned before.
1847 	 *
1848 	 * Only one flag, or 0, may be specified.
1849 	 */
1850 	if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
1851 		FILEDESC_UNLOCK(fdp);
1852 		return (EBADF);
1853 	}
1854 	if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
1855 		FILEDESC_UNLOCK(fdp);
1856 		return (EINVAL);
1857 	}
1858 	if (hold) {
1859 		fhold(fp);
1860 		FILEDESC_UNLOCK(fdp);
1861 	}
1862 	*fpp = fp;
1863 	return (0);
1864 }
1865 
1866 int
1867 fget(struct thread *td, int fd, struct file **fpp)
1868 {
1869 
1870 	return(_fget(td, fd, fpp, 0, 1));
1871 }
1872 
1873 int
1874 fget_read(struct thread *td, int fd, struct file **fpp)
1875 {
1876 
1877 	return(_fget(td, fd, fpp, FREAD, 1));
1878 }
1879 
1880 int
1881 fget_write(struct thread *td, int fd, struct file **fpp)
1882 {
1883 
1884 	return(_fget(td, fd, fpp, FWRITE, 1));
1885 }
1886 
1887 /*
1888  * Like fget() but loads the underlying vnode, or returns an error if
1889  * the descriptor does not represent a vnode.  Note that pipes use vnodes
1890  * but never have VM objects (so VOP_GETVOBJECT() calls will return an
1891  * error).  The returned vnode will be vref()d.
1892  */
1893 static __inline int
1894 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
1895 {
1896 	struct file *fp;
1897 	int error;
1898 
1899 	*vpp = NULL;
1900 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
1901 		return (error);
1902 	if (fp->f_vnode == NULL) {
1903 		error = EINVAL;
1904 	} else {
1905 		*vpp = fp->f_vnode;
1906 		vref(*vpp);
1907 	}
1908 	FILEDESC_UNLOCK(td->td_proc->p_fd);
1909 	return (error);
1910 }
1911 
1912 int
1913 fgetvp(struct thread *td, int fd, struct vnode **vpp)
1914 {
1915 
1916 	return (_fgetvp(td, fd, vpp, 0));
1917 }
1918 
1919 int
1920 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
1921 {
1922 
1923 	return (_fgetvp(td, fd, vpp, FREAD));
1924 }
1925 
1926 int
1927 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
1928 {
1929 
1930 	return (_fgetvp(td, fd, vpp, FWRITE));
1931 }
1932 
1933 /*
1934  * Like fget() but loads the underlying socket, or returns an error if
1935  * the descriptor does not represent a socket.
1936  *
1937  * We bump the ref count on the returned socket.  XXX Also obtain the SX
1938  * lock in the future.
1939  */
1940 int
1941 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
1942 {
1943 	struct file *fp;
1944 	int error;
1945 
1946 	*spp = NULL;
1947 	if (fflagp != NULL)
1948 		*fflagp = 0;
1949 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
1950 		return (error);
1951 	if (fp->f_type != DTYPE_SOCKET) {
1952 		error = ENOTSOCK;
1953 	} else {
1954 		*spp = fp->f_data;
1955 		if (fflagp)
1956 			*fflagp = fp->f_flag;
1957 		soref(*spp);
1958 	}
1959 	FILEDESC_UNLOCK(td->td_proc->p_fd);
1960 	return (error);
1961 }
1962 
1963 /*
1964  * Drop the reference count on the the socket and XXX release the SX lock in
1965  * the future.  The last reference closes the socket.
1966  */
1967 void
1968 fputsock(struct socket *so)
1969 {
1970 
1971 	sorele(so);
1972 }
1973 
1974 /*
1975  * Drop reference on struct file passed in, may call closef if the
1976  * reference hits zero.
1977  * Expects struct file locked, and will unlock it.
1978  */
1979 int
1980 fdrop_locked(fp, td)
1981 	struct file *fp;
1982 	struct thread *td;
1983 {
1984 	struct flock lf;
1985 	struct vnode *vp;
1986 	int error;
1987 
1988 	FILE_LOCK_ASSERT(fp, MA_OWNED);
1989 
1990 	if (--fp->f_count > 0) {
1991 		FILE_UNLOCK(fp);
1992 		return (0);
1993 	}
1994 	/* We have the last ref so we can proceed without the file lock. */
1995 	FILE_UNLOCK(fp);
1996 	mtx_lock(&Giant);
1997 	if (fp->f_count < 0)
1998 		panic("fdrop: count < 0");
1999 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
2000 		lf.l_whence = SEEK_SET;
2001 		lf.l_start = 0;
2002 		lf.l_len = 0;
2003 		lf.l_type = F_UNLCK;
2004 		vp = fp->f_vnode;
2005 		(void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2006 	}
2007 	if (fp->f_ops != &badfileops)
2008 		error = fo_close(fp, td);
2009 	else
2010 		error = 0;
2011 	ffree(fp);
2012 	mtx_unlock(&Giant);
2013 	return (error);
2014 }
2015 
2016 /*
2017  * Apply an advisory lock on a file descriptor.
2018  *
2019  * Just attempt to get a record lock of the requested type on
2020  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2021  */
2022 #ifndef _SYS_SYSPROTO_H_
2023 struct flock_args {
2024 	int	fd;
2025 	int	how;
2026 };
2027 #endif
2028 /*
2029  * MPSAFE
2030  */
2031 /* ARGSUSED */
2032 int
2033 flock(td, uap)
2034 	struct thread *td;
2035 	struct flock_args *uap;
2036 {
2037 	struct file *fp;
2038 	struct vnode *vp;
2039 	struct flock lf;
2040 	int error;
2041 
2042 	if ((error = fget(td, uap->fd, &fp)) != 0)
2043 		return (error);
2044 	if (fp->f_type != DTYPE_VNODE) {
2045 		fdrop(fp, td);
2046 		return (EOPNOTSUPP);
2047 	}
2048 
2049 	mtx_lock(&Giant);
2050 	vp = fp->f_vnode;
2051 	lf.l_whence = SEEK_SET;
2052 	lf.l_start = 0;
2053 	lf.l_len = 0;
2054 	if (uap->how & LOCK_UN) {
2055 		lf.l_type = F_UNLCK;
2056 		FILE_LOCK(fp);
2057 		fp->f_flag &= ~FHASLOCK;
2058 		FILE_UNLOCK(fp);
2059 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2060 		goto done2;
2061 	}
2062 	if (uap->how & LOCK_EX)
2063 		lf.l_type = F_WRLCK;
2064 	else if (uap->how & LOCK_SH)
2065 		lf.l_type = F_RDLCK;
2066 	else {
2067 		error = EBADF;
2068 		goto done2;
2069 	}
2070 	FILE_LOCK(fp);
2071 	fp->f_flag |= FHASLOCK;
2072 	FILE_UNLOCK(fp);
2073 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2074 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2075 done2:
2076 	fdrop(fp, td);
2077 	mtx_unlock(&Giant);
2078 	return (error);
2079 }
2080 
2081 /*
2082  * File Descriptor pseudo-device driver (/dev/fd/).
2083  *
2084  * Opening minor device N dup()s the file (if any) connected to file
2085  * descriptor N belonging to the calling process.  Note that this driver
2086  * consists of only the ``open()'' routine, because all subsequent
2087  * references to this file will be direct to the other driver.
2088  */
2089 /* ARGSUSED */
2090 static int
2091 fdopen(dev, mode, type, td)
2092 	dev_t dev;
2093 	int mode, type;
2094 	struct thread *td;
2095 {
2096 
2097 	/*
2098 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2099 	 * the file descriptor being sought for duplication. The error
2100 	 * return ensures that the vnode for this device will be released
2101 	 * by vn_open. Open will detect this special error and take the
2102 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2103 	 * will simply report the error.
2104 	 */
2105 	td->td_dupfd = dev2unit(dev);
2106 	return (ENODEV);
2107 }
2108 
2109 /*
2110  * Duplicate the specified descriptor to a free descriptor.
2111  */
2112 int
2113 dupfdopen(td, fdp, indx, dfd, mode, error)
2114 	struct thread *td;
2115 	struct filedesc *fdp;
2116 	int indx, dfd;
2117 	int mode;
2118 	int error;
2119 {
2120 	struct file *wfp;
2121 	struct file *fp;
2122 
2123 	/*
2124 	 * If the to-be-dup'd fd number is greater than the allowed number
2125 	 * of file descriptors, or the fd to be dup'd has already been
2126 	 * closed, then reject.
2127 	 */
2128 	FILEDESC_LOCK(fdp);
2129 	if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2130 	    (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2131 		FILEDESC_UNLOCK(fdp);
2132 		return (EBADF);
2133 	}
2134 
2135 	/*
2136 	 * There are two cases of interest here.
2137 	 *
2138 	 * For ENODEV simply dup (dfd) to file descriptor
2139 	 * (indx) and return.
2140 	 *
2141 	 * For ENXIO steal away the file structure from (dfd) and
2142 	 * store it in (indx).  (dfd) is effectively closed by
2143 	 * this operation.
2144 	 *
2145 	 * Any other error code is just returned.
2146 	 */
2147 	switch (error) {
2148 	case ENODEV:
2149 		/*
2150 		 * Check that the mode the file is being opened for is a
2151 		 * subset of the mode of the existing descriptor.
2152 		 */
2153 		FILE_LOCK(wfp);
2154 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2155 			FILE_UNLOCK(wfp);
2156 			FILEDESC_UNLOCK(fdp);
2157 			return (EACCES);
2158 		}
2159 		fp = fdp->fd_ofiles[indx];
2160 #if 0
2161 		if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED)
2162 			(void) munmapfd(td, indx);
2163 #endif
2164 		fdp->fd_ofiles[indx] = wfp;
2165 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2166 		fhold_locked(wfp);
2167 		FILE_UNLOCK(wfp);
2168 		if (indx > fdp->fd_lastfile)
2169 			fdp->fd_lastfile = indx;
2170 		if (fp != NULL)
2171 			FILE_LOCK(fp);
2172 		FILEDESC_UNLOCK(fdp);
2173 		/*
2174 		 * We now own the reference to fp that the ofiles[] array
2175 		 * used to own.  Release it.
2176 		 */
2177 		if (fp != NULL)
2178 			fdrop_locked(fp, td);
2179 		return (0);
2180 
2181 	case ENXIO:
2182 		/*
2183 		 * Steal away the file pointer from dfd and stuff it into indx.
2184 		 */
2185 		fp = fdp->fd_ofiles[indx];
2186 #if 0
2187 		if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED)
2188 			(void) munmapfd(td, indx);
2189 #endif
2190 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2191 		fdp->fd_ofiles[dfd] = NULL;
2192 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2193 		fdp->fd_ofileflags[dfd] = 0;
2194 
2195 		/*
2196 		 * Complete the clean up of the filedesc structure by
2197 		 * recomputing the various hints.
2198 		 */
2199 		if (indx > fdp->fd_lastfile) {
2200 			fdp->fd_lastfile = indx;
2201 		} else {
2202 			while (fdp->fd_lastfile > 0 &&
2203 			   fdp->fd_ofiles[fdp->fd_lastfile] == NULL) {
2204 				fdp->fd_lastfile--;
2205 			}
2206 			if (dfd < fdp->fd_freefile)
2207 				fdp->fd_freefile = dfd;
2208 		}
2209 		if (fp != NULL)
2210 			FILE_LOCK(fp);
2211 		FILEDESC_UNLOCK(fdp);
2212 
2213 		/*
2214 		 * we now own the reference to fp that the ofiles[] array
2215 		 * used to own.  Release it.
2216 		 */
2217 		if (fp != NULL)
2218 			fdrop_locked(fp, td);
2219 		return (0);
2220 
2221 	default:
2222 		FILEDESC_UNLOCK(fdp);
2223 		return (error);
2224 	}
2225 	/* NOTREACHED */
2226 }
2227 
2228 
2229 struct filedesc_to_leader *
2230 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2231 			 struct filedesc *fdp,
2232 			 struct proc *leader)
2233 {
2234 	struct filedesc_to_leader *fdtol;
2235 
2236 	MALLOC(fdtol, struct filedesc_to_leader *,
2237 	       sizeof(struct filedesc_to_leader),
2238 	       M_FILEDESC_TO_LEADER,
2239 	       M_WAITOK);
2240 	fdtol->fdl_refcount = 1;
2241 	fdtol->fdl_holdcount = 0;
2242 	fdtol->fdl_wakeup = 0;
2243 	fdtol->fdl_leader = leader;
2244 	if (old != NULL) {
2245 		FILEDESC_LOCK(fdp);
2246 		fdtol->fdl_next = old->fdl_next;
2247 		fdtol->fdl_prev = old;
2248 		old->fdl_next = fdtol;
2249 		fdtol->fdl_next->fdl_prev = fdtol;
2250 		FILEDESC_UNLOCK(fdp);
2251 	} else {
2252 		fdtol->fdl_next = fdtol;
2253 		fdtol->fdl_prev = fdtol;
2254 	}
2255 	return fdtol;
2256 }
2257 
2258 /*
2259  * Get file structures.
2260  */
2261 static int
2262 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2263 {
2264 	struct xfile xf;
2265 	struct filedesc *fdp;
2266 	struct file *fp;
2267 	struct proc *p;
2268 	int error, n;
2269 
2270 	/*
2271 	 * Note: because the number of file descriptors is calculated
2272 	 * in different ways for sizing vs returning the data,
2273 	 * there is information leakage from the first loop.  However,
2274 	 * it is of a similar order of magnitude to the leakage from
2275 	 * global system statistics such as kern.openfiles.
2276 	 */
2277 	sysctl_wire_old_buffer(req, 0);
2278 	if (req->oldptr == NULL) {
2279 		n = 16;		/* A slight overestimate. */
2280 		sx_slock(&filelist_lock);
2281 		LIST_FOREACH(fp, &filehead, f_list) {
2282 			/*
2283 			 * We should grab the lock, but this is an
2284 			 * estimate, so does it really matter?
2285 			 */
2286 			/* mtx_lock(fp->f_mtxp); */
2287 			n += fp->f_count;
2288 			/* mtx_unlock(f->f_mtxp); */
2289 		}
2290 		sx_sunlock(&filelist_lock);
2291 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2292 	}
2293 	error = 0;
2294 	bzero(&xf, sizeof(xf));
2295 	xf.xf_size = sizeof(xf);
2296 	sx_slock(&allproc_lock);
2297 	LIST_FOREACH(p, &allproc, p_list) {
2298 		PROC_LOCK(p);
2299 		if (p_cansee(req->td, p) != 0) {
2300 			PROC_UNLOCK(p);
2301 			continue;
2302 		}
2303 		xf.xf_pid = p->p_pid;
2304 		xf.xf_uid = p->p_ucred->cr_uid;
2305 		PROC_UNLOCK(p);
2306 		mtx_lock(&fdesc_mtx);
2307 		if ((fdp = p->p_fd) == NULL) {
2308 			mtx_unlock(&fdesc_mtx);
2309 			continue;
2310 		}
2311 		FILEDESC_LOCK(fdp);
2312 		for (n = 0; n < fdp->fd_nfiles; ++n) {
2313 			if ((fp = fdp->fd_ofiles[n]) == NULL)
2314 				continue;
2315 			xf.xf_fd = n;
2316 			xf.xf_file = fp;
2317 			xf.xf_data = fp->f_data;
2318 			xf.xf_type = fp->f_type;
2319 			xf.xf_count = fp->f_count;
2320 			xf.xf_msgcount = fp->f_msgcount;
2321 			xf.xf_offset = fp->f_offset;
2322 			xf.xf_flag = fp->f_flag;
2323 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2324 			if (error)
2325 				break;
2326 		}
2327 		FILEDESC_UNLOCK(fdp);
2328 		mtx_unlock(&fdesc_mtx);
2329 		if (error)
2330 			break;
2331 	}
2332 	sx_sunlock(&allproc_lock);
2333 	return (error);
2334 }
2335 
2336 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2337     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2338 
2339 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2340     &maxfilesperproc, 0, "Maximum files allowed open per process");
2341 
2342 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2343     &maxfiles, 0, "Maximum number of files");
2344 
2345 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2346     &nfiles, 0, "System-wide number of open files");
2347 
2348 static void
2349 fildesc_drvinit(void *unused)
2350 {
2351 	dev_t dev;
2352 
2353 	dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2354 	make_dev_alias(dev, "stdin");
2355 	dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2356 	make_dev_alias(dev, "stdout");
2357 	dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2358 	make_dev_alias(dev, "stderr");
2359 }
2360 
2361 static fo_rdwr_t	badfo_readwrite;
2362 static fo_ioctl_t	badfo_ioctl;
2363 static fo_poll_t	badfo_poll;
2364 static fo_kqfilter_t	badfo_kqfilter;
2365 static fo_stat_t	badfo_stat;
2366 static fo_close_t	badfo_close;
2367 
2368 struct fileops badfileops = {
2369 	.fo_read = badfo_readwrite,
2370 	.fo_write = badfo_readwrite,
2371 	.fo_ioctl = badfo_ioctl,
2372 	.fo_poll = badfo_poll,
2373 	.fo_kqfilter = badfo_kqfilter,
2374 	.fo_stat = badfo_stat,
2375 	.fo_close = badfo_close,
2376 };
2377 
2378 static int
2379 badfo_readwrite(fp, uio, active_cred, flags, td)
2380 	struct file *fp;
2381 	struct uio *uio;
2382 	struct ucred *active_cred;
2383 	struct thread *td;
2384 	int flags;
2385 {
2386 
2387 	return (EBADF);
2388 }
2389 
2390 static int
2391 badfo_ioctl(fp, com, data, active_cred, td)
2392 	struct file *fp;
2393 	u_long com;
2394 	void *data;
2395 	struct ucred *active_cred;
2396 	struct thread *td;
2397 {
2398 
2399 	return (EBADF);
2400 }
2401 
2402 static int
2403 badfo_poll(fp, events, active_cred, td)
2404 	struct file *fp;
2405 	int events;
2406 	struct ucred *active_cred;
2407 	struct thread *td;
2408 {
2409 
2410 	return (0);
2411 }
2412 
2413 static int
2414 badfo_kqfilter(fp, kn)
2415 	struct file *fp;
2416 	struct knote *kn;
2417 {
2418 
2419 	return (0);
2420 }
2421 
2422 static int
2423 badfo_stat(fp, sb, active_cred, td)
2424 	struct file *fp;
2425 	struct stat *sb;
2426 	struct ucred *active_cred;
2427 	struct thread *td;
2428 {
2429 
2430 	return (EBADF);
2431 }
2432 
2433 static int
2434 badfo_close(fp, td)
2435 	struct file *fp;
2436 	struct thread *td;
2437 {
2438 
2439 	return (EBADF);
2440 }
2441 
2442 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2443 					fildesc_drvinit,NULL)
2444 
2445 static void filelistinit(void *);
2446 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2447 
2448 /* ARGSUSED*/
2449 static void
2450 filelistinit(dummy)
2451 	void *dummy;
2452 {
2453 
2454 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2455 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2456 	sx_init(&filelist_lock, "filelist lock");
2457 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2458 }
2459