kernel: Add three new commands to fctnl
[dragonfly.git] / sys / kern / kern_descrip.c
CommitLineData
984263bc 1/*
29d211fb
MD
2 * Copyright (c) 2005 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Jeffrey Hsu.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 *
984263bc
MD
35 * Copyright (c) 1982, 1986, 1989, 1991, 1993
36 * The Regents of the University of California. All rights reserved.
37 * (c) UNIX System Laboratories, Inc.
38 * All or some portions of this file are derived from material licensed
39 * to the University of California by American Telephone and Telegraph
40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41 * the permission of UNIX System Laboratories, Inc.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
dc71b7ab 51 * 3. Neither the name of the University nor the names of its contributors
984263bc
MD
52 * may be used to endorse or promote products derived from this software
53 * without specific prior written permission.
54 *
55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65 * SUCH DAMAGE.
66 *
67 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
c4cb6d8b 68 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
984263bc
MD
69 */
70
71#include "opt_compat.h"
72#include <sys/param.h>
73#include <sys/systm.h>
74#include <sys/malloc.h>
75#include <sys/sysproto.h>
76#include <sys/conf.h>
fef8985e 77#include <sys/device.h>
a724d72b 78#include <sys/file.h>
984263bc
MD
79#include <sys/filedesc.h>
80#include <sys/kernel.h>
81#include <sys/sysctl.h>
82#include <sys/vnode.h>
83#include <sys/proc.h>
fad57d0e 84#include <sys/nlookup.h>
984263bc
MD
85#include <sys/stat.h>
86#include <sys/filio.h>
87#include <sys/fcntl.h>
88#include <sys/unistd.h>
89#include <sys/resourcevar.h>
90#include <sys/event.h>
dda4b42b 91#include <sys/kern_syscall.h>
1c55bd1c 92#include <sys/kcore.h>
7b124c9f 93#include <sys/kinfo.h>
ea8f324c 94#include <sys/un.h>
984263bc
MD
95
96#include <vm/vm.h>
97#include <vm/vm_extern.h>
98
e43a034f 99#include <sys/thread2.h>
dadab5e9 100#include <sys/file2.h>
85fceac1 101#include <sys/spinlock2.h>
dadab5e9 102
5b287bba
MD
103static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
104static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
105static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
5b287bba 106static void ffree(struct file *fp);
fa541be6 107
984263bc
MD
108static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
109static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
110 "file desc to leader structures");
111MALLOC_DEFINE(M_FILE, "file", "Open file structure");
112static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
113
1e864525
MD
114static struct krate krate_uidinfo = { .freq = 1 };
115
984263bc
MD
116static d_open_t fdopen;
117#define NUMFDESC 64
118
119#define CDEV_MAJOR 22
fef8985e 120static struct dev_ops fildesc_ops = {
88abd8b5 121 { "FD", 0, 0 },
fef8985e 122 .d_open = fdopen,
984263bc
MD
123};
124
984263bc
MD
125/*
126 * Descriptor management.
127 */
2dd63755
MD
128static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
129static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
130static int nfiles; /* actual number of open files */
984263bc
MD
131extern int cmask;
132
8247b2f9
MD
133/*
134 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
5b287bba
MD
135 *
136 * MPSAFE - must be called with fdp->fd_spin exclusively held
8247b2f9
MD
137 */
138static __inline
139void
5b287bba 140fdfixup_locked(struct filedesc *fdp, int fd)
8247b2f9
MD
141{
142 if (fd < fdp->fd_freefile) {
143 fdp->fd_freefile = fd;
144 }
145 while (fdp->fd_lastfile >= 0 &&
146 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
147 fdp->fd_files[fdp->fd_lastfile].reserved == 0
148 ) {
149 --fdp->fd_lastfile;
150 }
151}
152
984263bc
MD
153/*
154 * System calls on descriptors.
5b287bba
MD
155 *
156 * MPSAFE
984263bc 157 */
984263bc 158int
753fd850 159sys_getdtablesize(struct getdtablesize_args *uap)
984263bc 160{
41c20dac 161 struct proc *p = curproc;
5b287bba 162 struct plimit *limit = p->p_limit;
d37c8f7f 163 int dtsize;
984263bc 164
287a8577 165 spin_lock(&limit->p_spin);
d37c8f7f
MD
166 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
167 dtsize = INT_MAX;
168 else
169 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
287a8577 170 spin_unlock(&limit->p_spin);
a7b16922 171
d37c8f7f
MD
172 if (dtsize > maxfilesperproc)
173 dtsize = maxfilesperproc;
174 if (dtsize < minfilesperproc)
175 dtsize = minfilesperproc;
176 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
177 dtsize = maxfilesperuser;
178 uap->sysmsg_result = dtsize;
984263bc
MD
179 return (0);
180}
181
182/*
183 * Duplicate a file descriptor to a particular value.
184 *
185 * note: keep in mind that a potential race condition exists when closing
186 * descriptors from a shared descriptor table (via rfork).
5b287bba
MD
187 *
188 * MPSAFE
984263bc 189 */
984263bc 190int
753fd850 191sys_dup2(struct dup2_args *uap)
984263bc 192{
dda4b42b 193 int error;
973c11b9 194 int fd = 0;
dda4b42b 195
973c11b9
MD
196 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
197 uap->sysmsg_fds[0] = fd;
dda4b42b
DRJ
198
199 return (error);
984263bc
MD
200}
201
202/*
203 * Duplicate a file descriptor.
5b287bba
MD
204 *
205 * MPSAFE
984263bc 206 */
984263bc 207int
753fd850 208sys_dup(struct dup_args *uap)
984263bc 209{
dda4b42b 210 int error;
973c11b9 211 int fd = 0;
984263bc 212
973c11b9
MD
213 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
214 uap->sysmsg_fds[0] = fd;
dda4b42b
DRJ
215
216 return (error);
984263bc
MD
217}
218
5b287bba
MD
219/*
220 * MPALMOSTSAFE - acquires mplock for fp operations
221 */
984263bc 222int
87de5057 223kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
984263bc 224{
dadab5e9
MD
225 struct thread *td = curthread;
226 struct proc *p = td->td_proc;
41c20dac 227 struct file *fp;
984263bc 228 struct vnode *vp;
984263bc 229 u_int newmin;
9ba76b73 230 u_int oflags;
3a907475 231 u_int nflags;
dda4b42b 232 int tmp, error, flg = F_POSIX;
984263bc 233
dadab5e9
MD
234 KKASSERT(p);
235
5b287bba
MD
236 /*
237 * Operations on file descriptors that do not require a file pointer.
238 */
dda4b42b 239 switch (cmd) {
984263bc 240 case F_GETFD:
85fceac1
MD
241 error = fgetfdflags(p->p_fd, fd, &tmp);
242 if (error == 0)
243 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
5b287bba 244 return (error);
984263bc
MD
245
246 case F_SETFD:
85fceac1
MD
247 if (dat->fc_cloexec & FD_CLOEXEC)
248 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
249 else
250 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
5b287bba
MD
251 return (error);
252 case F_DUPFD:
253 newmin = dat->fc_fd;
254 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
255 return (error);
a2988f1b
JM
256 case F_DUP2FD:
257 newmin = dat->fc_fd;
258 error = kern_dup(DUP_FIXED, fd, newmin, &dat->fc_fd);
259 return (error);
260 case F_DUPFD_CLOEXEC:
261 newmin = dat->fc_fd;
262 error = kern_dup(DUP_VARIABLE | DUP_CLOEXEC, fd, newmin,
263 &dat->fc_fd);
264 return (error);
265 case F_DUP2FD_CLOEXEC:
266 newmin = dat->fc_fd;
267 error = kern_dup(DUP_FIXED | DUP_CLOEXEC, fd, newmin,
268 &dat->fc_fd);
269 return (error);
5b287bba 270 default:
85fceac1 271 break;
5b287bba 272 }
984263bc 273
5b287bba
MD
274 /*
275 * Operations on file pointers
276 */
277 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
278 return (EBADF);
279
5b287bba 280 switch (cmd) {
984263bc 281 case F_GETFL:
dda4b42b 282 dat->fc_flags = OFLAGS(fp->f_flag);
85fceac1
MD
283 error = 0;
284 break;
984263bc
MD
285
286 case F_SETFL:
3a907475
MD
287 oflags = fp->f_flag;
288 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
289 nflags |= oflags & ~FCNTLFLAGS;
290
f58b505b 291 error = 0;
3a907475
MD
292 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
293 error = EINVAL;
294 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
e8cdd284 295 tmp = nflags & FASYNC;
87baaf0c
MD
296 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
297 cred, NULL);
f58b505b 298 }
3a907475
MD
299 if (error == 0)
300 fp->f_flag = nflags;
85fceac1 301 break;
984263bc
MD
302
303 case F_GETOWN:
87baaf0c
MD
304 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
305 cred, NULL);
85fceac1 306 break;
984263bc
MD
307
308 case F_SETOWN:
87baaf0c
MD
309 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
310 cred, NULL);
85fceac1 311 break;
984263bc
MD
312
313 case F_SETLKW:
314 flg |= F_WAIT;
315 /* Fall into F_SETLK */
316
317 case F_SETLK:
85fceac1
MD
318 if (fp->f_type != DTYPE_VNODE) {
319 error = EBADF;
320 break;
321 }
984263bc
MD
322 vp = (struct vnode *)fp->f_data;
323
324 /*
325 * copyin/lockop may block
326 */
dda4b42b
DRJ
327 if (dat->fc_flock.l_whence == SEEK_CUR)
328 dat->fc_flock.l_start += fp->f_offset;
984263bc 329
dda4b42b 330 switch (dat->fc_flock.l_type) {
984263bc
MD
331 case F_RDLCK:
332 if ((fp->f_flag & FREAD) == 0) {
333 error = EBADF;
334 break;
335 }
4643740a 336 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) {
616516c8 337 lwkt_gettoken(&p->p_leader->p_token);
4643740a 338 p->p_leader->p_flags |= P_ADVLOCK;
616516c8
MD
339 lwkt_reltoken(&p->p_leader->p_token);
340 }
984263bc 341 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
dda4b42b 342 &dat->fc_flock, flg);
984263bc
MD
343 break;
344 case F_WRLCK:
345 if ((fp->f_flag & FWRITE) == 0) {
346 error = EBADF;
347 break;
348 }
4643740a 349 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) {
616516c8 350 lwkt_gettoken(&p->p_leader->p_token);
4643740a 351 p->p_leader->p_flags |= P_ADVLOCK;
616516c8
MD
352 lwkt_reltoken(&p->p_leader->p_token);
353 }
984263bc 354 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
dda4b42b 355 &dat->fc_flock, flg);
984263bc
MD
356 break;
357 case F_UNLCK:
358 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
dda4b42b 359 &dat->fc_flock, F_POSIX);
984263bc
MD
360 break;
361 default:
362 error = EINVAL;
363 break;
364 }
85fceac1
MD
365
366 /*
367 * It is possible to race a close() on the descriptor while
368 * we were blocked getting the lock. If this occurs the
369 * close might not have caught the lock.
370 */
cf9f4e88 371 if (checkfdclosed(p->p_fd, fd, fp)) {
dda4b42b
DRJ
372 dat->fc_flock.l_whence = SEEK_SET;
373 dat->fc_flock.l_start = 0;
374 dat->fc_flock.l_len = 0;
375 dat->fc_flock.l_type = F_UNLCK;
984263bc 376 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
dda4b42b 377 F_UNLCK, &dat->fc_flock, F_POSIX);
984263bc 378 }
85fceac1 379 break;
984263bc
MD
380
381 case F_GETLK:
85fceac1
MD
382 if (fp->f_type != DTYPE_VNODE) {
383 error = EBADF;
384 break;
385 }
984263bc
MD
386 vp = (struct vnode *)fp->f_data;
387 /*
388 * copyin/lockop may block
389 */
dda4b42b
DRJ
390 if (dat->fc_flock.l_type != F_RDLCK &&
391 dat->fc_flock.l_type != F_WRLCK &&
392 dat->fc_flock.l_type != F_UNLCK) {
85fceac1
MD
393 error = EINVAL;
394 break;
984263bc 395 }
dda4b42b
DRJ
396 if (dat->fc_flock.l_whence == SEEK_CUR)
397 dat->fc_flock.l_start += fp->f_offset;
984263bc 398 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
dda4b42b 399 &dat->fc_flock, F_POSIX);
85fceac1 400 break;
984263bc 401 default:
85fceac1
MD
402 error = EINVAL;
403 break;
984263bc 404 }
5b287bba 405
85fceac1
MD
406 fdrop(fp);
407 return (error);
984263bc
MD
408}
409
dda4b42b
DRJ
410/*
411 * The file control system call.
5b287bba
MD
412 *
413 * MPSAFE
dda4b42b
DRJ
414 */
415int
753fd850 416sys_fcntl(struct fcntl_args *uap)
dda4b42b
DRJ
417{
418 union fcntl_dat dat;
419 int error;
420
421 switch (uap->cmd) {
422 case F_DUPFD:
a2988f1b
JM
423 case F_DUP2FD:
424 case F_DUPFD_CLOEXEC:
425 case F_DUP2FD_CLOEXEC:
dda4b42b
DRJ
426 dat.fc_fd = uap->arg;
427 break;
428 case F_SETFD:
429 dat.fc_cloexec = uap->arg;
430 break;
431 case F_SETFL:
432 dat.fc_flags = uap->arg;
433 break;
434 case F_SETOWN:
435 dat.fc_owner = uap->arg;
436 break;
437 case F_SETLKW:
438 case F_SETLK:
439 case F_GETLK:
440 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
5b287bba 441 sizeof(struct flock));
dda4b42b
DRJ
442 if (error)
443 return (error);
444 break;
445 }
446
9910d07b 447 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
dda4b42b
DRJ
448
449 if (error == 0) {
450 switch (uap->cmd) {
451 case F_DUPFD:
a2988f1b
JM
452 case F_DUP2FD:
453 case F_DUPFD_CLOEXEC:
454 case F_DUP2FD_CLOEXEC:
dda4b42b
DRJ
455 uap->sysmsg_result = dat.fc_fd;
456 break;
457 case F_GETFD:
458 uap->sysmsg_result = dat.fc_cloexec;
459 break;
460 case F_GETFL:
461 uap->sysmsg_result = dat.fc_flags;
462 break;
463 case F_GETOWN:
464 uap->sysmsg_result = dat.fc_owner;
a2988f1b 465 break;
dda4b42b
DRJ
466 case F_GETLK:
467 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
468 sizeof(struct flock));
469 break;
470 }
471 }
472
473 return (error);
474}
475
984263bc
MD
476/*
477 * Common code for dup, dup2, and fcntl(F_DUPFD).
dda4b42b 478 *
a2988f1b
JM
479 * There are three type flags: DUP_FIXED, DUP_VARIABLE, and DUP_CLOEXEC.
480 * The first two flags are mutually exclusive, and the third is optional.
481 * DUP_FIXED tells kern_dup() to destructively dup over an existing file
482 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup()
483 * to find the lowest unused file descriptor that is greater than or
484 * equal to "new". DUP_CLOEXEC, which works with either of the first
485 * two flags, sets the close-on-exec flag on the "new" file descriptor.
5b287bba
MD
486 *
487 * MPSAFE
984263bc 488 */
dda4b42b 489int
a2988f1b 490kern_dup(int flags, int old, int new, int *res)
984263bc 491{
dda4b42b
DRJ
492 struct thread *td = curthread;
493 struct proc *p = td->td_proc;
494 struct filedesc *fdp = p->p_fd;
984263bc
MD
495 struct file *fp;
496 struct file *delfp;
259b8ea0 497 int oldflags;
984263bc 498 int holdleaders;
d37c8f7f 499 int dtsize;
dda4b42b
DRJ
500 int error, newfd;
501
502 /*
503 * Verify that we have a valid descriptor to dup from and
504 * possibly to dup to.
d37c8f7f
MD
505 *
506 * NOTE: maxfilesperuser is not applicable to dup()
dda4b42b 507 */
259b8ea0 508retry:
d37c8f7f
MD
509 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
510 dtsize = INT_MAX;
511 else
512 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
513 if (dtsize > maxfilesperproc)
514 dtsize = maxfilesperproc;
515 if (dtsize < minfilesperproc)
516 dtsize = minfilesperproc;
517
518 if (new < 0 || new > dtsize)
5b287bba 519 return (EINVAL);
d37c8f7f 520
287a8577 521 spin_lock(&fdp->fd_spin);
5b287bba 522 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
287a8577 523 spin_unlock(&fdp->fd_spin);
dda4b42b 524 return (EBADF);
5b287bba 525 }
a2988f1b 526 if ((flags & DUP_FIXED) && old == new) {
dda4b42b 527 *res = new;
a2988f1b
JM
528 if (flags & DUP_CLOEXEC)
529 fdp->fd_files[new].fileflags |= UF_EXCLOSE;
287a8577 530 spin_unlock(&fdp->fd_spin);
dda4b42b
DRJ
531 return (0);
532 }
0679adc4 533 fp = fdp->fd_files[old].fp;
259b8ea0 534 oldflags = fdp->fd_files[old].fileflags;
5b287bba 535 fhold(fp); /* MPSAFE - can be called with a spinlock held */
dda4b42b
DRJ
536
537 /*
259b8ea0
MD
538 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
539 * if the requested descriptor is beyond the current table size.
540 *
541 * This can block. Retry if the source descriptor no longer matches
542 * or if our expectation in the expansion case races.
543 *
544 * If we are not expanding or allocating a new decriptor, then reset
545 * the target descriptor to a reserved state so we have a uniform
546 * setup for the next code block.
dda4b42b 547 */
a2988f1b 548 if ((flags & DUP_VARIABLE) || new >= fdp->fd_nfiles) {
287a8577 549 spin_unlock(&fdp->fd_spin);
dda4b42b 550 error = fdalloc(p, new, &newfd);
287a8577 551 spin_lock(&fdp->fd_spin);
dda4b42b 552 if (error) {
287a8577 553 spin_unlock(&fdp->fd_spin);
9f87144f 554 fdrop(fp);
dda4b42b
DRJ
555 return (error);
556 }
5b287bba
MD
557 /*
558 * Check for ripout
559 */
259b8ea0 560 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
5b287bba 561 fsetfd_locked(fdp, NULL, newfd);
287a8577 562 spin_unlock(&fdp->fd_spin);
259b8ea0
MD
563 fdrop(fp);
564 goto retry;
565 }
5b287bba
MD
566 /*
567 * Check for expansion race
568 */
a2988f1b 569 if ((flags & DUP_VARIABLE) == 0 && new != newfd) {
5b287bba 570 fsetfd_locked(fdp, NULL, newfd);
287a8577 571 spin_unlock(&fdp->fd_spin);
259b8ea0
MD
572 fdrop(fp);
573 goto retry;
574 }
5b287bba
MD
575 /*
576 * Check for ripout, newfd reused old (this case probably
577 * can't occur).
578 */
259b8ea0 579 if (old == newfd) {
5b287bba 580 fsetfd_locked(fdp, NULL, newfd);
287a8577 581 spin_unlock(&fdp->fd_spin);
259b8ea0
MD
582 fdrop(fp);
583 goto retry;
584 }
dda4b42b 585 new = newfd;
259b8ea0
MD
586 delfp = NULL;
587 } else {
588 if (fdp->fd_files[new].reserved) {
287a8577 589 spin_unlock(&fdp->fd_spin);
259b8ea0 590 fdrop(fp);
6ea70f76 591 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
259b8ea0
MD
592 tsleep(fdp, 0, "fdres", hz);
593 goto retry;
594 }
dda4b42b 595
259b8ea0
MD
596 /*
597 * If the target descriptor was never allocated we have
598 * to allocate it. If it was we have to clean out the
5b287bba
MD
599 * old descriptor. delfp inherits the ref from the
600 * descriptor table.
259b8ea0
MD
601 */
602 delfp = fdp->fd_files[new].fp;
603 fdp->fd_files[new].fp = NULL;
604 fdp->fd_files[new].reserved = 1;
605 if (delfp == NULL) {
5b287bba 606 fdreserve_locked(fdp, new, 1);
259b8ea0
MD
607 if (new > fdp->fd_lastfile)
608 fdp->fd_lastfile = new;
dda4b42b 609 }
259b8ea0 610
dda4b42b 611 }
984263bc 612
5b287bba
MD
613 /*
614 * NOTE: still holding an exclusive spinlock
615 */
616
984263bc 617 /*
259b8ea0
MD
618 * If a descriptor is being overwritten we may hve to tell
619 * fdfree() to sleep to ensure that all relevant process
620 * leaders can be traversed in closef().
984263bc 621 */
984263bc 622 if (delfp != NULL && p->p_fdtol != NULL) {
984263bc
MD
623 fdp->fd_holdleaderscount++;
624 holdleaders = 1;
259b8ea0 625 } else {
984263bc 626 holdleaders = 0;
259b8ea0 627 }
a2988f1b 628 KASSERT(delfp == NULL || (flags & DUP_FIXED),
259b8ea0 629 ("dup() picked an open file"));
984263bc
MD
630
631 /*
259b8ea0
MD
632 * Duplicate the source descriptor, update lastfile. If the new
633 * descriptor was not allocated and we aren't replacing an existing
634 * descriptor we have to mark the descriptor as being in use.
635 *
636 * The fd_files[] array inherits fp's hold reference.
984263bc 637 */
5b287bba 638 fsetfd_locked(fdp, fp, new);
a2988f1b
JM
639 if ((flags & DUP_CLOEXEC) != 0)
640 fdp->fd_files[new].fileflags = oldflags | UF_EXCLOSE;
641 else
642 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
287a8577 643 spin_unlock(&fdp->fd_spin);
5b287bba 644 fdrop(fp);
dda4b42b 645 *res = new;
984263bc
MD
646
647 /*
648 * If we dup'd over a valid file, we now own the reference to it
649 * and must dispose of it using closef() semantics (as if a
650 * close() were performed on it).
651 */
652 if (delfp) {
5a4b22b1 653 if (SLIST_FIRST(&delfp->f_klist))
34e191bd 654 knote_fdclose(delfp, fdp, new);
b8477cda 655 closef(delfp, p);
984263bc 656 if (holdleaders) {
287a8577 657 spin_lock(&fdp->fd_spin);
984263bc
MD
658 fdp->fd_holdleaderscount--;
659 if (fdp->fd_holdleaderscount == 0 &&
660 fdp->fd_holdleaderswakeup != 0) {
661 fdp->fd_holdleaderswakeup = 0;
287a8577 662 spin_unlock(&fdp->fd_spin);
984263bc 663 wakeup(&fdp->fd_holdleaderscount);
5b287bba 664 } else {
287a8577 665 spin_unlock(&fdp->fd_spin);
984263bc
MD
666 }
667 }
668 }
669 return (0);
670}
671
672/*
673 * If sigio is on the list associated with a process or process group,
674 * disable signalling from the device, remove sigio from the list and
675 * free sigio.
b5c4d81f
MD
676 *
677 * MPSAFE
984263bc
MD
678 */
679void
58c2553a 680funsetown(struct sigio **sigiop)
984263bc 681{
58c2553a
MD
682 struct pgrp *pgrp;
683 struct proc *p;
684 struct sigio *sigio;
685
686 if ((sigio = *sigiop) != NULL) {
687 lwkt_gettoken(&proc_token); /* protect sigio */
688 KKASSERT(sigiop == sigio->sio_myref);
689 sigio = *sigiop;
690 *sigiop = NULL;
691 lwkt_reltoken(&proc_token);
692 }
984263bc
MD
693 if (sigio == NULL)
694 return;
58c2553a 695
984263bc 696 if (sigio->sio_pgid < 0) {
58c2553a
MD
697 pgrp = sigio->sio_pgrp;
698 sigio->sio_pgrp = NULL;
699 lwkt_gettoken(&pgrp->pg_token);
700 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio);
701 lwkt_reltoken(&pgrp->pg_token);
702 pgrel(pgrp);
984263bc 703 } else /* if ((*sigiop)->sio_pgid > 0) */ {
58c2553a
MD
704 p = sigio->sio_proc;
705 sigio->sio_proc = NULL;
706 PHOLD(p);
707 lwkt_gettoken(&p->p_token);
708 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
709 lwkt_reltoken(&p->p_token);
710 PRELE(p);
984263bc
MD
711 }
712 crfree(sigio->sio_ucred);
58c2553a 713 sigio->sio_ucred = NULL;
efda3bd0 714 kfree(sigio, M_SIGIO);
984263bc
MD
715}
716
b5c4d81f 717/*
58c2553a
MD
718 * Free a list of sigio structures. Caller is responsible for ensuring
719 * that the list is MPSAFE.
b5c4d81f
MD
720 *
721 * MPSAFE
722 */
984263bc 723void
7bf8660a 724funsetownlst(struct sigiolst *sigiolst)
984263bc
MD
725{
726 struct sigio *sigio;
727
728 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
58c2553a 729 funsetown(sigio->sio_myref);
984263bc
MD
730}
731
732/*
733 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
734 *
735 * After permission checking, add a sigio structure to the sigio list for
736 * the process or process group.
b5c4d81f
MD
737 *
738 * MPSAFE
984263bc
MD
739 */
740int
7bf8660a 741fsetown(pid_t pgid, struct sigio **sigiop)
984263bc 742{
58c2553a
MD
743 struct proc *proc = NULL;
744 struct pgrp *pgrp = NULL;
984263bc 745 struct sigio *sigio;
b5c4d81f 746 int error;
984263bc
MD
747
748 if (pgid == 0) {
58c2553a 749 funsetown(sigiop);
984263bc
MD
750 return (0);
751 }
b5c4d81f 752
984263bc
MD
753 if (pgid > 0) {
754 proc = pfind(pgid);
b5c4d81f
MD
755 if (proc == NULL) {
756 error = ESRCH;
757 goto done;
758 }
984263bc
MD
759
760 /*
761 * Policy - Don't allow a process to FSETOWN a process
762 * in another session.
763 *
764 * Remove this test to allow maximum flexibility or
765 * restrict FSETOWN to the current process or process
766 * group for maximum safety.
767 */
b5c4d81f
MD
768 if (proc->p_session != curproc->p_session) {
769 error = EPERM;
770 goto done;
771 }
984263bc
MD
772 } else /* if (pgid < 0) */ {
773 pgrp = pgfind(-pgid);
b5c4d81f
MD
774 if (pgrp == NULL) {
775 error = ESRCH;
776 goto done;
777 }
984263bc
MD
778
779 /*
780 * Policy - Don't allow a process to FSETOWN a process
781 * in another session.
782 *
783 * Remove this test to allow maximum flexibility or
784 * restrict FSETOWN to the current process or process
785 * group for maximum safety.
786 */
b5c4d81f
MD
787 if (pgrp->pg_session != curproc->p_session) {
788 error = EPERM;
789 goto done;
790 }
984263bc 791 }
58c2553a 792 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO);
984263bc 793 if (pgid > 0) {
58c2553a
MD
794 KKASSERT(pgrp == NULL);
795 lwkt_gettoken(&proc->p_token);
984263bc
MD
796 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
797 sigio->sio_proc = proc;
58c2553a 798 lwkt_reltoken(&proc->p_token);
984263bc 799 } else {
58c2553a
MD
800 KKASSERT(proc == NULL);
801 lwkt_gettoken(&pgrp->pg_token);
984263bc
MD
802 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
803 sigio->sio_pgrp = pgrp;
58c2553a
MD
804 lwkt_reltoken(&pgrp->pg_token);
805 pgrp = NULL;
984263bc
MD
806 }
807 sigio->sio_pgid = pgid;
9910d07b 808 sigio->sio_ucred = crhold(curthread->td_ucred);
984263bc 809 /* It would be convenient if p_ruid was in ucred. */
9910d07b 810 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
984263bc 811 sigio->sio_myref = sigiop;
58c2553a
MD
812
813 lwkt_gettoken(&proc_token);
814 while (*sigiop)
815 funsetown(sigiop);
984263bc 816 *sigiop = sigio;
58c2553a 817 lwkt_reltoken(&proc_token);
b5c4d81f
MD
818 error = 0;
819done:
58c2553a
MD
820 if (pgrp)
821 pgrel(pgrp);
822 if (proc)
823 PRELE(proc);
b5c4d81f 824 return (error);
984263bc
MD
825}
826
827/*
828 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
b5c4d81f
MD
829 *
830 * MPSAFE
984263bc
MD
831 */
832pid_t
b5c4d81f 833fgetown(struct sigio **sigiop)
984263bc 834{
b5c4d81f
MD
835 struct sigio *sigio;
836 pid_t own;
837
838 lwkt_gettoken(&proc_token);
839 sigio = *sigiop;
840 own = (sigio != NULL ? sigio->sio_pgid : 0);
841 lwkt_reltoken(&proc_token);
842
843 return (own);
984263bc
MD
844}
845
4336d5df
JS
846/*
847 * Close many file descriptors.
5b287bba
MD
848 *
849 * MPSAFE
4336d5df 850 */
4336d5df 851int
753fd850 852sys_closefrom(struct closefrom_args *uap)
4336d5df
JS
853{
854 return(kern_closefrom(uap->fd));
855}
856
5b287bba
MD
857/*
858 * Close all file descriptors greater then or equal to fd
859 *
860 * MPSAFE
861 */
4336d5df
JS
862int
863kern_closefrom(int fd)
864{
865 struct thread *td = curthread;
866 struct proc *p = td->td_proc;
867 struct filedesc *fdp;
4336d5df
JS
868
869 KKASSERT(p);
870 fdp = p->p_fd;
871
5b287bba
MD
872 if (fd < 0)
873 return (EINVAL);
5e8cfac8 874
259b8ea0
MD
875 /*
876 * NOTE: This function will skip unassociated descriptors and
877 * reserved descriptors that have not yet been assigned.
878 * fd_lastfile can change as a side effect of kern_close().
879 */
287a8577 880 spin_lock(&fdp->fd_spin);
259b8ea0
MD
881 while (fd <= fdp->fd_lastfile) {
882 if (fdp->fd_files[fd].fp != NULL) {
287a8577 883 spin_unlock(&fdp->fd_spin);
5b287bba 884 /* ok if this races another close */
259b8ea0
MD
885 if (kern_close(fd) == EINTR)
886 return (EINTR);
287a8577 887 spin_lock(&fdp->fd_spin);
259b8ea0
MD
888 }
889 ++fd;
890 }
287a8577 891 spin_unlock(&fdp->fd_spin);
4336d5df
JS
892 return (0);
893}
894
984263bc
MD
895/*
896 * Close a file descriptor.
5b287bba
MD
897 *
898 * MPSAFE
984263bc 899 */
984263bc 900int
753fd850 901sys_close(struct close_args *uap)
12693083
MD
902{
903 return(kern_close(uap->fd));
904}
905
5b287bba 906/*
a081e067 907 * MPSAFE
5b287bba 908 */
12693083
MD
909int
910kern_close(int fd)
984263bc 911{
dadab5e9
MD
912 struct thread *td = curthread;
913 struct proc *p = td->td_proc;
914 struct filedesc *fdp;
41c20dac 915 struct file *fp;
984263bc
MD
916 int error;
917 int holdleaders;
918
dadab5e9
MD
919 KKASSERT(p);
920 fdp = p->p_fd;
921
287a8577 922 spin_lock(&fdp->fd_spin);
5b287bba 923 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
287a8577 924 spin_unlock(&fdp->fd_spin);
984263bc 925 return (EBADF);
5b287bba 926 }
984263bc
MD
927 holdleaders = 0;
928 if (p->p_fdtol != NULL) {
929 /*
930 * Ask fdfree() to sleep to ensure that all relevant
931 * process leaders can be traversed in closef().
932 */
933 fdp->fd_holdleaderscount++;
934 holdleaders = 1;
935 }
936
937 /*
938 * we now hold the fp reference that used to be owned by the descriptor
939 * array.
940 */
287a8577 941 spin_unlock(&fdp->fd_spin);
5a4b22b1 942 if (SLIST_FIRST(&fp->f_klist))
ccafe911 943 knote_fdclose(fp, fdp, fd);
b8477cda 944 error = closef(fp, p);
984263bc 945 if (holdleaders) {
287a8577 946 spin_lock(&fdp->fd_spin);
984263bc
MD
947 fdp->fd_holdleaderscount--;
948 if (fdp->fd_holdleaderscount == 0 &&
949 fdp->fd_holdleaderswakeup != 0) {
950 fdp->fd_holdleaderswakeup = 0;
287a8577 951 spin_unlock(&fdp->fd_spin);
984263bc 952 wakeup(&fdp->fd_holdleaderscount);
5b287bba 953 } else {
287a8577 954 spin_unlock(&fdp->fd_spin);
984263bc
MD
955 }
956 }
957 return (error);
958}
959
004d2de5
MD
960/*
961 * shutdown_args(int fd, int how)
962 */
963int
964kern_shutdown(int fd, int how)
965{
966 struct thread *td = curthread;
967 struct proc *p = td->td_proc;
004d2de5
MD
968 struct file *fp;
969 int error;
970
971 KKASSERT(p);
972
5b287bba 973 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
004d2de5 974 return (EBADF);
87de5057 975 error = fo_shutdown(fp, how);
9f87144f 976 fdrop(fp);
004d2de5
MD
977
978 return (error);
979}
980
3919ced0
MD
981/*
982 * MPALMOSTSAFE
983 */
004d2de5 984int
753fd850 985sys_shutdown(struct shutdown_args *uap)
004d2de5
MD
986{
987 int error;
988
989 error = kern_shutdown(uap->s, uap->how);
990
991 return (error);
992}
993
1ee6e3c6
MD
994/*
995 * MPSAFE
996 */
984263bc 997int
8f6f8622 998kern_fstat(int fd, struct stat *ub)
984263bc 999{
dadab5e9
MD
1000 struct thread *td = curthread;
1001 struct proc *p = td->td_proc;
41c20dac 1002 struct file *fp;
984263bc
MD
1003 int error;
1004
dadab5e9 1005 KKASSERT(p);
8f6f8622 1006
5b287bba 1007 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
984263bc 1008 return (EBADF);
9910d07b 1009 error = fo_stat(fp, ub, td->td_ucred);
9f87144f 1010 fdrop(fp);
8f6f8622 1011
984263bc
MD
1012 return (error);
1013}
984263bc
MD
1014
1015/*
1016 * Return status information about a file descriptor.
1ee6e3c6
MD
1017 *
1018 * MPSAFE
984263bc 1019 */
984263bc 1020int
753fd850 1021sys_fstat(struct fstat_args *uap)
984263bc 1022{
8f6f8622 1023 struct stat st;
984263bc
MD
1024 int error;
1025
8f6f8622
DRJ
1026 error = kern_fstat(uap->fd, &st);
1027
984263bc 1028 if (error == 0)
8f6f8622 1029 error = copyout(&st, uap->sb, sizeof(st));
984263bc
MD
1030 return (error);
1031}
1032
984263bc
MD
1033/*
1034 * Return pathconf information about a file descriptor.
3919ced0
MD
1035 *
1036 * MPALMOSTSAFE
984263bc 1037 */
984263bc 1038int
753fd850 1039sys_fpathconf(struct fpathconf_args *uap)
984263bc 1040{
dadab5e9
MD
1041 struct thread *td = curthread;
1042 struct proc *p = td->td_proc;
984263bc
MD
1043 struct file *fp;
1044 struct vnode *vp;
1045 int error = 0;
1046
5b287bba
MD
1047 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
1048 return (EBADF);
984263bc
MD
1049
1050 switch (fp->f_type) {
1051 case DTYPE_PIPE:
1052 case DTYPE_SOCKET:
1053 if (uap->name != _PC_PIPE_BUF) {
1054 error = EINVAL;
1055 } else {
c7114eea 1056 uap->sysmsg_result = PIPE_BUF;
984263bc
MD
1057 error = 0;
1058 }
1059 break;
1060 case DTYPE_FIFO:
1061 case DTYPE_VNODE:
1062 vp = (struct vnode *)fp->f_data;
d557216f 1063 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
984263bc
MD
1064 break;
1065 default:
1066 error = EOPNOTSUPP;
1067 break;
1068 }
9f87144f 1069 fdrop(fp);
984263bc
MD
1070 return(error);
1071}
1072
984263bc 1073static int fdexpand;
0c52fa62
SG
1074SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0,
1075 "Number of times a file table has been expanded");
984263bc 1076
5b287bba
MD
1077/*
1078 * Grow the file table so it can hold through descriptor (want).
1079 *
1080 * The fdp's spinlock must be held exclusively on entry and may be held
1081 * exclusively on return. The spinlock may be cycled by the routine.
1082 *
1083 * MPSAFE
1084 */
69908319 1085static void
5b287bba 1086fdgrow_locked(struct filedesc *fdp, int want)
69908319 1087{
0679adc4
MD
1088 struct fdnode *newfiles;
1089 struct fdnode *oldfiles;
69908319
JH
1090 int nf, extra;
1091
1092 nf = fdp->fd_nfiles;
1093 do {
1094 /* nf has to be of the form 2^n - 1 */
1095 nf = 2 * nf + 1;
1096 } while (nf <= want);
1097
287a8577 1098 spin_unlock(&fdp->fd_spin);
efda3bd0 1099 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
287a8577 1100 spin_lock(&fdp->fd_spin);
69908319
JH
1101
1102 /*
5b287bba
MD
1103 * We could have raced another extend while we were not holding
1104 * the spinlock.
69908319
JH
1105 */
1106 if (fdp->fd_nfiles >= nf) {
287a8577 1107 spin_unlock(&fdp->fd_spin);
efda3bd0 1108 kfree(newfiles, M_FILEDESC);
287a8577 1109 spin_lock(&fdp->fd_spin);
69908319
JH
1110 return;
1111 }
69908319
JH
1112 /*
1113 * Copy the existing ofile and ofileflags arrays
1114 * and zero the new portion of each array.
1115 */
1116 extra = nf - fdp->fd_nfiles;
0679adc4
MD
1117 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1118 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1119
1120 oldfiles = fdp->fd_files;
1121 fdp->fd_files = newfiles;
69908319 1122 fdp->fd_nfiles = nf;
0679adc4 1123
5b287bba 1124 if (oldfiles != fdp->fd_builtin_files) {
287a8577 1125 spin_unlock(&fdp->fd_spin);
efda3bd0 1126 kfree(oldfiles, M_FILEDESC);
287a8577 1127 spin_lock(&fdp->fd_spin);
5b287bba 1128 }
69908319
JH
1129 fdexpand++;
1130}
1131
1132/*
1133 * Number of nodes in right subtree, including the root.
1134 */
1135static __inline int
1136right_subtree_size(int n)
1137{
1138 return (n ^ (n | (n + 1)));
1139}
1140
1141/*
1142 * Bigger ancestor.
1143 */
1144static __inline int
1145right_ancestor(int n)
1146{
1147 return (n | (n + 1));
1148}
1149
1150/*
1151 * Smaller ancestor.
1152 */
1153static __inline int
1154left_ancestor(int n)
1155{
1156 return ((n & (n + 1)) - 1);
1157}
1158
5b287bba
MD
1159/*
1160 * Traverse the in-place binary tree buttom-up adjusting the allocation
1161 * count so scans can determine where free descriptors are located.
1162 *
1163 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1164 */
fa541be6 1165static
69908319 1166void
5b287bba 1167fdreserve_locked(struct filedesc *fdp, int fd, int incr)
69908319
JH
1168{
1169 while (fd >= 0) {
0679adc4
MD
1170 fdp->fd_files[fd].allocated += incr;
1171 KKASSERT(fdp->fd_files[fd].allocated >= 0);
69908319
JH
1172 fd = left_ancestor(fd);
1173 }
1174}
1175
1176/*
259b8ea0
MD
1177 * Reserve a file descriptor for the process. If no error occurs, the
1178 * caller MUST at some point call fsetfd() or assign a file pointer
1179 * or dispose of the reservation.
5b287bba
MD
1180 *
1181 * MPSAFE
69908319 1182 */
984263bc 1183int
dda4b42b 1184fdalloc(struct proc *p, int want, int *result)
984263bc 1185{
41c20dac 1186 struct filedesc *fdp = p->p_fd;
d37c8f7f 1187 struct uidinfo *uip;
69908319
JH
1188 int fd, rsize, rsum, node, lim;
1189
d37c8f7f
MD
1190 /*
1191 * Check dtable size limit
1192 */
287a8577 1193 spin_lock(&p->p_limit->p_spin);
d37c8f7f
MD
1194 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1195 lim = INT_MAX;
1196 else
1197 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
287a8577 1198 spin_unlock(&p->p_limit->p_spin);
a7b16922 1199
d37c8f7f
MD
1200 if (lim > maxfilesperproc)
1201 lim = maxfilesperproc;
1202 if (lim < minfilesperproc)
1203 lim = minfilesperproc;
69908319
JH
1204 if (want >= lim)
1205 return (EMFILE);
d37c8f7f
MD
1206
1207 /*
1208 * Check that the user has not run out of descriptors (non-root only).
1209 * As a safety measure the dtable is allowed to have at least
1210 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1211 */
1212 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1213 uip = p->p_ucred->cr_uidinfo;
1214 if (uip->ui_openfiles > maxfilesperuser) {
1e864525 1215 krateprintf(&krate_uidinfo,
d37c8f7f
MD
1216 "Warning: user %d pid %d (%s) ran out of "
1217 "file descriptors (%d/%d)\n",
1218 p->p_ucred->cr_uid, (int)p->p_pid,
1219 p->p_comm,
1220 uip->ui_openfiles, maxfilesperuser);
1221 return(ENFILE);
1222 }
1223 }
1224
1225 /*
1226 * Grow the dtable if necessary
1227 */
287a8577 1228 spin_lock(&fdp->fd_spin);
69908319 1229 if (want >= fdp->fd_nfiles)
5b287bba 1230 fdgrow_locked(fdp, want);
984263bc
MD
1231
1232 /*
1233 * Search for a free descriptor starting at the higher
1234 * of want or fd_freefile. If that fails, consider
1235 * expanding the ofile array.
259b8ea0
MD
1236 *
1237 * NOTE! the 'allocated' field is a cumulative recursive allocation
1238 * count. If we happen to see a value of 0 then we can shortcut
1239 * our search. Otherwise we run through through the tree going
1240 * down branches we know have free descriptor(s) until we hit a
1241 * leaf node. The leaf node will be free but will not necessarily
1242 * have an allocated field of 0.
984263bc 1243 */
69908319
JH
1244retry:
1245 /* move up the tree looking for a subtree with a free node */
1246 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1247 fd = right_ancestor(fd)) {
0679adc4 1248 if (fdp->fd_files[fd].allocated == 0)
69908319 1249 goto found;
984263bc 1250
69908319 1251 rsize = right_subtree_size(fd);
0679adc4 1252 if (fdp->fd_files[fd].allocated == rsize)
69908319 1253 continue; /* right subtree full */
984263bc
MD
1254
1255 /*
69908319
JH
1256 * Free fd is in the right subtree of the tree rooted at fd.
1257 * Call that subtree R. Look for the smallest (leftmost)
1258 * subtree of R with an unallocated fd: continue moving
1259 * down the left branch until encountering a full left
1260 * subtree, then move to the right.
984263bc 1261 */
69908319
JH
1262 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1263 node = fd + rsize;
0679adc4
MD
1264 rsum += fdp->fd_files[node].allocated;
1265 if (fdp->fd_files[fd].allocated == rsum + rsize) {
69908319 1266 fd = node; /* move to the right */
0679adc4 1267 if (fdp->fd_files[node].allocated == 0)
69908319
JH
1268 goto found;
1269 rsum = 0;
1270 }
984263bc 1271 }
69908319 1272 goto found;
984263bc 1273 }
69908319
JH
1274
1275 /*
1276 * No space in current array. Expand?
1277 */
5b287bba 1278 if (fdp->fd_nfiles >= lim) {
287a8577 1279 spin_unlock(&fdp->fd_spin);
69908319 1280 return (EMFILE);
5b287bba
MD
1281 }
1282 fdgrow_locked(fdp, want);
69908319
JH
1283 goto retry;
1284
1285found:
1286 KKASSERT(fd < fdp->fd_nfiles);
69908319
JH
1287 if (fd > fdp->fd_lastfile)
1288 fdp->fd_lastfile = fd;
1289 if (want <= fdp->fd_freefile)
1290 fdp->fd_freefile = fd;
1291 *result = fd;
0679adc4 1292 KKASSERT(fdp->fd_files[fd].fp == NULL);
259b8ea0
MD
1293 KKASSERT(fdp->fd_files[fd].reserved == 0);
1294 fdp->fd_files[fd].fileflags = 0;
1295 fdp->fd_files[fd].reserved = 1;
5b287bba 1296 fdreserve_locked(fdp, fd, 1);
287a8577 1297 spin_unlock(&fdp->fd_spin);
984263bc
MD
1298 return (0);
1299}
1300
1301/*
1302 * Check to see whether n user file descriptors
1303 * are available to the process p.
5b287bba
MD
1304 *
1305 * MPSAFE
984263bc
MD
1306 */
1307int
7bf8660a 1308fdavail(struct proc *p, int n)
984263bc 1309{
41c20dac 1310 struct filedesc *fdp = p->p_fd;
0679adc4 1311 struct fdnode *fdnode;
41c20dac 1312 int i, lim, last;
984263bc 1313
287a8577 1314 spin_lock(&p->p_limit->p_spin);
d37c8f7f
MD
1315 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1316 lim = INT_MAX;
1317 else
1318 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
287a8577 1319 spin_unlock(&p->p_limit->p_spin);
a7b16922 1320
d37c8f7f
MD
1321 if (lim > maxfilesperproc)
1322 lim = maxfilesperproc;
1323 if (lim < minfilesperproc)
1324 lim = minfilesperproc;
984263bc 1325
287a8577 1326 spin_lock(&fdp->fd_spin);
5b287bba 1327 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
287a8577 1328 spin_unlock(&fdp->fd_spin);
5b287bba
MD
1329 return (1);
1330 }
984263bc 1331 last = min(fdp->fd_nfiles, lim);
0679adc4
MD
1332 fdnode = &fdp->fd_files[fdp->fd_freefile];
1333 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
5b287bba 1334 if (fdnode->fp == NULL && --n <= 0) {
287a8577 1335 spin_unlock(&fdp->fd_spin);
984263bc 1336 return (1);
5b287bba 1337 }
984263bc 1338 }
287a8577 1339 spin_unlock(&fdp->fd_spin);
984263bc
MD
1340 return (0);
1341}
1342
b8477cda
MD
1343/*
1344 * Revoke open descriptors referencing (f_data, f_type)
1345 *
1346 * Any revoke executed within a prison is only able to
1347 * revoke descriptors for processes within that prison.
1348 *
1349 * Returns 0 on success or an error code.
1350 */
1351struct fdrevoke_info {
1352 void *data;
1353 short type;
1354 short unused;
1355 int count;
ea8f324c 1356 int intransit;
b8477cda
MD
1357 struct ucred *cred;
1358 struct file *nfp;
1359};
1360
ea8f324c
MD
1361static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1362static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
b8477cda
MD
1363
1364int
1365fdrevoke(void *f_data, short f_type, struct ucred *cred)
1366{
1367 struct fdrevoke_info info;
1368 int error;
1369
1370 bzero(&info, sizeof(info));
1371 info.data = f_data;
1372 info.type = f_type;
1373 info.cred = cred;
1374 error = falloc(NULL, &info.nfp, NULL);
1375 if (error)
1376 return (error);
ea8f324c
MD
1377
1378 /*
1379 * Scan the file pointer table once. dups do not dup file pointers,
1380 * only descriptors, so there is no leak. Set FREVOKED on the fps
1381 * being revoked.
1382 */
1383 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1384
1385 /*
1386 * If any fps were marked track down the related descriptors
1387 * and close them. Any dup()s at this point will notice
1388 * the FREVOKED already set in the fp and do the right thing.
1389 *
1390 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1391 * socket) bumped the intransit counter and will require a
1392 * scan. Races against fps leaving the socket are closed by
1393 * the socket code checking for FREVOKED.
1394 */
1395 if (info.count)
1396 allproc_scan(fdrevoke_proc_callback, &info);
1397 if (info.intransit)
1398 unp_revoke_gc(info.nfp);
b8477cda 1399 fdrop(info.nfp);
ea8f324c 1400 return(0);
b8477cda
MD
1401}
1402
ea8f324c
MD
1403/*
1404 * Locate matching file pointers directly.
2aa0ab56
MD
1405 *
1406 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
ea8f324c 1407 */
b8477cda 1408static int
ea8f324c
MD
1409fdrevoke_check_callback(struct file *fp, void *vinfo)
1410{
1411 struct fdrevoke_info *info = vinfo;
1412
1413 /*
1414 * File pointers already flagged for revokation are skipped.
1415 */
1416 if (fp->f_flag & FREVOKED)
1417 return(0);
1418
1419 /*
1420 * If revoking from a prison file pointers created outside of
1421 * that prison, or file pointers without creds, cannot be revoked.
1422 */
1423 if (info->cred->cr_prison &&
1424 (fp->f_cred == NULL ||
1425 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1426 return(0);
1427 }
1428
1429 /*
1430 * If the file pointer matches then mark it for revocation. The
1431 * flag is currently only used by unp_revoke_gc().
1432 *
1433 * info->count is a heuristic and can race in a SMP environment.
1434 */
1435 if (info->data == fp->f_data && info->type == fp->f_type) {
1436 atomic_set_int(&fp->f_flag, FREVOKED);
1437 info->count += fp->f_count;
1438 if (fp->f_msgcount)
1439 ++info->intransit;
1440 }
1441 return(0);
1442}
1443
1444/*
1445 * Locate matching file pointers via process descriptor tables.
1446 */
1447static int
1448fdrevoke_proc_callback(struct proc *p, void *vinfo)
b8477cda
MD
1449{
1450 struct fdrevoke_info *info = vinfo;
1451 struct filedesc *fdp;
1452 struct file *fp;
1453 int n;
1454
1455 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1456 return(0);
1457 if (info->cred->cr_prison &&
1458 info->cred->cr_prison != p->p_ucred->cr_prison) {
1459 return(0);
1460 }
1461
1462 /*
1463 * If the controlling terminal of the process matches the
1464 * vnode being revoked we clear the controlling terminal.
1465 *
1466 * The normal spec_close() may not catch this because it
1467 * uses curproc instead of p.
1468 */
1469 if (p->p_session && info->type == DTYPE_VNODE &&
1470 info->data == p->p_session->s_ttyvp) {
1471 p->p_session->s_ttyvp = NULL;
1472 vrele(info->data);
1473 }
1474
1475 /*
0a4a9c77 1476 * Softref the fdp to prevent it from being destroyed
b8477cda 1477 */
287a8577 1478 spin_lock(&p->p_spin);
0a4a9c77 1479 if ((fdp = p->p_fd) == NULL) {
287a8577 1480 spin_unlock(&p->p_spin);
b8477cda 1481 return(0);
0a4a9c77
MD
1482 }
1483 atomic_add_int(&fdp->fd_softrefs, 1);
287a8577 1484 spin_unlock(&p->p_spin);
0a4a9c77
MD
1485
1486 /*
1487 * Locate and close any matching file descriptors.
1488 */
287a8577 1489 spin_lock(&fdp->fd_spin);
b8477cda
MD
1490 for (n = 0; n < fdp->fd_nfiles; ++n) {
1491 if ((fp = fdp->fd_files[n].fp) == NULL)
1492 continue;
ea8f324c 1493 if (fp->f_flag & FREVOKED) {
b8477cda
MD
1494 fhold(info->nfp);
1495 fdp->fd_files[n].fp = info->nfp;
287a8577 1496 spin_unlock(&fdp->fd_spin);
ccafe911 1497 knote_fdclose(fp, fdp, n); /* XXX */
b8477cda 1498 closef(fp, p);
287a8577 1499 spin_lock(&fdp->fd_spin);
ea8f324c 1500 --info->count;
b8477cda
MD
1501 }
1502 }
287a8577 1503 spin_unlock(&fdp->fd_spin);
0a4a9c77 1504 atomic_subtract_int(&fdp->fd_softrefs, 1);
b8477cda
MD
1505 return(0);
1506}
1507
984263bc 1508/*
39f91578 1509 * falloc:
259b8ea0
MD
1510 * Create a new open file structure and reserve a file decriptor
1511 * for the process that refers to it.
1512 *
f3a2d8c4
MD
1513 * Root creds are checked using lp, or assumed if lp is NULL. If
1514 * resultfd is non-NULL then lp must also be non-NULL. No file
1515 * descriptor is reserved (and no process context is needed) if
1516 * resultfd is NULL.
fad57d0e 1517 *
259b8ea0
MD
1518 * A file pointer with a refcount of 1 is returned. Note that the
1519 * file pointer is NOT associated with the descriptor. If falloc
1520 * returns success, fsetfd() MUST be called to either associate the
1521 * file pointer or clear the reservation.
5b287bba 1522 *
2dd63755 1523 * MPSAFE
984263bc
MD
1524 */
1525int
f3a2d8c4 1526falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
984263bc 1527{
c4cb6d8b
HP
1528 static struct timeval lastfail;
1529 static int curfail;
fad57d0e 1530 struct file *fp;
f3a2d8c4 1531 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
fad57d0e 1532 int error;
c4cb6d8b 1533
fad57d0e
MD
1534 fp = NULL;
1535
1536 /*
1537 * Handle filetable full issues and root overfill.
1538 */
60ee93b9 1539 if (nfiles >= maxfiles - maxfilesrootres &&
f3a2d8c4 1540 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
c4cb6d8b 1541 if (ppsratecheck(&lastfail, &curfail, 1)) {
f3a2d8c4
MD
1542 kprintf("kern.maxfiles limit exceeded by uid %d, "
1543 "please see tuning(7).\n",
1544 cred->cr_ruid);
c4cb6d8b 1545 }
fad57d0e
MD
1546 error = ENFILE;
1547 goto done;
984263bc 1548 }
fad57d0e 1549
984263bc
MD
1550 /*
1551 * Allocate a new file descriptor.
984263bc 1552 */
efda3bd0 1553 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
85fceac1 1554 spin_init(&fp->f_spin);
ccafe911 1555 SLIST_INIT(&fp->f_klist);
984263bc 1556 fp->f_count = 1;
984263bc
MD
1557 fp->f_ops = &badfileops;
1558 fp->f_seqcount = 1;
d37c8f7f 1559 fsetcred(fp, cred);
287a8577 1560 spin_lock(&filehead_spin);
2dd63755 1561 nfiles++;
fad57d0e 1562 LIST_INSERT_HEAD(&filehead, fp, f_list);
287a8577 1563 spin_unlock(&filehead_spin);
fad57d0e 1564 if (resultfd) {
f3a2d8c4 1565 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
9f87144f 1566 fdrop(fp);
fad57d0e 1567 fp = NULL;
39f91578 1568 }
a235f7bb 1569 } else {
fad57d0e 1570 error = 0;
984263bc 1571 }
fad57d0e
MD
1572done:
1573 *resultfp = fp;
64f33bc8 1574 return (error);
fad57d0e
MD
1575}
1576
85fceac1 1577/*
cf9f4e88
MD
1578 * Check for races against a file descriptor by determining that the
1579 * file pointer is still associated with the specified file descriptor,
1580 * and a close is not currently in progress.
1581 *
85fceac1
MD
1582 * MPSAFE
1583 */
85fceac1 1584int
cf9f4e88 1585checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
85fceac1
MD
1586{
1587 int error;
1588
0846e4ce 1589 spin_lock_shared(&fdp->fd_spin);
cf9f4e88 1590 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
85fceac1
MD
1591 error = EBADF;
1592 else
1593 error = 0;
0846e4ce 1594 spin_unlock_shared(&fdp->fd_spin);
85fceac1
MD
1595 return (error);
1596}
1597
fad57d0e 1598/*
259b8ea0
MD
1599 * Associate a file pointer with a previously reserved file descriptor.
1600 * This function always succeeds.
1601 *
1602 * If fp is NULL, the file descriptor is returned to the pool.
fad57d0e 1603 */
fad57d0e 1604
5b287bba
MD
1605/*
1606 * MPSAFE (exclusive spinlock must be held on call)
1607 */
1608static void
1609fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1610{
259b8ea0
MD
1611 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1612 KKASSERT(fdp->fd_files[fd].reserved != 0);
1613 if (fp) {
fad57d0e 1614 fhold(fp);
259b8ea0
MD
1615 fdp->fd_files[fd].fp = fp;
1616 fdp->fd_files[fd].reserved = 0;
1617 } else {
1618 fdp->fd_files[fd].reserved = 0;
5b287bba
MD
1619 fdreserve_locked(fdp, fd, -1);
1620 fdfixup_locked(fdp, fd);
fad57d0e 1621 }
984263bc
MD
1622}
1623
5b287bba
MD
1624/*
1625 * MPSAFE
1626 */
1627void
f3a2d8c4 1628fsetfd(struct filedesc *fdp, struct file *fp, int fd)
5b287bba 1629{
287a8577 1630 spin_lock(&fdp->fd_spin);
5b287bba 1631 fsetfd_locked(fdp, fp, fd);
287a8577 1632 spin_unlock(&fdp->fd_spin);
5b287bba
MD
1633}
1634
1635/*
1636 * MPSAFE (exclusive spinlock must be held on call)
1637 */
fa541be6 1638static
259b8ea0 1639struct file *
5b287bba 1640funsetfd_locked(struct filedesc *fdp, int fd)
69908319 1641{
259b8ea0
MD
1642 struct file *fp;
1643
1644 if ((unsigned)fd >= fdp->fd_nfiles)
1645 return (NULL);
1646 if ((fp = fdp->fd_files[fd].fp) == NULL)
1647 return (NULL);
0679adc4
MD
1648 fdp->fd_files[fd].fp = NULL;
1649 fdp->fd_files[fd].fileflags = 0;
259b8ea0 1650
5b287bba
MD
1651 fdreserve_locked(fdp, fd, -1);
1652 fdfixup_locked(fdp, fd);
259b8ea0 1653 return(fp);
69908319
JH
1654}
1655
85fceac1
MD
1656/*
1657 * MPSAFE
1658 */
1659int
1660fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1661{
1662 int error;
1663
287a8577 1664 spin_lock(&fdp->fd_spin);
85fceac1
MD
1665 if (((u_int)fd) >= fdp->fd_nfiles) {
1666 error = EBADF;
1667 } else if (fdp->fd_files[fd].fp == NULL) {
1668 error = EBADF;
1669 } else {
1670 *flagsp = fdp->fd_files[fd].fileflags;
1671 error = 0;
1672 }
287a8577 1673 spin_unlock(&fdp->fd_spin);
85fceac1
MD
1674 return (error);
1675}
1676
1677/*
1678 * MPSAFE
1679 */
228b401d
MD
1680int
1681fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1682{
85fceac1
MD
1683 int error;
1684
287a8577 1685 spin_lock(&fdp->fd_spin);
85fceac1
MD
1686 if (((u_int)fd) >= fdp->fd_nfiles) {
1687 error = EBADF;
1688 } else if (fdp->fd_files[fd].fp == NULL) {
1689 error = EBADF;
1690 } else {
1691 fdp->fd_files[fd].fileflags |= add_flags;
1692 error = 0;
1693 }
287a8577 1694 spin_unlock(&fdp->fd_spin);
85fceac1 1695 return (error);
228b401d
MD
1696}
1697
85fceac1
MD
1698/*
1699 * MPSAFE
1700 */
228b401d
MD
1701int
1702fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1703{
85fceac1
MD
1704 int error;
1705
287a8577 1706 spin_lock(&fdp->fd_spin);
85fceac1
MD
1707 if (((u_int)fd) >= fdp->fd_nfiles) {
1708 error = EBADF;
1709 } else if (fdp->fd_files[fd].fp == NULL) {
1710 error = EBADF;
1711 } else {
1712 fdp->fd_files[fd].fileflags &= ~rem_flags;
1713 error = 0;
1714 }
287a8577 1715 spin_unlock(&fdp->fd_spin);
85fceac1 1716 return (error);
228b401d
MD
1717}
1718
d37c8f7f
MD
1719/*
1720 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1721 */
a235f7bb 1722void
d37c8f7f 1723fsetcred(struct file *fp, struct ucred *ncr)
a235f7bb 1724{
d37c8f7f
MD
1725 struct ucred *ocr;
1726 struct uidinfo *uip;
1727
1728 ocr = fp->f_cred;
1729 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1730 if (ocr) {
1731 uip = ocr->cr_uidinfo;
1732 atomic_add_int(&uip->ui_openfiles, -1);
1733 }
1734 if (ncr) {
1735 uip = ncr->cr_uidinfo;
1736 atomic_add_int(&uip->ui_openfiles, 1);
1737 }
1738 }
1739 if (ncr)
1740 crhold(ncr);
1741 fp->f_cred = ncr;
1742 if (ocr)
1743 crfree(ocr);
a235f7bb
MD
1744}
1745
984263bc
MD
1746/*
1747 * Free a file descriptor.
1748 */
5b287bba 1749static
984263bc 1750void
7bf8660a 1751ffree(struct file *fp)
984263bc
MD
1752{
1753 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
287a8577 1754 spin_lock(&filehead_spin);
984263bc 1755 LIST_REMOVE(fp, f_list);
2dd63755 1756 nfiles--;
287a8577 1757 spin_unlock(&filehead_spin);
d37c8f7f 1758 fsetcred(fp, NULL);
28623bf9
MD
1759 if (fp->f_nchandle.ncp)
1760 cache_drop(&fp->f_nchandle);
efda3bd0 1761 kfree(fp, M_FILE);
984263bc
MD
1762}
1763
228b401d
MD
1764/*
1765 * called from init_main, initialize filedesc0 for proc0.
1766 */
1767void
1768fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1769{
1770 p0->p_fd = fdp0;
1771 p0->p_fdtol = NULL;
1772 fdp0->fd_refcnt = 1;
1773 fdp0->fd_cmask = cmask;
1774 fdp0->fd_files = fdp0->fd_builtin_files;
1775 fdp0->fd_nfiles = NDFILE;
259b8ea0 1776 fdp0->fd_lastfile = -1;
85fceac1 1777 spin_init(&fdp0->fd_spin);
228b401d
MD
1778}
1779
984263bc
MD
1780/*
1781 * Build a new filedesc structure.
5b287bba
MD
1782 *
1783 * NOT MPSAFE (vref)
984263bc
MD
1784 */
1785struct filedesc *
7bf8660a 1786fdinit(struct proc *p)
984263bc 1787{
0679adc4 1788 struct filedesc *newfdp;
41c20dac 1789 struct filedesc *fdp = p->p_fd;
984263bc 1790
efda3bd0 1791 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
287a8577 1792 spin_lock(&fdp->fd_spin);
bccde7a3 1793 if (fdp->fd_cdir) {
0679adc4
MD
1794 newfdp->fd_cdir = fdp->fd_cdir;
1795 vref(newfdp->fd_cdir);
28623bf9 1796 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
690a3127 1797 }
bccde7a3
MD
1798
1799 /*
1800 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1801 * proc0, but should unconditionally exist in other processes.
1802 */
1803 if (fdp->fd_rdir) {
0679adc4
MD
1804 newfdp->fd_rdir = fdp->fd_rdir;
1805 vref(newfdp->fd_rdir);
28623bf9 1806 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
bccde7a3
MD
1807 }
1808 if (fdp->fd_jdir) {
0679adc4
MD
1809 newfdp->fd_jdir = fdp->fd_jdir;
1810 vref(newfdp->fd_jdir);
28623bf9 1811 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
690a3127 1812 }
287a8577 1813 spin_unlock(&fdp->fd_spin);
690a3127 1814
984263bc 1815 /* Create the file descriptor table. */
0679adc4
MD
1816 newfdp->fd_refcnt = 1;
1817 newfdp->fd_cmask = cmask;
1818 newfdp->fd_files = newfdp->fd_builtin_files;
1819 newfdp->fd_nfiles = NDFILE;
259b8ea0 1820 newfdp->fd_lastfile = -1;
85fceac1 1821 spin_init(&newfdp->fd_spin);
0679adc4
MD
1822
1823 return (newfdp);
984263bc
MD
1824}
1825
1826/*
1827 * Share a filedesc structure.
5b287bba
MD
1828 *
1829 * MPSAFE
984263bc
MD
1830 */
1831struct filedesc *
7bf8660a 1832fdshare(struct proc *p)
984263bc 1833{
5b287bba
MD
1834 struct filedesc *fdp;
1835
1836 fdp = p->p_fd;
287a8577 1837 spin_lock(&fdp->fd_spin);
5b287bba 1838 fdp->fd_refcnt++;
287a8577 1839 spin_unlock(&fdp->fd_spin);
5b287bba 1840 return (fdp);
984263bc
MD
1841}
1842
1843/*
1844 * Copy a filedesc structure.
5b287bba
MD
1845 *
1846 * MPSAFE
984263bc 1847 */
2994659f
VS
1848int
1849fdcopy(struct proc *p, struct filedesc **fpp)
984263bc 1850{
5b287bba
MD
1851 struct filedesc *fdp = p->p_fd;
1852 struct filedesc *newfdp;
0679adc4 1853 struct fdnode *fdnode;
41c20dac 1854 int i;
5b287bba 1855 int ni;
984263bc 1856
5b287bba
MD
1857 /*
1858 * Certain daemons might not have file descriptors.
1859 */
984263bc 1860 if (fdp == NULL)
2994659f 1861 return (0);
984263bc 1862
5b287bba
MD
1863 /*
1864 * Allocate the new filedesc and fd_files[] array. This can race
1865 * with operations by other threads on the fdp so we have to be
1866 * careful.
1867 */
2994659f
VS
1868 newfdp = kmalloc(sizeof(struct filedesc),
1869 M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK);
1870 if (newfdp == NULL) {
1871 *fpp = NULL;
1872 return (-1);
1873 }
5b287bba 1874again:
287a8577 1875 spin_lock(&fdp->fd_spin);
5b287bba
MD
1876 if (fdp->fd_lastfile < NDFILE) {
1877 newfdp->fd_files = newfdp->fd_builtin_files;
1878 i = NDFILE;
1879 } else {
1880 /*
1881 * We have to allocate (N^2-1) entries for our in-place
1882 * binary tree. Allow the table to shrink.
1883 */
1884 i = fdp->fd_nfiles;
1885 ni = (i - 1) / 2;
1886 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1887 i = ni;
1888 ni = (i - 1) / 2;
1889 }
287a8577 1890 spin_unlock(&fdp->fd_spin);
77652cad 1891 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
5b287bba
MD
1892 M_FILEDESC, M_WAITOK | M_ZERO);
1893
1894 /*
1895 * Check for race, retry
1896 */
287a8577 1897 spin_lock(&fdp->fd_spin);
5b287bba 1898 if (i <= fdp->fd_lastfile) {
287a8577 1899 spin_unlock(&fdp->fd_spin);
efda3bd0 1900 kfree(newfdp->fd_files, M_FILEDESC);
5b287bba
MD
1901 goto again;
1902 }
1903 }
1904
1905 /*
1906 * Dup the remaining fields. vref() and cache_hold() can be
1907 * safely called while holding the read spinlock on fdp.
1908 *
1909 * The read spinlock on fdp is still being held.
1910 *
1911 * NOTE: vref and cache_hold calls for the case where the vnode
1912 * or cache entry already has at least one ref may be called
1913 * while holding spin locks.
1914 */
1915 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
597aea93 1916 vref(newfdp->fd_cdir);
28623bf9 1917 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
690a3127 1918 }
6bdbb368
DR
1919 /*
1920 * We must check for fd_rdir here, at least for now because
1921 * the init process is created before we have access to the
1922 * rootvode to take a reference to it.
1923 */
5b287bba 1924 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
597aea93 1925 vref(newfdp->fd_rdir);
28623bf9 1926 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
690a3127 1927 }
5b287bba 1928 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
597aea93 1929 vref(newfdp->fd_jdir);
28623bf9 1930 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
690a3127 1931 }
984263bc 1932 newfdp->fd_refcnt = 1;
984263bc 1933 newfdp->fd_nfiles = i;
5b287bba
MD
1934 newfdp->fd_lastfile = fdp->fd_lastfile;
1935 newfdp->fd_freefile = fdp->fd_freefile;
1936 newfdp->fd_cmask = fdp->fd_cmask;
5b287bba 1937 spin_init(&newfdp->fd_spin);
984263bc
MD
1938
1939 /*
5b287bba
MD
1940 * Copy the descriptor table through (i). This also copies the
1941 * allocation state. Then go through and ref the file pointers
1942 * and clean up any KQ descriptors.
1943 *
259b8ea0
MD
1944 * kq descriptors cannot be copied. Since we haven't ref'd the
1945 * copied files yet we can ignore the return value from funsetfd().
5b287bba
MD
1946 *
1947 * The read spinlock on fdp is still being held.
984263bc 1948 */
5b287bba
MD
1949 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1950 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
259b8ea0
MD
1951 fdnode = &newfdp->fd_files[i];
1952 if (fdnode->reserved) {
5b287bba 1953 fdreserve_locked(newfdp, i, -1);
259b8ea0 1954 fdnode->reserved = 0;
5b287bba
MD
1955 fdfixup_locked(newfdp, i);
1956 } else if (fdnode->fp) {
1957 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1958 (void)funsetfd_locked(newfdp, i);
1959 } else {
1960 fhold(fdnode->fp);
1961 }
259b8ea0 1962 }
984263bc 1963 }
287a8577 1964 spin_unlock(&fdp->fd_spin);
2994659f
VS
1965 *fpp = newfdp;
1966 return (0);
984263bc
MD
1967}
1968
1969/*
1970 * Release a filedesc structure.
5b287bba
MD
1971 *
1972 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
984263bc
MD
1973 */
1974void
0a4a9c77 1975fdfree(struct proc *p, struct filedesc *repl)
984263bc 1976{
0a4a9c77 1977 struct filedesc *fdp;
0679adc4 1978 struct fdnode *fdnode;
41c20dac 1979 int i;
984263bc
MD
1980 struct filedesc_to_leader *fdtol;
1981 struct file *fp;
1982 struct vnode *vp;
1983 struct flock lf;
1984
0a4a9c77
MD
1985 /*
1986 * Certain daemons might not have file descriptors.
1987 */
464e801e
MD
1988 fdp = p->p_fd;
1989 if (fdp == NULL) {
1990 p->p_fd = repl;
984263bc 1991 return;
464e801e 1992 }
984263bc 1993
5b287bba 1994 /*
0a4a9c77 1995 * Severe messing around to follow.
5b287bba 1996 */
287a8577 1997 spin_lock(&fdp->fd_spin);
5b287bba 1998
984263bc
MD
1999 /* Check for special need to clear POSIX style locks */
2000 fdtol = p->p_fdtol;
2001 if (fdtol != NULL) {
2002 KASSERT(fdtol->fdl_refcount > 0,
2003 ("filedesc_to_refcount botch: fdl_refcount=%d",
2004 fdtol->fdl_refcount));
2005 if (fdtol->fdl_refcount == 1 &&
4643740a 2006 (p->p_leader->p_flags & P_ADVLOCK) != 0) {
259b8ea0
MD
2007 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2008 fdnode = &fdp->fd_files[i];
0679adc4 2009 if (fdnode->fp == NULL ||
259b8ea0 2010 fdnode->fp->f_type != DTYPE_VNODE) {
984263bc 2011 continue;
259b8ea0 2012 }
0679adc4 2013 fp = fdnode->fp;
984263bc 2014 fhold(fp);
287a8577 2015 spin_unlock(&fdp->fd_spin);
5b287bba 2016
984263bc
MD
2017 lf.l_whence = SEEK_SET;
2018 lf.l_start = 0;
2019 lf.l_len = 0;
2020 lf.l_type = F_UNLCK;
2021 vp = (struct vnode *)fp->f_data;
2022 (void) VOP_ADVLOCK(vp,
2023 (caddr_t)p->p_leader,
2024 F_UNLCK,
2025 &lf,
2026 F_POSIX);
9f87144f 2027 fdrop(fp);
287a8577 2028 spin_lock(&fdp->fd_spin);
984263bc
MD
2029 }
2030 }
2031 retry:
2032 if (fdtol->fdl_refcount == 1) {
2033 if (fdp->fd_holdleaderscount > 0 &&
4643740a 2034 (p->p_leader->p_flags & P_ADVLOCK) != 0) {
984263bc
MD
2035 /*
2036 * close() or do_dup() has cleared a reference
2037 * in a shared file descriptor table.
2038 */
2039 fdp->fd_holdleaderswakeup = 1;
e590ee86 2040 ssleep(&fdp->fd_holdleaderscount,
7749886d 2041 &fdp->fd_spin, 0, "fdlhold", 0);
984263bc
MD
2042 goto retry;
2043 }
2044 if (fdtol->fdl_holdcount > 0) {
2045 /*
2046 * Ensure that fdtol->fdl_leader
2047 * remains valid in closef().
2048 */
2049 fdtol->fdl_wakeup = 1;
e590ee86 2050 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
984263bc
MD
2051 goto retry;
2052 }
2053 }
2054 fdtol->fdl_refcount--;
2055 if (fdtol->fdl_refcount == 0 &&
2056 fdtol->fdl_holdcount == 0) {
2057 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2058 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
5b287bba 2059 } else {
984263bc 2060 fdtol = NULL;
5b287bba 2061 }
984263bc 2062 p->p_fdtol = NULL;
5b287bba 2063 if (fdtol != NULL) {
287a8577 2064 spin_unlock(&fdp->fd_spin);
efda3bd0 2065 kfree(fdtol, M_FILEDESC_TO_LEADER);
287a8577 2066 spin_lock(&fdp->fd_spin);
5b287bba 2067 }
984263bc 2068 }
5b287bba 2069 if (--fdp->fd_refcnt > 0) {
287a8577
AH
2070 spin_unlock(&fdp->fd_spin);
2071 spin_lock(&p->p_spin);
464e801e 2072 p->p_fd = repl;
287a8577 2073 spin_unlock(&p->p_spin);
984263bc 2074 return;
5b287bba 2075 }
464e801e
MD
2076
2077 /*
2078 * Even though we are the last reference to the structure allproc
2079 * scans may still reference the structure. Maintain proper
2080 * locks until we can replace p->p_fd.
2081 *
2082 * Also note that kqueue's closef still needs to reference the
2083 * fdp via p->p_fd, so we have to close the descriptors before
2084 * we replace p->p_fd.
2085 */
2086 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2087 if (fdp->fd_files[i].fp) {
2088 fp = funsetfd_locked(fdp, i);
2089 if (fp) {
287a8577 2090 spin_unlock(&fdp->fd_spin);
5a4b22b1 2091 if (SLIST_FIRST(&fp->f_klist))
34e191bd 2092 knote_fdclose(fp, fdp, i);
464e801e 2093 closef(fp, p);
287a8577 2094 spin_lock(&fdp->fd_spin);
464e801e
MD
2095 }
2096 }
2097 }
287a8577 2098 spin_unlock(&fdp->fd_spin);
5b287bba 2099
464e801e
MD
2100 /*
2101 * Interlock against an allproc scan operations (typically frevoke).
2102 */
287a8577 2103 spin_lock(&p->p_spin);
464e801e 2104 p->p_fd = repl;
287a8577 2105 spin_unlock(&p->p_spin);
464e801e 2106
0a4a9c77
MD
2107 /*
2108 * Wait for any softrefs to go away. This race rarely occurs so
2109 * we can use a non-critical-path style poll/sleep loop. The
2110 * race only occurs against allproc scans.
2111 *
2112 * No new softrefs can occur with the fdp disconnected from the
2113 * process.
2114 */
2115 if (fdp->fd_softrefs) {
2116 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2117 while (fdp->fd_softrefs)
2118 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2119 }
2120
0679adc4 2121 if (fdp->fd_files != fdp->fd_builtin_files)
efda3bd0 2122 kfree(fdp->fd_files, M_FILEDESC);
690a3127 2123 if (fdp->fd_cdir) {
28623bf9 2124 cache_drop(&fdp->fd_ncdir);
984263bc 2125 vrele(fdp->fd_cdir);
690a3127 2126 }
bccde7a3 2127 if (fdp->fd_rdir) {
28623bf9 2128 cache_drop(&fdp->fd_nrdir);
bccde7a3
MD
2129 vrele(fdp->fd_rdir);
2130 }
690a3127 2131 if (fdp->fd_jdir) {
28623bf9 2132 cache_drop(&fdp->fd_njdir);
984263bc 2133 vrele(fdp->fd_jdir);
690a3127 2134 }
efda3bd0 2135 kfree(fdp, M_FILEDESC);
984263bc
MD
2136}
2137
fa541be6
MD
2138/*
2139 * Retrieve and reference the file pointer associated with a descriptor.
85fceac1
MD
2140 *
2141 * MPSAFE
fa541be6
MD
2142 */
2143struct file *
2144holdfp(struct filedesc *fdp, int fd, int flag)
2145{
2146 struct file* fp;
2147
0846e4ce 2148 spin_lock_shared(&fdp->fd_spin);
85fceac1
MD
2149 if (((u_int)fd) >= fdp->fd_nfiles) {
2150 fp = NULL;
2151 goto done;
2152 }
fa541be6 2153 if ((fp = fdp->fd_files[fd].fp) == NULL)
85fceac1
MD
2154 goto done;
2155 if ((fp->f_flag & flag) == 0 && flag != -1) {
2156 fp = NULL;
2157 goto done;
2158 }
fa541be6 2159 fhold(fp);
85fceac1 2160done:
0846e4ce 2161 spin_unlock_shared(&fdp->fd_spin);
fa541be6
MD
2162 return (fp);
2163}
2164
2165/*
2166 * holdsock() - load the struct file pointer associated
2167 * with a socket into *fpp. If an error occurs, non-zero
2168 * will be returned and *fpp will be set to NULL.
5b287bba
MD
2169 *
2170 * MPSAFE
fa541be6
MD
2171 */
2172int
5b287bba 2173holdsock(struct filedesc *fdp, int fd, struct file **fpp)
fa541be6
MD
2174{
2175 struct file *fp;
5b287bba 2176 int error;
fa541be6 2177
0846e4ce 2178 spin_lock_shared(&fdp->fd_spin);
5b287bba
MD
2179 if ((unsigned)fd >= fdp->fd_nfiles) {
2180 error = EBADF;
2181 fp = NULL;
2182 goto done;
2183 }
2184 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2185 error = EBADF;
2186 goto done;
2187 }
2188 if (fp->f_type != DTYPE_SOCKET) {
2189 error = ENOTSOCK;
2190 goto done;
2191 }
fa541be6 2192 fhold(fp);
5b287bba
MD
2193 error = 0;
2194done:
0846e4ce 2195 spin_unlock_shared(&fdp->fd_spin);
fa541be6
MD
2196 *fpp = fp;
2197 return (error);
2198}
2199
228b401d 2200/*
5b287bba
MD
2201 * Convert a user file descriptor to a held file pointer.
2202 *
2203 * MPSAFE
228b401d
MD
2204 */
2205int
5b287bba 2206holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
228b401d
MD
2207{
2208 struct file *fp;
5b287bba
MD
2209 int error;
2210
0846e4ce 2211 spin_lock_shared(&fdp->fd_spin);
5b287bba
MD
2212 if ((unsigned)fd >= fdp->fd_nfiles) {
2213 error = EBADF;
2214 fp = NULL;
2215 goto done;
2216 }
2217 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2218 error = EBADF;
2219 goto done;
2220 }
2221 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
8f9f2535 2222 fp = NULL;
5b287bba
MD
2223 error = EINVAL;
2224 goto done;
2225 }
2226 fhold(fp);
2227 error = 0;
2228done:
0846e4ce 2229 spin_unlock_shared(&fdp->fd_spin);
228b401d 2230 *fpp = fp;
5b287bba 2231 return (error);
228b401d
MD
2232}
2233
984263bc
MD
2234/*
2235 * For setugid programs, we don't want to people to use that setugidness
2236 * to generate error messages which write to a file which otherwise would
2237 * otherwise be off-limits to the process.
2238 *
2239 * This is a gross hack to plug the hole. A better solution would involve
2240 * a special vop or other form of generalized access control mechanism. We
2241 * go ahead and just reject all procfs file systems accesses as dangerous.
2242 *
2243 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2244 * sufficient. We also don't for check setugidness since we know we are.
2245 */
2246static int
2247is_unsafe(struct file *fp)
2248{
2249 if (fp->f_type == DTYPE_VNODE &&
2250 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2251 return (1);
2252 return (0);
2253}
2254
2255/*
2256 * Make this setguid thing safe, if at all possible.
5b287bba
MD
2257 *
2258 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
984263bc
MD
2259 */
2260void
dadab5e9 2261setugidsafety(struct proc *p)
984263bc
MD
2262{
2263 struct filedesc *fdp = p->p_fd;
41c20dac 2264 int i;
984263bc
MD
2265
2266 /* Certain daemons might not have file descriptors. */
2267 if (fdp == NULL)
2268 return;
2269
2270 /*
0679adc4 2271 * note: fdp->fd_files may be reallocated out from under us while
984263bc
MD
2272 * we are blocked in a close. Be careful!
2273 */
2274 for (i = 0; i <= fdp->fd_lastfile; i++) {
2275 if (i > 2)
2276 break;
0679adc4 2277 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
984263bc
MD
2278 struct file *fp;
2279
984263bc
MD
2280 /*
2281 * NULL-out descriptor prior to close to avoid
2282 * a race while close blocks.
2283 */
ccafe911
MD
2284 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2285 knote_fdclose(fp, fdp, i);
b8477cda 2286 closef(fp, p);
ccafe911 2287 }
984263bc
MD
2288 }
2289 }
984263bc
MD
2290}
2291
2292/*
2293 * Close any files on exec?
5b287bba
MD
2294 *
2295 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
984263bc
MD
2296 */
2297void
dadab5e9 2298fdcloseexec(struct proc *p)
984263bc
MD
2299{
2300 struct filedesc *fdp = p->p_fd;
41c20dac 2301 int i;
984263bc
MD
2302
2303 /* Certain daemons might not have file descriptors. */
2304 if (fdp == NULL)
2305 return;
2306
2307 /*
0679adc4
MD
2308 * We cannot cache fd_files since operations may block and rip
2309 * them out from under us.
984263bc
MD
2310 */
2311 for (i = 0; i <= fdp->fd_lastfile; i++) {
0679adc4
MD
2312 if (fdp->fd_files[i].fp != NULL &&
2313 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
984263bc
MD
2314 struct file *fp;
2315
984263bc
MD
2316 /*
2317 * NULL-out descriptor prior to close to avoid
2318 * a race while close blocks.
2319 */
ccafe911
MD
2320 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2321 knote_fdclose(fp, fdp, i);
b8477cda 2322 closef(fp, p);
ccafe911 2323 }
984263bc
MD
2324 }
2325 }
984263bc
MD
2326}
2327
2328/*
2329 * It is unsafe for set[ug]id processes to be started with file
2330 * descriptors 0..2 closed, as these descriptors are given implicit
2331 * significance in the Standard C library. fdcheckstd() will create a
2332 * descriptor referencing /dev/null for each of stdin, stdout, and
2333 * stderr that is not already open.
5b287bba
MD
2334 *
2335 * NOT MPSAFE - calls falloc, vn_open, etc
984263bc
MD
2336 */
2337int
f3a2d8c4 2338fdcheckstd(struct lwp *lp)
984263bc 2339{
fad57d0e 2340 struct nlookupdata nd;
dadab5e9
MD
2341 struct filedesc *fdp;
2342 struct file *fp;
973c11b9 2343 int retval;
259b8ea0 2344 int i, error, flags, devnull;
984263bc 2345
f3a2d8c4 2346 fdp = lp->lwp_proc->p_fd;
ea8f324c
MD
2347 if (fdp == NULL)
2348 return (0);
2349 devnull = -1;
2350 error = 0;
2351 for (i = 0; i < 3; i++) {
0679adc4 2352 if (fdp->fd_files[i].fp != NULL)
fad57d0e
MD
2353 continue;
2354 if (devnull < 0) {
f3a2d8c4 2355 if ((error = falloc(lp, &fp, &devnull)) != 0)
fad57d0e
MD
2356 break;
2357
2358 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2359 NLC_FOLLOW|NLC_LOCKVP);
2360 flags = FREAD | FWRITE;
2361 if (error == 0)
2362 error = vn_open(&nd, fp, flags, 0);
2363 if (error == 0)
f3a2d8c4 2364 fsetfd(fdp, fp, devnull);
259b8ea0 2365 else
f3a2d8c4 2366 fsetfd(fdp, NULL, devnull);
9f87144f 2367 fdrop(fp);
fad57d0e
MD
2368 nlookup_done(&nd);
2369 if (error)
2370 break;
259b8ea0 2371 KKASSERT(i == devnull);
fad57d0e
MD
2372 } else {
2373 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2374 if (error != 0)
2375 break;
2376 }
ea8f324c
MD
2377 }
2378 return (error);
984263bc
MD
2379}
2380
2381/*
2382 * Internal form of close.
2383 * Decrement reference count on file structure.
dadab5e9 2384 * Note: td and/or p may be NULL when closing a file
984263bc 2385 * that was being passed in a message.
5b287bba
MD
2386 *
2387 * MPALMOSTSAFE - acquires mplock for VOP operations
984263bc
MD
2388 */
2389int
b8477cda 2390closef(struct file *fp, struct proc *p)
984263bc
MD
2391{
2392 struct vnode *vp;
2393 struct flock lf;
2394 struct filedesc_to_leader *fdtol;
2395
2396 if (fp == NULL)
2397 return (0);
b8477cda 2398
984263bc
MD
2399 /*
2400 * POSIX record locking dictates that any close releases ALL
2401 * locks owned by this process. This is handled by setting
2402 * a flag in the unlock to free ONLY locks obeying POSIX
2403 * semantics, and not to free BSD-style file locks.
2404 * If the descriptor was in a message, POSIX-style locks
2405 * aren't passed with the descriptor.
2406 */
7749886d
MD
2407 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2408 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2409 ) {
4643740a 2410 if ((p->p_leader->p_flags & P_ADVLOCK) != 0) {
984263bc
MD
2411 lf.l_whence = SEEK_SET;
2412 lf.l_start = 0;
2413 lf.l_len = 0;
2414 lf.l_type = F_UNLCK;
2415 vp = (struct vnode *)fp->f_data;
2416 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2417 &lf, F_POSIX);
2418 }
2419 fdtol = p->p_fdtol;
2420 if (fdtol != NULL) {
b5c4d81f 2421 lwkt_gettoken(&p->p_token);
984263bc
MD
2422 /*
2423 * Handle special case where file descriptor table
2424 * is shared between multiple process leaders.
2425 */
2426 for (fdtol = fdtol->fdl_next;
2427 fdtol != p->p_fdtol;
2428 fdtol = fdtol->fdl_next) {
4643740a 2429 if ((fdtol->fdl_leader->p_flags &
984263bc
MD
2430 P_ADVLOCK) == 0)
2431 continue;
2432 fdtol->fdl_holdcount++;
2433 lf.l_whence = SEEK_SET;
2434 lf.l_start = 0;
2435 lf.l_len = 0;
2436 lf.l_type = F_UNLCK;
2437 vp = (struct vnode *)fp->f_data;
2438 (void) VOP_ADVLOCK(vp,
661fc1a0 2439 (caddr_t)fdtol->fdl_leader,
984263bc
MD
2440 F_UNLCK, &lf, F_POSIX);
2441 fdtol->fdl_holdcount--;
2442 if (fdtol->fdl_holdcount == 0 &&
2443 fdtol->fdl_wakeup != 0) {
2444 fdtol->fdl_wakeup = 0;
2445 wakeup(fdtol);
2446 }
2447 }
b5c4d81f 2448 lwkt_reltoken(&p->p_token);
984263bc
MD
2449 }
2450 }
9f87144f 2451 return (fdrop(fp));
984263bc
MD
2452}
2453
85fceac1
MD
2454/*
2455 * MPSAFE
2456 *
2457 * fhold() can only be called if f_count is already at least 1 (i.e. the
2458 * caller of fhold() already has a reference to the file pointer in some
5b287bba
MD
2459 * manner or other).
2460 *
bc1eb676
MD
2461 * f_count is not spin-locked. Instead, atomic ops are used for
2462 * incrementing, decrementing, and handling the 1->0 transition.
85fceac1
MD
2463 */
2464void
2465fhold(struct file *fp)
2466{
2467 atomic_add_int(&fp->f_count, 1);
2468}
2469
2470/*
bc1eb676 2471 * fdrop() - drop a reference to a descriptor
85fceac1 2472 *
5b287bba 2473 * MPALMOSTSAFE - acquires mplock for final close sequence
85fceac1 2474 */
984263bc 2475int
9f87144f 2476fdrop(struct file *fp)
984263bc
MD
2477{
2478 struct flock lf;
2479 struct vnode *vp;
2480 int error;
2481
bc1eb676
MD
2482 /*
2483 * A combined fetch and subtract is needed to properly detect
2484 * 1->0 transitions, otherwise two cpus dropping from a ref
2485 * count of 2 might both try to run the 1->0 code.
2486 */
2487 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
984263bc 2488 return (0);
85fceac1 2489
34e191bd 2490 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
5b287bba 2491
85fceac1
MD
2492 /*
2493 * The last reference has gone away, we own the fp structure free
2494 * and clear.
2495 */
984263bc
MD
2496 if (fp->f_count < 0)
2497 panic("fdrop: count < 0");
7749886d
MD
2498 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2499 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2500 ) {
984263bc
MD
2501 lf.l_whence = SEEK_SET;
2502 lf.l_start = 0;
2503 lf.l_len = 0;
2504 lf.l_type = F_UNLCK;
2505 vp = (struct vnode *)fp->f_data;
71c18fe3 2506 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
984263bc
MD
2507 }
2508 if (fp->f_ops != &badfileops)
87de5057 2509 error = fo_close(fp);
984263bc
MD
2510 else
2511 error = 0;
2512 ffree(fp);
2513 return (error);
2514}
2515
2516/*
2517 * Apply an advisory lock on a file descriptor.
2518 *
2519 * Just attempt to get a record lock of the requested type on
2520 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3919ced0
MD
2521 *
2522 * MPALMOSTSAFE
984263bc 2523 */
984263bc 2524int
753fd850 2525sys_flock(struct flock_args *uap)
984263bc 2526{
41c20dac 2527 struct proc *p = curproc;
41c20dac 2528 struct file *fp;
984263bc
MD
2529 struct vnode *vp;
2530 struct flock lf;
5b287bba 2531 int error;
984263bc 2532
5b287bba 2533 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
984263bc 2534 return (EBADF);
5b287bba
MD
2535 if (fp->f_type != DTYPE_VNODE) {
2536 error = EOPNOTSUPP;
2537 goto done;
2538 }
984263bc
MD
2539 vp = (struct vnode *)fp->f_data;
2540 lf.l_whence = SEEK_SET;
2541 lf.l_start = 0;
2542 lf.l_len = 0;
2543 if (uap->how & LOCK_UN) {
2544 lf.l_type = F_UNLCK;
2545 fp->f_flag &= ~FHASLOCK;
5b287bba
MD
2546 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2547 goto done;
984263bc
MD
2548 }
2549 if (uap->how & LOCK_EX)
2550 lf.l_type = F_WRLCK;
2551 else if (uap->how & LOCK_SH)
2552 lf.l_type = F_RDLCK;
5b287bba
MD
2553 else {
2554 error = EBADF;
2555 goto done;
2556 }
984263bc
MD
2557 fp->f_flag |= FHASLOCK;
2558 if (uap->how & LOCK_NB)
5b287bba
MD
2559 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2560 else
2561 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2562done:
2563 fdrop(fp);
2564 return (error);
984263bc
MD
2565}
2566
2567/*
2568 * File Descriptor pseudo-device driver (/dev/fd/).
2569 *
2570 * Opening minor device N dup()s the file (if any) connected to file
2571 * descriptor N belonging to the calling process. Note that this driver
2572 * consists of only the ``open()'' routine, because all subsequent
2573 * references to this file will be direct to the other driver.
2574 */
984263bc 2575static int
fef8985e 2576fdopen(struct dev_open_args *ap)
984263bc 2577{
fef8985e
MD
2578 thread_t td = curthread;
2579
fbfe4e7d 2580 KKASSERT(td->td_lwp != NULL);
984263bc
MD
2581
2582 /*
fbfe4e7d 2583 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
984263bc
MD
2584 * the file descriptor being sought for duplication. The error
2585 * return ensures that the vnode for this device will be released
2586 * by vn_open. Open will detect this special error and take the
2587 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2588 * will simply report the error.
2589 */
fef8985e 2590 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
984263bc
MD
2591 return (ENODEV);
2592}
2593
2594/*
259b8ea0
MD
2595 * The caller has reserved the file descriptor dfd for us. On success we
2596 * must fsetfd() it. On failure the caller will clean it up.
5b287bba 2597 *
aac0aabd 2598 * MPSAFE
984263bc
MD
2599 */
2600int
f3a2d8c4 2601dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
984263bc 2602{
41c20dac 2603 struct file *wfp;
259b8ea0 2604 struct file *xfp;
ea8f324c 2605 int werror;
984263bc 2606
259b8ea0 2607 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
984263bc 2608 return (EBADF);
984263bc 2609
ea8f324c
MD
2610 /*
2611 * Close a revoke/dup race. Duping a descriptor marked as revoked
2612 * will dup a dummy descriptor instead of the real one.
2613 */
2614 if (wfp->f_flag & FREVOKED) {
2615 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2616 fdrop(wfp);
2617 wfp = NULL;
2618 werror = falloc(NULL, &wfp, NULL);
2619 if (werror)
2620 return (werror);
2621 }
2622
984263bc
MD
2623 /*
2624 * There are two cases of interest here.
2625 *
259b8ea0 2626 * For ENODEV simply dup sfd to file descriptor dfd and return.
984263bc 2627 *
259b8ea0
MD
2628 * For ENXIO steal away the file structure from sfd and store it
2629 * dfd. sfd is effectively closed by this operation.
984263bc
MD
2630 *
2631 * Any other error code is just returned.
2632 */
2633 switch (error) {
2634 case ENODEV:
2635 /*
2636 * Check that the mode the file is being opened for is a
2637 * subset of the mode of the existing descriptor.
2638 */
ea8f324c
MD
2639 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2640 error = EACCES;
2641 break;
2642 }
287a8577 2643 spin_lock(&fdp->fd_spin);
259b8ea0 2644 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
aac0aabd 2645 fsetfd_locked(fdp, wfp, dfd);
287a8577 2646 spin_unlock(&fdp->fd_spin);
259b8ea0
MD
2647 error = 0;
2648 break;
984263bc
MD
2649 case ENXIO:
2650 /*
2651 * Steal away the file pointer from dfd, and stuff it into indx.
2652 */
287a8577 2653 spin_lock(&fdp->fd_spin);
259b8ea0 2654 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
f3a2d8c4 2655 fsetfd(fdp, wfp, dfd);
aac0aabd 2656 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
287a8577 2657 spin_unlock(&fdp->fd_spin);
259b8ea0 2658 fdrop(xfp);
aac0aabd 2659 } else {
287a8577 2660 spin_unlock(&fdp->fd_spin);
aac0aabd 2661 }
259b8ea0
MD
2662 error = 0;
2663 break;
984263bc 2664 default:
259b8ea0 2665 break;
984263bc 2666 }
259b8ea0
MD
2667 fdrop(wfp);
2668 return (error);
984263bc
MD
2669}
2670
5b287bba
MD
2671/*
2672 * NOT MPSAFE - I think these refer to a common file descriptor table
2673 * and we need to spinlock that to link fdtol in.
2674 */
984263bc
MD
2675struct filedesc_to_leader *
2676filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2677 struct proc *leader)
2678{
2679 struct filedesc_to_leader *fdtol;
2680
77652cad 2681 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
b5c4d81f 2682 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO);
984263bc
MD
2683 fdtol->fdl_refcount = 1;
2684 fdtol->fdl_holdcount = 0;
2685 fdtol->fdl_wakeup = 0;
2686 fdtol->fdl_leader = leader;
2687 if (old != NULL) {
2688 fdtol->fdl_next = old->fdl_next;
2689 fdtol->fdl_prev = old;
2690 old->fdl_next = fdtol;
2691 fdtol->fdl_next->fdl_prev = fdtol;
2692 } else {
2693 fdtol->fdl_next = fdtol;
2694 fdtol->fdl_prev = fdtol;
2695 }
2696 return fdtol;
2697}
2698
2dd63755
MD
2699/*
2700 * Scan all file pointers in the system. The callback is made with
ea8f324c 2701 * the master list spinlock held exclusively.
2dd63755
MD
2702 *
2703 * MPSAFE
2dd63755
MD
2704 */
2705void
2706allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2707{
2708 struct file *fp;
2709 int res;
2710
287a8577 2711 spin_lock(&filehead_spin);
2dd63755 2712 LIST_FOREACH(fp, &filehead, f_list) {
2dd63755 2713 res = callback(fp, data);
2dd63755
MD
2714 if (res < 0)
2715 break;
2716 }
287a8577 2717 spin_unlock(&filehead_spin);
2dd63755
MD
2718}
2719
984263bc
MD
2720/*
2721 * Get file structures.
5b287bba
MD
2722 *
2723 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
984263bc 2724 */
8fa76237
MD
2725
2726struct sysctl_kern_file_info {
2727 int count;
2728 int error;
2729 struct sysctl_req *req;
2730};
2731
2732static int sysctl_kern_file_callback(struct proc *p, void *data);
2733
984263bc
MD
2734static int
2735sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2736{
8fa76237 2737 struct sysctl_kern_file_info info;
984263bc
MD
2738
2739 /*
7b124c9f
JS
2740 * Note: because the number of file descriptors is calculated
2741 * in different ways for sizing vs returning the data,
2742 * there is information leakage from the first loop. However,
2743 * it is of a similar order of magnitude to the leakage from
2744 * global system statistics such as kern.openfiles.
6d132b4d
MD
2745 *
2746 * When just doing a count, note that we cannot just count
2747 * the elements and add f_count via the filehead list because
2748 * threaded processes share their descriptor table and f_count might
2749 * still be '1' in that case.
e38910da
MD
2750 *
2751 * Since the SYSCTL op can block, we must hold the process to
2752 * prevent it being ripped out from under us either in the
2753 * file descriptor loop or in the greater LIST_FOREACH. The
2754 * process may be in varying states of disrepair. If the process
2755 * is in SZOMB we may have caught it just as it is being removed
2756 * from the allproc list, we must skip it in that case to maintain
2757 * an unbroken chain through the allproc list.
984263bc 2758 */
8fa76237
MD
2759 info.count = 0;
2760 info.error = 0;
2761 info.req = req;
2762 allproc_scan(sysctl_kern_file_callback, &info);
6d132b4d
MD
2763
2764 /*
2765 * When just calculating the size, overestimate a bit to try to
2766 * prevent system activity from causing the buffer-fill call
2767 * to fail later on.
2768 */
2769 if (req->oldptr == NULL) {
8fa76237
MD
2770 info.count = (info.count + 16) + (info.count / 10);
2771 info.error = SYSCTL_OUT(req, NULL,
2772 info.count * sizeof(struct kinfo_file));
6d132b4d 2773 }
8fa76237
MD
2774 return (info.error);
2775}
2776
2777static int
2778sysctl_kern_file_callback(struct proc *p, void *data)
2779{
2780 struct sysctl_kern_file_info *info = data;
2781 struct kinfo_file kf;
2782 struct filedesc *fdp;
2783 struct file *fp;
2784 uid_t uid;
2785 int n;
2786
416d05d7 2787 if (p->p_stat == SIDL || p->p_stat == SZOMB)
8fa76237 2788 return(0);
9910d07b 2789 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
8fa76237 2790 return(0);
0a4a9c77
MD
2791
2792 /*
2793 * Softref the fdp to prevent it from being destroyed
2794 */
287a8577 2795 spin_lock(&p->p_spin);
0a4a9c77 2796 if ((fdp = p->p_fd) == NULL) {
287a8577 2797 spin_unlock(&p->p_spin);
8fa76237 2798 return(0);
0a4a9c77
MD
2799 }
2800 atomic_add_int(&fdp->fd_softrefs, 1);
287a8577 2801 spin_unlock(&p->p_spin);
0a4a9c77
MD
2802
2803 /*
2804 * The fdp's own spinlock prevents the contents from being
2805 * modified.
2806 */
0846e4ce 2807 spin_lock_shared(&fdp->fd_spin);
8fa76237
MD
2808 for (n = 0; n < fdp->fd_nfiles; ++n) {
2809 if ((fp = fdp->fd_files[n].fp) == NULL)
2810 continue;
2811 if (info->req->oldptr == NULL) {
2812 ++info->count;
2813 } else {
2814 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2815 kcore_make_file(&kf, fp, p->p_pid, uid, n);
0846e4ce 2816 spin_unlock_shared(&fdp->fd_spin);
8fa76237 2817 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
0846e4ce 2818 spin_lock_shared(&fdp->fd_spin);
8fa76237
MD
2819 if (info->error)
2820 break;
2821 }
2822 }
0846e4ce 2823 spin_unlock_shared(&fdp->fd_spin);
0a4a9c77 2824 atomic_subtract_int(&fdp->fd_softrefs, 1);
8fa76237
MD
2825 if (info->error)
2826 return(-1);
2827 return(0);
984263bc
MD
2828}
2829
2830SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2831 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2832
d37c8f7f
MD
2833SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2834 &minfilesperproc, 0, "Minimum files allowed open per process");
984263bc
MD
2835SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2836 &maxfilesperproc, 0, "Maximum files allowed open per process");
d37c8f7f
MD
2837SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2838 &maxfilesperuser, 0, "Maximum files allowed open per user");
984263bc
MD
2839
2840SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2841 &maxfiles, 0, "Maximum number of files");
2842
60ee93b9
MD
2843SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2844 &maxfilesrootres, 0, "Descriptors reserved for root use");
2845
984263bc
MD
2846SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2847 &nfiles, 0, "System-wide number of open files");
2848
2849static void
2850fildesc_drvinit(void *unused)
2851{
2852 int fd;
2853
e4c9c0c8 2854 for (fd = 0; fd < NUMFDESC; fd++) {
fef8985e 2855 make_dev(&fildesc_ops, fd,
3e82b46c 2856 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
e4c9c0c8 2857 }
cd29885a 2858
fef8985e
MD
2859 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2860 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2861 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
984263bc
MD
2862}
2863
d9b2033e
MD
2864/*
2865 * MPSAFE
2866 */
984263bc 2867struct fileops badfileops = {
b2d248cb
MD
2868 .fo_read = badfo_readwrite,
2869 .fo_write = badfo_readwrite,
2870 .fo_ioctl = badfo_ioctl,
b2d248cb
MD
2871 .fo_kqfilter = badfo_kqfilter,
2872 .fo_stat = badfo_stat,
2873 .fo_close = badfo_close,
2874 .fo_shutdown = badfo_shutdown
984263bc
MD
2875};
2876
a724d72b 2877int
dadab5e9
MD
2878badfo_readwrite(
2879 struct file *fp,
2880 struct uio *uio,
2881 struct ucred *cred,
87de5057 2882 int flags
dadab5e9 2883) {
984263bc
MD
2884 return (EBADF);
2885}
2886
a724d72b 2887int
87baaf0c
MD
2888badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2889 struct ucred *cred, struct sysmsg *msgv)
984263bc 2890{
984263bc
MD
2891 return (EBADF);
2892}
2893
d9b2033e 2894/*
304d60e8
MD
2895 * Must return an error to prevent registration, typically
2896 * due to a revoked descriptor (file_filtops assigned).
d9b2033e 2897 */
a724d72b 2898int
dadab5e9 2899badfo_kqfilter(struct file *fp, struct knote *kn)
984263bc 2900{
304d60e8 2901 return (EOPNOTSUPP);
984263bc
MD
2902}
2903
304d60e8
MD
2904/*
2905 * MPSAFE
2906 */
a724d72b 2907int
87de5057 2908badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
984263bc 2909{
984263bc
MD
2910 return (EBADF);
2911}
2912
d9b2033e
MD
2913/*
2914 * MPSAFE
2915 */
a724d72b 2916int
87de5057 2917badfo_close(struct file *fp)
984263bc 2918{
984263bc
MD
2919 return (EBADF);
2920}
2921
d9b2033e
MD
2922/*
2923 * MPSAFE
2924 */
a724d72b 2925int
87de5057 2926badfo_shutdown(struct file *fp, int how)
004d2de5
MD
2927{
2928 return (EBADF);
2929}
2930
d9b2033e
MD
2931/*
2932 * MPSAFE
2933 */
004d2de5 2934int
87de5057 2935nofo_shutdown(struct file *fp, int how)
004d2de5
MD
2936{
2937 return (EOPNOTSUPP);
2938}
2939
984263bc
MD
2940SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2941 fildesc_drvinit,NULL)