__P() removal
[dragonfly.git] / sys / kern / kern_descrip.c
CommitLineData
984263bc
MD
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 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.17 2003/06/06 20:21:32 tegge Exp $
402ed7e1 40 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.12 2003/08/26 21:09:02 rob Exp $
984263bc
MD
41 */
42
43#include "opt_compat.h"
44#include <sys/param.h>
45#include <sys/systm.h>
46#include <sys/malloc.h>
47#include <sys/sysproto.h>
48#include <sys/conf.h>
49#include <sys/filedesc.h>
50#include <sys/kernel.h>
51#include <sys/sysctl.h>
52#include <sys/vnode.h>
53#include <sys/proc.h>
54#include <sys/namei.h>
55#include <sys/file.h>
56#include <sys/stat.h>
57#include <sys/filio.h>
58#include <sys/fcntl.h>
59#include <sys/unistd.h>
60#include <sys/resourcevar.h>
61#include <sys/event.h>
62
63#include <vm/vm.h>
64#include <vm/vm_extern.h>
65
dadab5e9
MD
66#include <sys/file2.h>
67
984263bc
MD
68static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
69static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
70 "file desc to leader structures");
71MALLOC_DEFINE(M_FILE, "file", "Open file structure");
72static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
73
74static d_open_t fdopen;
75#define NUMFDESC 64
76
77#define CDEV_MAJOR 22
78static struct cdevsw fildesc_cdevsw = {
fabb8ceb
MD
79 /* name */ "FD",
80 /* maj */ CDEV_MAJOR,
81 /* flags */ 0,
82 /* port */ NULL,
83 /* autoq */ 0,
84
984263bc
MD
85 /* open */ fdopen,
86 /* close */ noclose,
87 /* read */ noread,
88 /* write */ nowrite,
89 /* ioctl */ noioctl,
90 /* poll */ nopoll,
91 /* mmap */ nommap,
92 /* strategy */ nostrategy,
984263bc 93 /* dump */ nodump,
fabb8ceb 94 /* psize */ nopsize
984263bc
MD
95};
96
402ed7e1
RG
97static int do_dup (struct filedesc *fdp, int old, int new, register_t *retval, struct proc *p);
98static int badfo_readwrite (struct file *fp, struct uio *uio,
99 struct ucred *cred, int flags, struct thread *td);
100static int badfo_ioctl (struct file *fp, u_long com, caddr_t data,
101 struct thread *td);
102static int badfo_poll (struct file *fp, int events,
103 struct ucred *cred, struct thread *td);
104static int badfo_kqfilter (struct file *fp, struct knote *kn);
105static int badfo_stat (struct file *fp, struct stat *sb, struct thread *td);
106static int badfo_close (struct file *fp, struct thread *td);
984263bc
MD
107
108/*
109 * Descriptor management.
110 */
111struct filelist filehead; /* head of list of open files */
112int nfiles; /* actual number of open files */
113extern int cmask;
114
115/*
116 * System calls on descriptors.
117 */
984263bc
MD
118/* ARGSUSED */
119int
41c20dac 120getdtablesize(struct getdtablesize_args *uap)
984263bc 121{
41c20dac 122 struct proc *p = curproc;
984263bc 123
c7114eea 124 uap->sysmsg_result =
984263bc
MD
125 min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
126 return (0);
127}
128
129/*
130 * Duplicate a file descriptor to a particular value.
131 *
132 * note: keep in mind that a potential race condition exists when closing
133 * descriptors from a shared descriptor table (via rfork).
134 */
984263bc
MD
135/* ARGSUSED */
136int
41c20dac 137dup2(struct dup2_args *uap)
984263bc 138{
41c20dac
MD
139 struct proc *p = curproc;
140 struct filedesc *fdp = p->p_fd;
141 u_int old = uap->from, new = uap->to;
984263bc
MD
142 int i, error;
143
144retry:
145 if (old >= fdp->fd_nfiles ||
146 fdp->fd_ofiles[old] == NULL ||
147 new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
148 new >= maxfilesperproc) {
149 return (EBADF);
150 }
151 if (old == new) {
c7114eea 152 uap->sysmsg_result = new;
984263bc
MD
153 return (0);
154 }
155 if (new >= fdp->fd_nfiles) {
156 if ((error = fdalloc(p, new, &i)))
157 return (error);
158 /*
159 * fdalloc() may block, retest everything.
160 */
161 goto retry;
162 }
c7114eea 163 return (do_dup(fdp, (int)old, (int)new, uap->sysmsg_fds, p));
984263bc
MD
164}
165
166/*
167 * Duplicate a file descriptor.
168 */
984263bc
MD
169/* ARGSUSED */
170int
41c20dac 171dup(struct dup_args *uap)
984263bc 172{
41c20dac
MD
173 struct proc *p = curproc;
174 struct filedesc *fdp;
984263bc
MD
175 u_int old;
176 int new, error;
177
178 old = uap->fd;
179 fdp = p->p_fd;
180 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL)
181 return (EBADF);
182 if ((error = fdalloc(p, 0, &new)))
183 return (error);
c7114eea 184 return (do_dup(fdp, (int)old, new, uap->sysmsg_fds, p));
984263bc
MD
185}
186
187/*
188 * The file control system call.
189 */
984263bc
MD
190/* ARGSUSED */
191int
41c20dac 192fcntl(struct fcntl_args *uap)
984263bc 193{
dadab5e9
MD
194 struct thread *td = curthread;
195 struct proc *p = td->td_proc;
41c20dac
MD
196 struct filedesc *fdp = p->p_fd;
197 struct file *fp;
198 char *pop;
984263bc
MD
199 struct vnode *vp;
200 int i, tmp, error, flg = F_POSIX;
201 struct flock fl;
202 u_int newmin;
203
dadab5e9
MD
204 KKASSERT(p);
205
984263bc
MD
206 if ((unsigned)uap->fd >= fdp->fd_nfiles ||
207 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
208 return (EBADF);
209 pop = &fdp->fd_ofileflags[uap->fd];
210
211 switch (uap->cmd) {
212 case F_DUPFD:
213 newmin = uap->arg;
214 if (newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
215 newmin >= maxfilesperproc)
216 return (EINVAL);
217 if ((error = fdalloc(p, newmin, &i)))
218 return (error);
c7114eea 219 return (do_dup(fdp, uap->fd, i, uap->sysmsg_fds, p));
984263bc
MD
220
221 case F_GETFD:
c7114eea 222 uap->sysmsg_result = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
984263bc
MD
223 return (0);
224
225 case F_SETFD:
226 *pop = (*pop &~ UF_EXCLOSE) |
227 (uap->arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
228 return (0);
229
230 case F_GETFL:
c7114eea 231 uap->sysmsg_result = OFLAGS(fp->f_flag);
984263bc
MD
232 return (0);
233
234 case F_SETFL:
235 fhold(fp);
236 fp->f_flag &= ~FCNTLFLAGS;
237 fp->f_flag |= FFLAGS(uap->arg & ~O_ACCMODE) & FCNTLFLAGS;
238 tmp = fp->f_flag & FNONBLOCK;
dadab5e9 239 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td);
984263bc 240 if (error) {
dadab5e9 241 fdrop(fp, td);
984263bc
MD
242 return (error);
243 }
244 tmp = fp->f_flag & FASYNC;
dadab5e9 245 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, td);
984263bc 246 if (!error) {
dadab5e9 247 fdrop(fp, td);
984263bc
MD
248 return (0);
249 }
250 fp->f_flag &= ~FNONBLOCK;
251 tmp = 0;
dadab5e9
MD
252 (void)fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, td);
253 fdrop(fp, td);
984263bc
MD
254 return (error);
255
256 case F_GETOWN:
257 fhold(fp);
c7114eea 258 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)uap->sysmsg_fds, td);
dadab5e9 259 fdrop(fp, td);
984263bc
MD
260 return(error);
261
262 case F_SETOWN:
263 fhold(fp);
dadab5e9
MD
264 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&uap->arg, td);
265 fdrop(fp, td);
984263bc
MD
266 return(error);
267
268 case F_SETLKW:
269 flg |= F_WAIT;
270 /* Fall into F_SETLK */
271
272 case F_SETLK:
273 if (fp->f_type != DTYPE_VNODE)
274 return (EBADF);
275 vp = (struct vnode *)fp->f_data;
276
277 /*
278 * copyin/lockop may block
279 */
280 fhold(fp);
281 /* Copy in the lock structure */
282 error = copyin((caddr_t)(intptr_t)uap->arg, (caddr_t)&fl,
283 sizeof(fl));
284 if (error) {
dadab5e9 285 fdrop(fp, td);
984263bc
MD
286 return (error);
287 }
288 if (fl.l_whence == SEEK_CUR)
289 fl.l_start += fp->f_offset;
290
291 switch (fl.l_type) {
292 case F_RDLCK:
293 if ((fp->f_flag & FREAD) == 0) {
294 error = EBADF;
295 break;
296 }
297 p->p_leader->p_flag |= P_ADVLOCK;
298 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
299 &fl, flg);
300 break;
301 case F_WRLCK:
302 if ((fp->f_flag & FWRITE) == 0) {
303 error = EBADF;
304 break;
305 }
306 p->p_leader->p_flag |= P_ADVLOCK;
307 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
308 &fl, flg);
309 break;
310 case F_UNLCK:
311 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
312 &fl, F_POSIX);
313 break;
314 default:
315 error = EINVAL;
316 break;
317 }
318 /* Check for race with close */
319 if ((unsigned) uap->fd >= fdp->fd_nfiles ||
320 fp != fdp->fd_ofiles[uap->fd]) {
321 fl.l_whence = SEEK_SET;
322 fl.l_start = 0;
323 fl.l_len = 0;
324 fl.l_type = F_UNLCK;
325 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
326 F_UNLCK, &fl, F_POSIX);
327 }
dadab5e9 328 fdrop(fp, td);
984263bc
MD
329 return(error);
330
331 case F_GETLK:
332 if (fp->f_type != DTYPE_VNODE)
333 return (EBADF);
334 vp = (struct vnode *)fp->f_data;
335 /*
336 * copyin/lockop may block
337 */
338 fhold(fp);
339 /* Copy in the lock structure */
340 error = copyin((caddr_t)(intptr_t)uap->arg, (caddr_t)&fl,
341 sizeof(fl));
342 if (error) {
dadab5e9 343 fdrop(fp, td);
984263bc
MD
344 return (error);
345 }
346 if (fl.l_type != F_RDLCK && fl.l_type != F_WRLCK &&
347 fl.l_type != F_UNLCK) {
dadab5e9 348 fdrop(fp, td);
984263bc
MD
349 return (EINVAL);
350 }
351 if (fl.l_whence == SEEK_CUR)
352 fl.l_start += fp->f_offset;
353 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
354 &fl, F_POSIX);
dadab5e9 355 fdrop(fp, td);
984263bc
MD
356 if (error == 0) {
357 error = copyout((caddr_t)&fl,
358 (caddr_t)(intptr_t)uap->arg, sizeof(fl));
359 }
360 return(error);
361 default:
362 return (EINVAL);
363 }
364 /* NOTREACHED */
365}
366
367/*
368 * Common code for dup, dup2, and fcntl(F_DUPFD).
369 */
370static int
371do_dup(fdp, old, new, retval, p)
41c20dac
MD
372 struct filedesc *fdp;
373 int old, new;
984263bc
MD
374 register_t *retval;
375 struct proc *p;
376{
dadab5e9 377 struct thread *td = p->p_thread;
984263bc
MD
378 struct file *fp;
379 struct file *delfp;
380 int holdleaders;
381
382 /*
383 * Save info on the descriptor being overwritten. We have
384 * to do the unmap now, but we cannot close it without
385 * introducing an ownership race for the slot.
386 */
387 delfp = fdp->fd_ofiles[new];
388 if (delfp != NULL && p->p_fdtol != NULL) {
389 /*
390 * Ask fdfree() to sleep to ensure that all relevant
391 * process leaders can be traversed in closef().
392 */
393 fdp->fd_holdleaderscount++;
394 holdleaders = 1;
395 } else
396 holdleaders = 0;
397#if 0
398 if (delfp && (fdp->fd_ofileflags[new] & UF_MAPPED))
399 (void) munmapfd(p, new);
400#endif
401
402 /*
403 * Duplicate the source descriptor, update lastfile
404 */
405 fp = fdp->fd_ofiles[old];
406 fdp->fd_ofiles[new] = fp;
407 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
408 fhold(fp);
409 if (new > fdp->fd_lastfile)
410 fdp->fd_lastfile = new;
411 *retval = new;
412
413 /*
414 * If we dup'd over a valid file, we now own the reference to it
415 * and must dispose of it using closef() semantics (as if a
416 * close() were performed on it).
417 */
418 if (delfp) {
dadab5e9 419 (void) closef(delfp, td);
984263bc
MD
420 if (holdleaders) {
421 fdp->fd_holdleaderscount--;
422 if (fdp->fd_holdleaderscount == 0 &&
423 fdp->fd_holdleaderswakeup != 0) {
424 fdp->fd_holdleaderswakeup = 0;
425 wakeup(&fdp->fd_holdleaderscount);
426 }
427 }
428 }
429 return (0);
430}
431
432/*
433 * If sigio is on the list associated with a process or process group,
434 * disable signalling from the device, remove sigio from the list and
435 * free sigio.
436 */
437void
438funsetown(sigio)
439 struct sigio *sigio;
440{
441 int s;
442
443 if (sigio == NULL)
444 return;
445 s = splhigh();
446 *(sigio->sio_myref) = NULL;
447 splx(s);
448 if (sigio->sio_pgid < 0) {
449 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
450 sigio, sio_pgsigio);
451 } else /* if ((*sigiop)->sio_pgid > 0) */ {
452 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
453 sigio, sio_pgsigio);
454 }
455 crfree(sigio->sio_ucred);
456 FREE(sigio, M_SIGIO);
457}
458
459/* Free a list of sigio structures. */
460void
461funsetownlst(sigiolst)
462 struct sigiolst *sigiolst;
463{
464 struct sigio *sigio;
465
466 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
467 funsetown(sigio);
468}
469
470/*
471 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
472 *
473 * After permission checking, add a sigio structure to the sigio list for
474 * the process or process group.
475 */
476int
477fsetown(pgid, sigiop)
478 pid_t pgid;
479 struct sigio **sigiop;
480{
481 struct proc *proc;
482 struct pgrp *pgrp;
483 struct sigio *sigio;
484 int s;
485
486 if (pgid == 0) {
487 funsetown(*sigiop);
488 return (0);
489 }
490 if (pgid > 0) {
491 proc = pfind(pgid);
492 if (proc == NULL)
493 return (ESRCH);
494
495 /*
496 * Policy - Don't allow a process to FSETOWN a process
497 * in another session.
498 *
499 * Remove this test to allow maximum flexibility or
500 * restrict FSETOWN to the current process or process
501 * group for maximum safety.
502 */
503 if (proc->p_session != curproc->p_session)
504 return (EPERM);
505
506 pgrp = NULL;
507 } else /* if (pgid < 0) */ {
508 pgrp = pgfind(-pgid);
509 if (pgrp == NULL)
510 return (ESRCH);
511
512 /*
513 * Policy - Don't allow a process to FSETOWN a process
514 * in another session.
515 *
516 * Remove this test to allow maximum flexibility or
517 * restrict FSETOWN to the current process or process
518 * group for maximum safety.
519 */
520 if (pgrp->pg_session != curproc->p_session)
521 return (EPERM);
522
523 proc = NULL;
524 }
525 funsetown(*sigiop);
526 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
527 if (pgid > 0) {
528 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
529 sigio->sio_proc = proc;
530 } else {
531 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
532 sigio->sio_pgrp = pgrp;
533 }
534 sigio->sio_pgid = pgid;
e9a372eb 535 sigio->sio_ucred = crhold(curproc->p_ucred);
984263bc 536 /* It would be convenient if p_ruid was in ucred. */
41c20dac 537 sigio->sio_ruid = curproc->p_ucred->cr_ruid;
984263bc
MD
538 sigio->sio_myref = sigiop;
539 s = splhigh();
540 *sigiop = sigio;
541 splx(s);
542 return (0);
543}
544
545/*
546 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
547 */
548pid_t
549fgetown(sigio)
550 struct sigio *sigio;
551{
552 return (sigio != NULL ? sigio->sio_pgid : 0);
553}
554
555/*
556 * Close a file descriptor.
557 */
984263bc
MD
558/* ARGSUSED */
559int
41c20dac 560close(struct close_args *uap)
984263bc 561{
dadab5e9
MD
562 struct thread *td = curthread;
563 struct proc *p = td->td_proc;
564 struct filedesc *fdp;
41c20dac
MD
565 struct file *fp;
566 int fd = uap->fd;
984263bc
MD
567 int error;
568 int holdleaders;
569
dadab5e9
MD
570 KKASSERT(p);
571 fdp = p->p_fd;
572
984263bc
MD
573 if ((unsigned)fd >= fdp->fd_nfiles ||
574 (fp = fdp->fd_ofiles[fd]) == NULL)
575 return (EBADF);
576#if 0
577 if (fdp->fd_ofileflags[fd] & UF_MAPPED)
578 (void) munmapfd(p, fd);
579#endif
580 fdp->fd_ofiles[fd] = NULL;
581 fdp->fd_ofileflags[fd] = 0;
582 holdleaders = 0;
583 if (p->p_fdtol != NULL) {
584 /*
585 * Ask fdfree() to sleep to ensure that all relevant
586 * process leaders can be traversed in closef().
587 */
588 fdp->fd_holdleaderscount++;
589 holdleaders = 1;
590 }
591
592 /*
593 * we now hold the fp reference that used to be owned by the descriptor
594 * array.
595 */
596 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
597 fdp->fd_lastfile--;
598 if (fd < fdp->fd_freefile)
599 fdp->fd_freefile = fd;
600 if (fd < fdp->fd_knlistsize)
601 knote_fdclose(p, fd);
dadab5e9 602 error = closef(fp, td);
984263bc
MD
603 if (holdleaders) {
604 fdp->fd_holdleaderscount--;
605 if (fdp->fd_holdleaderscount == 0 &&
606 fdp->fd_holdleaderswakeup != 0) {
607 fdp->fd_holdleaderswakeup = 0;
608 wakeup(&fdp->fd_holdleaderscount);
609 }
610 }
611 return (error);
612}
613
614#if defined(COMPAT_43) || defined(COMPAT_SUNOS)
615/*
616 * Return status information about a file descriptor.
617 */
984263bc
MD
618/* ARGSUSED */
619int
41c20dac 620ofstat(struct ofstat_args *uap)
984263bc 621{
dadab5e9
MD
622 struct thread *td = curthread;
623 struct proc *p = td->td_proc;
624 struct filedesc *fdp;
41c20dac 625 struct file *fp;
984263bc
MD
626 struct stat ub;
627 struct ostat oub;
628 int error;
629
dadab5e9
MD
630 KKASSERT(p);
631 fdp = p->p_fd;
984263bc
MD
632 if ((unsigned)uap->fd >= fdp->fd_nfiles ||
633 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
634 return (EBADF);
635 fhold(fp);
dadab5e9 636 error = fo_stat(fp, &ub, td);
984263bc
MD
637 if (error == 0) {
638 cvtstat(&ub, &oub);
639 error = copyout((caddr_t)&oub, (caddr_t)uap->sb, sizeof (oub));
640 }
dadab5e9 641 fdrop(fp, td);
984263bc
MD
642 return (error);
643}
644#endif /* COMPAT_43 || COMPAT_SUNOS */
645
646/*
647 * Return status information about a file descriptor.
648 */
984263bc
MD
649/* ARGSUSED */
650int
41c20dac 651fstat(struct fstat_args *uap)
984263bc 652{
dadab5e9
MD
653 struct thread *td = curthread;
654 struct proc *p = td->td_proc;
41c20dac
MD
655 struct filedesc *fdp = p->p_fd;
656 struct file *fp;
984263bc
MD
657 struct stat ub;
658 int error;
659
dadab5e9
MD
660 KKASSERT(p);
661 fdp = p->p_fd;
984263bc
MD
662 if ((unsigned)uap->fd >= fdp->fd_nfiles ||
663 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
664 return (EBADF);
665 fhold(fp);
dadab5e9 666 error = fo_stat(fp, &ub, td);
984263bc
MD
667 if (error == 0)
668 error = copyout((caddr_t)&ub, (caddr_t)uap->sb, sizeof (ub));
dadab5e9 669 fdrop(fp, td);
984263bc
MD
670 return (error);
671}
672
673/*
674 * Return status information about a file descriptor.
675 */
984263bc
MD
676/* ARGSUSED */
677int
41c20dac 678nfstat(struct nfstat_args *uap)
984263bc 679{
dadab5e9
MD
680 struct thread *td = curthread;
681 struct proc *p = td->td_proc;
682 struct filedesc *fdp;
41c20dac 683 struct file *fp;
984263bc
MD
684 struct stat ub;
685 struct nstat nub;
686 int error;
687
dadab5e9
MD
688 KKASSERT(p);
689 fdp = p->p_fd;
984263bc
MD
690 if ((unsigned)uap->fd >= fdp->fd_nfiles ||
691 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
692 return (EBADF);
693 fhold(fp);
dadab5e9 694 error = fo_stat(fp, &ub, td);
984263bc
MD
695 if (error == 0) {
696 cvtnstat(&ub, &nub);
697 error = copyout((caddr_t)&nub, (caddr_t)uap->sb, sizeof (nub));
698 }
dadab5e9 699 fdrop(fp, td);
984263bc
MD
700 return (error);
701}
702
703/*
704 * Return pathconf information about a file descriptor.
705 */
984263bc
MD
706/* ARGSUSED */
707int
41c20dac 708fpathconf(struct fpathconf_args *uap)
984263bc 709{
dadab5e9
MD
710 struct thread *td = curthread;
711 struct proc *p = td->td_proc;
712 struct filedesc *fdp;
984263bc
MD
713 struct file *fp;
714 struct vnode *vp;
715 int error = 0;
716
dadab5e9
MD
717 KKASSERT(p);
718 fdp = p->p_fd;
984263bc
MD
719 if ((unsigned)uap->fd >= fdp->fd_nfiles ||
720 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
721 return (EBADF);
722
723 fhold(fp);
724
725 switch (fp->f_type) {
726 case DTYPE_PIPE:
727 case DTYPE_SOCKET:
728 if (uap->name != _PC_PIPE_BUF) {
729 error = EINVAL;
730 } else {
c7114eea 731 uap->sysmsg_result = PIPE_BUF;
984263bc
MD
732 error = 0;
733 }
734 break;
735 case DTYPE_FIFO:
736 case DTYPE_VNODE:
737 vp = (struct vnode *)fp->f_data;
c7114eea 738 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
984263bc
MD
739 break;
740 default:
741 error = EOPNOTSUPP;
742 break;
743 }
dadab5e9 744 fdrop(fp, td);
984263bc
MD
745 return(error);
746}
747
748/*
749 * Allocate a file descriptor for the process.
750 */
751static int fdexpand;
752SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
753
754int
755fdalloc(p, want, result)
756 struct proc *p;
757 int want;
758 int *result;
759{
41c20dac
MD
760 struct filedesc *fdp = p->p_fd;
761 int i;
984263bc
MD
762 int lim, last, nfiles;
763 struct file **newofile;
764 char *newofileflags;
765
766 /*
767 * Search for a free descriptor starting at the higher
768 * of want or fd_freefile. If that fails, consider
769 * expanding the ofile array.
770 */
771 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
772 for (;;) {
773 last = min(fdp->fd_nfiles, lim);
774 if ((i = want) < fdp->fd_freefile)
775 i = fdp->fd_freefile;
776 for (; i < last; i++) {
777 if (fdp->fd_ofiles[i] == NULL) {
778 fdp->fd_ofileflags[i] = 0;
779 if (i > fdp->fd_lastfile)
780 fdp->fd_lastfile = i;
781 if (want <= fdp->fd_freefile)
782 fdp->fd_freefile = i;
783 *result = i;
784 return (0);
785 }
786 }
787
788 /*
789 * No space in current array. Expand?
790 */
791 if (fdp->fd_nfiles >= lim)
792 return (EMFILE);
793 if (fdp->fd_nfiles < NDEXTENT)
794 nfiles = NDEXTENT;
795 else
796 nfiles = 2 * fdp->fd_nfiles;
797 MALLOC(newofile, struct file **, nfiles * OFILESIZE,
798 M_FILEDESC, M_WAITOK);
799
800 /*
801 * deal with file-table extend race that might have occured
802 * when malloc was blocked.
803 */
804 if (fdp->fd_nfiles >= nfiles) {
805 FREE(newofile, M_FILEDESC);
806 continue;
807 }
808 newofileflags = (char *) &newofile[nfiles];
809 /*
810 * Copy the existing ofile and ofileflags arrays
811 * and zero the new portion of each array.
812 */
813 bcopy(fdp->fd_ofiles, newofile,
814 (i = sizeof(struct file *) * fdp->fd_nfiles));
815 bzero((char *)newofile + i, nfiles * sizeof(struct file *) - i);
816 bcopy(fdp->fd_ofileflags, newofileflags,
817 (i = sizeof(char) * fdp->fd_nfiles));
818 bzero(newofileflags + i, nfiles * sizeof(char) - i);
819 if (fdp->fd_nfiles > NDFILE)
820 FREE(fdp->fd_ofiles, M_FILEDESC);
821 fdp->fd_ofiles = newofile;
822 fdp->fd_ofileflags = newofileflags;
823 fdp->fd_nfiles = nfiles;
824 fdexpand++;
825 }
826 return (0);
827}
828
829/*
830 * Check to see whether n user file descriptors
831 * are available to the process p.
832 */
833int
834fdavail(p, n)
835 struct proc *p;
41c20dac 836 int n;
984263bc 837{
41c20dac
MD
838 struct filedesc *fdp = p->p_fd;
839 struct file **fpp;
840 int i, lim, last;
984263bc
MD
841
842 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
843 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
844 return (1);
845
846 last = min(fdp->fd_nfiles, lim);
847 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
848 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
849 if (*fpp == NULL && --n <= 0)
850 return (1);
851 }
852 return (0);
853}
854
855/*
856 * Create a new open file structure and allocate
857 * a file decriptor for the process that refers to it.
858 */
859int
860falloc(p, resultfp, resultfd)
41c20dac 861 struct proc *p;
984263bc
MD
862 struct file **resultfp;
863 int *resultfd;
864{
41c20dac 865 struct file *fp, *fq;
984263bc
MD
866 int error, i;
867
868 if (nfiles >= maxfiles) {
869 tablefull("file");
870 return (ENFILE);
871 }
872 /*
873 * Allocate a new file descriptor.
874 * If the process has file descriptor zero open, add to the list
875 * of open files at that point, otherwise put it at the front of
876 * the list of open files.
877 */
878 nfiles++;
879 MALLOC(fp, struct file *, sizeof(struct file), M_FILE, M_WAITOK);
880 bzero(fp, sizeof(struct file));
881
882 /*
883 * wait until after malloc (which may have blocked) returns before
884 * allocating the slot, else a race might have shrunk it if we had
885 * allocated it before the malloc.
886 */
887 if ((error = fdalloc(p, 0, &i))) {
888 nfiles--;
889 FREE(fp, M_FILE);
890 return (error);
891 }
892 fp->f_count = 1;
e9a372eb 893 fp->f_cred = crhold(p->p_ucred);
984263bc
MD
894 fp->f_ops = &badfileops;
895 fp->f_seqcount = 1;
984263bc
MD
896 if ((fq = p->p_fd->fd_ofiles[0])) {
897 LIST_INSERT_AFTER(fq, fp, f_list);
898 } else {
899 LIST_INSERT_HEAD(&filehead, fp, f_list);
900 }
901 p->p_fd->fd_ofiles[i] = fp;
902 if (resultfp)
903 *resultfp = fp;
904 if (resultfd)
905 *resultfd = i;
906 return (0);
907}
908
909/*
910 * Free a file descriptor.
911 */
912void
913ffree(fp)
41c20dac 914 struct file *fp;
984263bc
MD
915{
916 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
917 LIST_REMOVE(fp, f_list);
918 crfree(fp->f_cred);
919 nfiles--;
920 FREE(fp, M_FILE);
921}
922
923/*
924 * Build a new filedesc structure.
925 */
926struct filedesc *
927fdinit(p)
928 struct proc *p;
929{
41c20dac
MD
930 struct filedesc0 *newfdp;
931 struct filedesc *fdp = p->p_fd;
984263bc
MD
932
933 MALLOC(newfdp, struct filedesc0 *, sizeof(struct filedesc0),
934 M_FILEDESC, M_WAITOK);
935 bzero(newfdp, sizeof(struct filedesc0));
936 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
937 if (newfdp->fd_fd.fd_cdir)
938 VREF(newfdp->fd_fd.fd_cdir);
939 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
940 if (newfdp->fd_fd.fd_rdir)
941 VREF(newfdp->fd_fd.fd_rdir);
942 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
943 if (newfdp->fd_fd.fd_jdir)
944 VREF(newfdp->fd_fd.fd_jdir);
945
946 /* Create the file descriptor table. */
947 newfdp->fd_fd.fd_refcnt = 1;
948 newfdp->fd_fd.fd_cmask = cmask;
949 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
950 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
951 newfdp->fd_fd.fd_nfiles = NDFILE;
952 newfdp->fd_fd.fd_knlistsize = -1;
953
954 return (&newfdp->fd_fd);
955}
956
957/*
958 * Share a filedesc structure.
959 */
960struct filedesc *
961fdshare(p)
962 struct proc *p;
963{
964 p->p_fd->fd_refcnt++;
965 return (p->p_fd);
966}
967
968/*
969 * Copy a filedesc structure.
970 */
971struct filedesc *
972fdcopy(p)
973 struct proc *p;
974{
41c20dac
MD
975 struct filedesc *newfdp, *fdp = p->p_fd;
976 struct file **fpp;
977 int i;
984263bc
MD
978
979 /* Certain daemons might not have file descriptors. */
980 if (fdp == NULL)
981 return (NULL);
982
983 MALLOC(newfdp, struct filedesc *, sizeof(struct filedesc0),
984 M_FILEDESC, M_WAITOK);
985 bcopy(fdp, newfdp, sizeof(struct filedesc));
986 if (newfdp->fd_cdir)
987 VREF(newfdp->fd_cdir);
988 if (newfdp->fd_rdir)
989 VREF(newfdp->fd_rdir);
990 if (newfdp->fd_jdir)
991 VREF(newfdp->fd_jdir);
992 newfdp->fd_refcnt = 1;
993
994 /*
995 * If the number of open files fits in the internal arrays
996 * of the open file structure, use them, otherwise allocate
997 * additional memory for the number of descriptors currently
998 * in use.
999 */
1000 if (newfdp->fd_lastfile < NDFILE) {
1001 newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
1002 newfdp->fd_ofileflags =
1003 ((struct filedesc0 *) newfdp)->fd_dfileflags;
1004 i = NDFILE;
1005 } else {
1006 /*
1007 * Compute the smallest multiple of NDEXTENT needed
1008 * for the file descriptors currently in use,
1009 * allowing the table to shrink.
1010 */
1011 i = newfdp->fd_nfiles;
1012 while (i > 2 * NDEXTENT && i > newfdp->fd_lastfile * 2)
1013 i /= 2;
1014 MALLOC(newfdp->fd_ofiles, struct file **, i * OFILESIZE,
1015 M_FILEDESC, M_WAITOK);
1016 newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
1017 }
1018 newfdp->fd_nfiles = i;
1019 bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **));
1020 bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char));
1021
1022 /*
1023 * kq descriptors cannot be copied.
1024 */
1025 if (newfdp->fd_knlistsize != -1) {
1026 fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile];
1027 for (i = newfdp->fd_lastfile; i >= 0; i--, fpp--) {
1028 if (*fpp != NULL && (*fpp)->f_type == DTYPE_KQUEUE) {
1029 *fpp = NULL;
1030 if (i < newfdp->fd_freefile)
1031 newfdp->fd_freefile = i;
1032 }
1033 if (*fpp == NULL && i == newfdp->fd_lastfile && i > 0)
1034 newfdp->fd_lastfile--;
1035 }
1036 newfdp->fd_knlist = NULL;
1037 newfdp->fd_knlistsize = -1;
1038 newfdp->fd_knhash = NULL;
1039 newfdp->fd_knhashmask = 0;
1040 }
1041
1042 fpp = newfdp->fd_ofiles;
1043 for (i = newfdp->fd_lastfile; i-- >= 0; fpp++) {
1044 if (*fpp != NULL)
1045 fhold(*fpp);
1046 }
1047 return (newfdp);
1048}
1049
1050/*
1051 * Release a filedesc structure.
1052 */
1053void
41c20dac 1054fdfree(struct proc *p)
984263bc 1055{
dadab5e9 1056 struct thread *td = p->p_thread;
41c20dac 1057 struct filedesc *fdp = p->p_fd;
984263bc 1058 struct file **fpp;
41c20dac 1059 int i;
984263bc
MD
1060 struct filedesc_to_leader *fdtol;
1061 struct file *fp;
1062 struct vnode *vp;
1063 struct flock lf;
1064
1065 /* Certain daemons might not have file descriptors. */
1066 if (fdp == NULL)
1067 return;
1068
1069 /* Check for special need to clear POSIX style locks */
1070 fdtol = p->p_fdtol;
1071 if (fdtol != NULL) {
1072 KASSERT(fdtol->fdl_refcount > 0,
1073 ("filedesc_to_refcount botch: fdl_refcount=%d",
1074 fdtol->fdl_refcount));
1075 if (fdtol->fdl_refcount == 1 &&
1076 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1077 i = 0;
1078 fpp = fdp->fd_ofiles;
1079 for (i = 0, fpp = fdp->fd_ofiles;
1080 i < fdp->fd_lastfile;
1081 i++, fpp++) {
1082 if (*fpp == NULL ||
1083 (*fpp)->f_type != DTYPE_VNODE)
1084 continue;
1085 fp = *fpp;
1086 fhold(fp);
1087 lf.l_whence = SEEK_SET;
1088 lf.l_start = 0;
1089 lf.l_len = 0;
1090 lf.l_type = F_UNLCK;
1091 vp = (struct vnode *)fp->f_data;
1092 (void) VOP_ADVLOCK(vp,
1093 (caddr_t)p->p_leader,
1094 F_UNLCK,
1095 &lf,
1096 F_POSIX);
dadab5e9 1097 fdrop(fp, p->p_thread);
984263bc
MD
1098 fpp = fdp->fd_ofiles + i;
1099 }
1100 }
1101 retry:
1102 if (fdtol->fdl_refcount == 1) {
1103 if (fdp->fd_holdleaderscount > 0 &&
1104 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1105 /*
1106 * close() or do_dup() has cleared a reference
1107 * in a shared file descriptor table.
1108 */
1109 fdp->fd_holdleaderswakeup = 1;
1110 tsleep(&fdp->fd_holdleaderscount,
377d4740 1111 0, "fdlhold", 0);
984263bc
MD
1112 goto retry;
1113 }
1114 if (fdtol->fdl_holdcount > 0) {
1115 /*
1116 * Ensure that fdtol->fdl_leader
1117 * remains valid in closef().
1118 */
1119 fdtol->fdl_wakeup = 1;
377d4740 1120 tsleep(fdtol, 0, "fdlhold", 0);
984263bc
MD
1121 goto retry;
1122 }
1123 }
1124 fdtol->fdl_refcount--;
1125 if (fdtol->fdl_refcount == 0 &&
1126 fdtol->fdl_holdcount == 0) {
1127 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1128 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1129 } else
1130 fdtol = NULL;
1131 p->p_fdtol = NULL;
1132 if (fdtol != NULL)
1133 FREE(fdtol, M_FILEDESC_TO_LEADER);
1134 }
1135 if (--fdp->fd_refcnt > 0)
1136 return;
1137 /*
1138 * we are the last reference to the structure, we can
1139 * safely assume it will not change out from under us.
1140 */
1141 fpp = fdp->fd_ofiles;
1142 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1143 if (*fpp)
dadab5e9 1144 (void) closef(*fpp, td);
984263bc
MD
1145 }
1146 if (fdp->fd_nfiles > NDFILE)
1147 FREE(fdp->fd_ofiles, M_FILEDESC);
1148 if (fdp->fd_cdir)
1149 vrele(fdp->fd_cdir);
1150 if (fdp->fd_rdir)
1151 vrele(fdp->fd_rdir);
1152 if (fdp->fd_jdir)
1153 vrele(fdp->fd_jdir);
1154 if (fdp->fd_knlist)
1155 FREE(fdp->fd_knlist, M_KQUEUE);
1156 if (fdp->fd_knhash)
1157 FREE(fdp->fd_knhash, M_KQUEUE);
1158 FREE(fdp, M_FILEDESC);
1159}
1160
1161/*
1162 * For setugid programs, we don't want to people to use that setugidness
1163 * to generate error messages which write to a file which otherwise would
1164 * otherwise be off-limits to the process.
1165 *
1166 * This is a gross hack to plug the hole. A better solution would involve
1167 * a special vop or other form of generalized access control mechanism. We
1168 * go ahead and just reject all procfs file systems accesses as dangerous.
1169 *
1170 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1171 * sufficient. We also don't for check setugidness since we know we are.
1172 */
1173static int
1174is_unsafe(struct file *fp)
1175{
1176 if (fp->f_type == DTYPE_VNODE &&
1177 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
1178 return (1);
1179 return (0);
1180}
1181
1182/*
1183 * Make this setguid thing safe, if at all possible.
1184 */
1185void
dadab5e9 1186setugidsafety(struct proc *p)
984263bc 1187{
dadab5e9 1188 struct thread *td = p->p_thread;
984263bc 1189 struct filedesc *fdp = p->p_fd;
41c20dac 1190 int i;
984263bc
MD
1191
1192 /* Certain daemons might not have file descriptors. */
1193 if (fdp == NULL)
1194 return;
1195
1196 /*
1197 * note: fdp->fd_ofiles may be reallocated out from under us while
1198 * we are blocked in a close. Be careful!
1199 */
1200 for (i = 0; i <= fdp->fd_lastfile; i++) {
1201 if (i > 2)
1202 break;
1203 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1204 struct file *fp;
1205
1206#if 0
1207 if ((fdp->fd_ofileflags[i] & UF_MAPPED) != 0)
1208 (void) munmapfd(p, i);
1209#endif
1210 if (i < fdp->fd_knlistsize)
1211 knote_fdclose(p, i);
1212 /*
1213 * NULL-out descriptor prior to close to avoid
1214 * a race while close blocks.
1215 */
1216 fp = fdp->fd_ofiles[i];
1217 fdp->fd_ofiles[i] = NULL;
1218 fdp->fd_ofileflags[i] = 0;
1219 if (i < fdp->fd_freefile)
1220 fdp->fd_freefile = i;
dadab5e9 1221 (void) closef(fp, td);
984263bc
MD
1222 }
1223 }
1224 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
1225 fdp->fd_lastfile--;
1226}
1227
1228/*
1229 * Close any files on exec?
1230 */
1231void
dadab5e9 1232fdcloseexec(struct proc *p)
984263bc 1233{
dadab5e9 1234 struct thread *td = p->p_thread;
984263bc 1235 struct filedesc *fdp = p->p_fd;
41c20dac 1236 int i;
984263bc
MD
1237
1238 /* Certain daemons might not have file descriptors. */
1239 if (fdp == NULL)
1240 return;
1241
1242 /*
1243 * We cannot cache fd_ofiles or fd_ofileflags since operations
1244 * may block and rip them out from under us.
1245 */
1246 for (i = 0; i <= fdp->fd_lastfile; i++) {
1247 if (fdp->fd_ofiles[i] != NULL &&
1248 (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1249 struct file *fp;
1250
1251#if 0
1252 if (fdp->fd_ofileflags[i] & UF_MAPPED)
1253 (void) munmapfd(p, i);
1254#endif
1255 if (i < fdp->fd_knlistsize)
1256 knote_fdclose(p, i);
1257 /*
1258 * NULL-out descriptor prior to close to avoid
1259 * a race while close blocks.
1260 */
1261 fp = fdp->fd_ofiles[i];
1262 fdp->fd_ofiles[i] = NULL;
1263 fdp->fd_ofileflags[i] = 0;
1264 if (i < fdp->fd_freefile)
1265 fdp->fd_freefile = i;
dadab5e9 1266 (void) closef(fp, td);
984263bc
MD
1267 }
1268 }
1269 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL)
1270 fdp->fd_lastfile--;
1271}
1272
1273/*
1274 * It is unsafe for set[ug]id processes to be started with file
1275 * descriptors 0..2 closed, as these descriptors are given implicit
1276 * significance in the Standard C library. fdcheckstd() will create a
1277 * descriptor referencing /dev/null for each of stdin, stdout, and
1278 * stderr that is not already open.
1279 */
1280int
dadab5e9 1281fdcheckstd(struct proc *p)
984263bc 1282{
dadab5e9
MD
1283 struct thread *td = p->p_thread;
1284 struct nameidata nd;
1285 struct filedesc *fdp;
1286 struct file *fp;
1287 register_t retval;
1288 int fd, i, error, flags, devnull;
984263bc
MD
1289
1290 fdp = p->p_fd;
1291 if (fdp == NULL)
1292 return (0);
1293 devnull = -1;
1294 error = 0;
1295 for (i = 0; i < 3; i++) {
1296 if (fdp->fd_ofiles[i] != NULL)
1297 continue;
1298 if (devnull < 0) {
1299 error = falloc(p, &fp, &fd);
1300 if (error != 0)
1301 break;
dadab5e9
MD
1302 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE,
1303 "/dev/null", td);
984263bc
MD
1304 flags = FREAD | FWRITE;
1305 error = vn_open(&nd, flags, 0);
1306 if (error != 0) {
1307 fdp->fd_ofiles[i] = NULL;
dadab5e9 1308 fdrop(fp, td);
984263bc
MD
1309 break;
1310 }
1311 NDFREE(&nd, NDF_ONLY_PNBUF);
1312 fp->f_data = (caddr_t)nd.ni_vp;
1313 fp->f_flag = flags;
1314 fp->f_ops = &vnops;
1315 fp->f_type = DTYPE_VNODE;
dadab5e9 1316 VOP_UNLOCK(nd.ni_vp, 0, td);
984263bc
MD
1317 devnull = fd;
1318 } else {
1319 error = fdalloc(p, 0, &fd);
1320 if (error != 0)
1321 break;
1322 error = do_dup(fdp, devnull, fd, &retval, p);
1323 if (error != 0)
1324 break;
1325 }
1326 }
1327 return (error);
1328}
1329
1330/*
1331 * Internal form of close.
1332 * Decrement reference count on file structure.
dadab5e9 1333 * Note: td and/or p may be NULL when closing a file
984263bc
MD
1334 * that was being passed in a message.
1335 */
1336int
dadab5e9 1337closef(struct file *fp, struct thread *td)
984263bc
MD
1338{
1339 struct vnode *vp;
1340 struct flock lf;
1341 struct filedesc_to_leader *fdtol;
dadab5e9 1342 struct proc *p;
984263bc
MD
1343
1344 if (fp == NULL)
1345 return (0);
dadab5e9
MD
1346 if (td == NULL) {
1347 td = curthread;
1348 p = NULL; /* allow no proc association */
1349 } else {
1350 p = td->td_proc; /* can also be NULL */
1351 }
984263bc
MD
1352 /*
1353 * POSIX record locking dictates that any close releases ALL
1354 * locks owned by this process. This is handled by setting
1355 * a flag in the unlock to free ONLY locks obeying POSIX
1356 * semantics, and not to free BSD-style file locks.
1357 * If the descriptor was in a message, POSIX-style locks
1358 * aren't passed with the descriptor.
1359 */
1360 if (p != NULL &&
1361 fp->f_type == DTYPE_VNODE) {
1362 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
1363 lf.l_whence = SEEK_SET;
1364 lf.l_start = 0;
1365 lf.l_len = 0;
1366 lf.l_type = F_UNLCK;
1367 vp = (struct vnode *)fp->f_data;
1368 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
1369 &lf, F_POSIX);
1370 }
1371 fdtol = p->p_fdtol;
1372 if (fdtol != NULL) {
1373 /*
1374 * Handle special case where file descriptor table
1375 * is shared between multiple process leaders.
1376 */
1377 for (fdtol = fdtol->fdl_next;
1378 fdtol != p->p_fdtol;
1379 fdtol = fdtol->fdl_next) {
1380 if ((fdtol->fdl_leader->p_flag &
1381 P_ADVLOCK) == 0)
1382 continue;
1383 fdtol->fdl_holdcount++;
1384 lf.l_whence = SEEK_SET;
1385 lf.l_start = 0;
1386 lf.l_len = 0;
1387 lf.l_type = F_UNLCK;
1388 vp = (struct vnode *)fp->f_data;
1389 (void) VOP_ADVLOCK(vp,
1390 (caddr_t)p->p_leader,
1391 F_UNLCK, &lf, F_POSIX);
1392 fdtol->fdl_holdcount--;
1393 if (fdtol->fdl_holdcount == 0 &&
1394 fdtol->fdl_wakeup != 0) {
1395 fdtol->fdl_wakeup = 0;
1396 wakeup(fdtol);
1397 }
1398 }
1399 }
1400 }
dadab5e9 1401 return (fdrop(fp, td));
984263bc
MD
1402}
1403
1404int
dadab5e9 1405fdrop(struct file *fp, struct thread *td)
984263bc
MD
1406{
1407 struct flock lf;
1408 struct vnode *vp;
1409 int error;
1410
1411 if (--fp->f_count > 0)
1412 return (0);
1413 if (fp->f_count < 0)
1414 panic("fdrop: count < 0");
1415 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
1416 lf.l_whence = SEEK_SET;
1417 lf.l_start = 0;
1418 lf.l_len = 0;
1419 lf.l_type = F_UNLCK;
1420 vp = (struct vnode *)fp->f_data;
1421 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
1422 }
1423 if (fp->f_ops != &badfileops)
dadab5e9 1424 error = fo_close(fp, td);
984263bc
MD
1425 else
1426 error = 0;
1427 ffree(fp);
1428 return (error);
1429}
1430
1431/*
1432 * Apply an advisory lock on a file descriptor.
1433 *
1434 * Just attempt to get a record lock of the requested type on
1435 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
1436 */
984263bc
MD
1437/* ARGSUSED */
1438int
41c20dac 1439flock(struct flock_args *uap)
984263bc 1440{
41c20dac
MD
1441 struct proc *p = curproc;
1442 struct filedesc *fdp = p->p_fd;
1443 struct file *fp;
984263bc
MD
1444 struct vnode *vp;
1445 struct flock lf;
1446
1447 if ((unsigned)uap->fd >= fdp->fd_nfiles ||
1448 (fp = fdp->fd_ofiles[uap->fd]) == NULL)
1449 return (EBADF);
1450 if (fp->f_type != DTYPE_VNODE)
1451 return (EOPNOTSUPP);
1452 vp = (struct vnode *)fp->f_data;
1453 lf.l_whence = SEEK_SET;
1454 lf.l_start = 0;
1455 lf.l_len = 0;
1456 if (uap->how & LOCK_UN) {
1457 lf.l_type = F_UNLCK;
1458 fp->f_flag &= ~FHASLOCK;
1459 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK));
1460 }
1461 if (uap->how & LOCK_EX)
1462 lf.l_type = F_WRLCK;
1463 else if (uap->how & LOCK_SH)
1464 lf.l_type = F_RDLCK;
1465 else
1466 return (EBADF);
1467 fp->f_flag |= FHASLOCK;
1468 if (uap->how & LOCK_NB)
1469 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK));
1470 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT));
1471}
1472
1473/*
1474 * File Descriptor pseudo-device driver (/dev/fd/).
1475 *
1476 * Opening minor device N dup()s the file (if any) connected to file
1477 * descriptor N belonging to the calling process. Note that this driver
1478 * consists of only the ``open()'' routine, because all subsequent
1479 * references to this file will be direct to the other driver.
1480 */
1481/* ARGSUSED */
1482static int
41c20dac 1483fdopen(dev_t dev, int mode, int type, struct thread *td)
984263bc 1484{
41c20dac 1485 KKASSERT(td->td_proc != NULL);
984263bc
MD
1486
1487 /*
1488 * XXX Kludge: set curproc->p_dupfd to contain the value of the
1489 * the file descriptor being sought for duplication. The error
1490 * return ensures that the vnode for this device will be released
1491 * by vn_open. Open will detect this special error and take the
1492 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
1493 * will simply report the error.
1494 */
41c20dac 1495 td->td_proc->p_dupfd = minor(dev);
984263bc
MD
1496 return (ENODEV);
1497}
1498
1499/*
1500 * Duplicate the specified descriptor to a free descriptor.
1501 */
1502int
41c20dac 1503dupfdopen(struct filedesc *fdp, int indx, int dfd, int mode, int error)
984263bc 1504{
41c20dac 1505 struct file *wfp;
984263bc
MD
1506 struct file *fp;
1507
1508 /*
1509 * If the to-be-dup'd fd number is greater than the allowed number
1510 * of file descriptors, or the fd to be dup'd has already been
1511 * closed, then reject.
1512 */
1513 if ((u_int)dfd >= fdp->fd_nfiles ||
1514 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
1515 return (EBADF);
1516 }
1517
1518 /*
1519 * There are two cases of interest here.
1520 *
1521 * For ENODEV simply dup (dfd) to file descriptor
1522 * (indx) and return.
1523 *
1524 * For ENXIO steal away the file structure from (dfd) and
1525 * store it in (indx). (dfd) is effectively closed by
1526 * this operation.
1527 *
1528 * Any other error code is just returned.
1529 */
1530 switch (error) {
1531 case ENODEV:
1532 /*
1533 * Check that the mode the file is being opened for is a
1534 * subset of the mode of the existing descriptor.
1535 */
1536 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag)
1537 return (EACCES);
1538 fp = fdp->fd_ofiles[indx];
1539#if 0
1540 if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED)
1541 (void) munmapfd(p, indx);
1542#endif
1543 fdp->fd_ofiles[indx] = wfp;
1544 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1545 fhold(wfp);
1546 if (indx > fdp->fd_lastfile)
1547 fdp->fd_lastfile = indx;
1548 /*
1549 * we now own the reference to fp that the ofiles[] array
1550 * used to own. Release it.
1551 */
1552 if (fp)
dadab5e9 1553 fdrop(fp, curthread);
984263bc
MD
1554 return (0);
1555
1556 case ENXIO:
1557 /*
1558 * Steal away the file pointer from dfd, and stuff it into indx.
1559 */
1560 fp = fdp->fd_ofiles[indx];
1561#if 0
1562 if (fp && fdp->fd_ofileflags[indx] & UF_MAPPED)
1563 (void) munmapfd(p, indx);
1564#endif
1565 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
1566 fdp->fd_ofiles[dfd] = NULL;
1567 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1568 fdp->fd_ofileflags[dfd] = 0;
1569
1570 /*
1571 * we now own the reference to fp that the ofiles[] array
1572 * used to own. Release it.
1573 */
1574 if (fp)
dadab5e9 1575 fdrop(fp, curthread);
984263bc
MD
1576 /*
1577 * Complete the clean up of the filedesc structure by
1578 * recomputing the various hints.
1579 */
1580 if (indx > fdp->fd_lastfile) {
1581 fdp->fd_lastfile = indx;
1582 } else {
1583 while (fdp->fd_lastfile > 0 &&
1584 fdp->fd_ofiles[fdp->fd_lastfile] == NULL) {
1585 fdp->fd_lastfile--;
1586 }
1587 if (dfd < fdp->fd_freefile)
1588 fdp->fd_freefile = dfd;
1589 }
1590 return (0);
1591
1592 default:
1593 return (error);
1594 }
1595 /* NOTREACHED */
1596}
1597
1598
1599struct filedesc_to_leader *
1600filedesc_to_leader_alloc(struct filedesc_to_leader *old,
1601 struct proc *leader)
1602{
1603 struct filedesc_to_leader *fdtol;
1604
1605 MALLOC(fdtol, struct filedesc_to_leader *,
1606 sizeof(struct filedesc_to_leader),
1607 M_FILEDESC_TO_LEADER,
1608 M_WAITOK);
1609 fdtol->fdl_refcount = 1;
1610 fdtol->fdl_holdcount = 0;
1611 fdtol->fdl_wakeup = 0;
1612 fdtol->fdl_leader = leader;
1613 if (old != NULL) {
1614 fdtol->fdl_next = old->fdl_next;
1615 fdtol->fdl_prev = old;
1616 old->fdl_next = fdtol;
1617 fdtol->fdl_next->fdl_prev = fdtol;
1618 } else {
1619 fdtol->fdl_next = fdtol;
1620 fdtol->fdl_prev = fdtol;
1621 }
1622 return fdtol;
1623}
1624
1625/*
1626 * Get file structures.
1627 */
1628static int
1629sysctl_kern_file(SYSCTL_HANDLER_ARGS)
1630{
1631 int error;
1632 struct file *fp;
1633
1634 if (!req->oldptr) {
1635 /*
1636 * overestimate by 10 files
1637 */
1638 return (SYSCTL_OUT(req, 0, sizeof(filehead) +
1639 (nfiles + 10) * sizeof(struct file)));
1640 }
1641
1642 error = SYSCTL_OUT(req, (caddr_t)&filehead, sizeof(filehead));
1643 if (error)
1644 return (error);
1645
1646 /*
1647 * followed by an array of file structures
1648 */
1649 LIST_FOREACH(fp, &filehead, f_list) {
1650 error = SYSCTL_OUT(req, (caddr_t)fp, sizeof (struct file));
1651 if (error)
1652 return (error);
1653 }
1654 return (0);
1655}
1656
1657SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
1658 0, 0, sysctl_kern_file, "S,file", "Entire file table");
1659
1660SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
1661 &maxfilesperproc, 0, "Maximum files allowed open per process");
1662
1663SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
1664 &maxfiles, 0, "Maximum number of files");
1665
1666SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
1667 &nfiles, 0, "System-wide number of open files");
1668
1669static void
1670fildesc_drvinit(void *unused)
1671{
1672 int fd;
1673
1674 for (fd = 0; fd < NUMFDESC; fd++)
1675 make_dev(&fildesc_cdevsw, fd,
1676 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
1677 make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
1678 make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
1679 make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
1680}
1681
1682struct fileops badfileops = {
f53ede20
MD
1683 NULL, /* port */
1684 0, /* autoq */
984263bc
MD
1685 badfo_readwrite,
1686 badfo_readwrite,
1687 badfo_ioctl,
1688 badfo_poll,
1689 badfo_kqfilter,
1690 badfo_stat,
1691 badfo_close
1692};
1693
1694static int
dadab5e9
MD
1695badfo_readwrite(
1696 struct file *fp,
1697 struct uio *uio,
1698 struct ucred *cred,
1699 int flags,
1700 struct thread *td
1701) {
984263bc
MD
1702 return (EBADF);
1703}
1704
1705static int
dadab5e9 1706badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td)
984263bc 1707{
984263bc
MD
1708 return (EBADF);
1709}
1710
1711static int
dadab5e9 1712badfo_poll(struct file *fp, int events, struct ucred *cred, struct thread *td)
984263bc 1713{
984263bc
MD
1714 return (0);
1715}
1716
1717static int
dadab5e9 1718badfo_kqfilter(struct file *fp, struct knote *kn)
984263bc 1719{
984263bc
MD
1720 return (0);
1721}
1722
1723static int
dadab5e9 1724badfo_stat(struct file *fp, struct stat *sb, struct thread *td)
984263bc 1725{
984263bc
MD
1726 return (EBADF);
1727}
1728
1729static int
dadab5e9 1730badfo_close(struct file *fp, struct thread *td)
984263bc 1731{
984263bc
MD
1732 return (EBADF);
1733}
1734
1735SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
1736 fildesc_drvinit,NULL)