Add kernel-layer support for chflags checks, remove (most) from the VFS layer.
[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.
51 * 3. All advertising materials mentioning features or use of this software
52 * must display the following acknowledgement:
53 * This product includes software developed by the University of
54 * California, Berkeley and its contributors.
55 * 4. Neither the name of the University nor the names of its contributors
56 * may be used to endorse or promote products derived from this software
57 * without specific prior written permission.
58 *
59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69 * SUCH DAMAGE.
70 *
71 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
c4cb6d8b 72 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
8f9f2535 73 * $DragonFly: src/sys/kern/kern_descrip.c,v 1.79 2008/08/31 13:18:28 aggelos Exp $
984263bc
MD
74 */
75
76#include "opt_compat.h"
77#include <sys/param.h>
78#include <sys/systm.h>
79#include <sys/malloc.h>
80#include <sys/sysproto.h>
81#include <sys/conf.h>
fef8985e 82#include <sys/device.h>
984263bc
MD
83#include <sys/filedesc.h>
84#include <sys/kernel.h>
85#include <sys/sysctl.h>
86#include <sys/vnode.h>
87#include <sys/proc.h>
fad57d0e 88#include <sys/nlookup.h>
984263bc
MD
89#include <sys/file.h>
90#include <sys/stat.h>
91#include <sys/filio.h>
92#include <sys/fcntl.h>
93#include <sys/unistd.h>
94#include <sys/resourcevar.h>
95#include <sys/event.h>
dda4b42b 96#include <sys/kern_syscall.h>
1c55bd1c 97#include <sys/kcore.h>
7b124c9f 98#include <sys/kinfo.h>
ea8f324c 99#include <sys/un.h>
984263bc
MD
100
101#include <vm/vm.h>
102#include <vm/vm_extern.h>
103
e43a034f 104#include <sys/thread2.h>
dadab5e9 105#include <sys/file2.h>
85fceac1 106#include <sys/spinlock2.h>
dadab5e9 107
5b287bba
MD
108static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
109static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
110static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
85fceac1 111static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp);
5b287bba 112static void ffree(struct file *fp);
fa541be6 113
984263bc
MD
114static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
115static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
116 "file desc to leader structures");
117MALLOC_DEFINE(M_FILE, "file", "Open file structure");
118static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
119
120static d_open_t fdopen;
121#define NUMFDESC 64
122
123#define CDEV_MAJOR 22
fef8985e
MD
124static struct dev_ops fildesc_ops = {
125 { "FD", CDEV_MAJOR, 0 },
126 .d_open = fdopen,
984263bc
MD
127};
128
402ed7e1 129static int badfo_readwrite (struct file *fp, struct uio *uio,
87de5057 130 struct ucred *cred, int flags);
402ed7e1 131static int badfo_ioctl (struct file *fp, u_long com, caddr_t data,
87de5057
MD
132 struct ucred *cred);
133static int badfo_poll (struct file *fp, int events, struct ucred *cred);
402ed7e1 134static int badfo_kqfilter (struct file *fp, struct knote *kn);
87de5057
MD
135static int badfo_stat (struct file *fp, struct stat *sb, struct ucred *cred);
136static int badfo_close (struct file *fp);
137static int badfo_shutdown (struct file *fp, int how);
984263bc
MD
138
139/*
140 * Descriptor management.
141 */
2dd63755
MD
142static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
143static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
144static int nfiles; /* actual number of open files */
984263bc
MD
145extern int cmask;
146
147/*
8247b2f9 148 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
5b287bba
MD
149 *
150 * MPSAFE - must be called with fdp->fd_spin exclusively held
8247b2f9
MD
151 */
152static __inline
153void
5b287bba 154fdfixup_locked(struct filedesc *fdp, int fd)
8247b2f9
MD
155{
156 if (fd < fdp->fd_freefile) {
157 fdp->fd_freefile = fd;
158 }
159 while (fdp->fd_lastfile >= 0 &&
160 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
161 fdp->fd_files[fdp->fd_lastfile].reserved == 0
162 ) {
163 --fdp->fd_lastfile;
164 }
165}
166
167/*
984263bc 168 * System calls on descriptors.
5b287bba
MD
169 *
170 * MPSAFE
984263bc 171 */
984263bc 172int
753fd850 173sys_getdtablesize(struct getdtablesize_args *uap)
984263bc 174{
41c20dac 175 struct proc *p = curproc;
5b287bba 176 struct plimit *limit = p->p_limit;
984263bc 177
5b287bba 178 spin_lock_rd(&limit->p_spin);
c7114eea 179 uap->sysmsg_result =
5b287bba
MD
180 min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
181 spin_unlock_rd(&limit->p_spin);
984263bc
MD
182 return (0);
183}
184
185/*
186 * Duplicate a file descriptor to a particular value.
187 *
188 * note: keep in mind that a potential race condition exists when closing
189 * descriptors from a shared descriptor table (via rfork).
5b287bba
MD
190 *
191 * MPSAFE
984263bc 192 */
984263bc 193int
753fd850 194sys_dup2(struct dup2_args *uap)
984263bc 195{
dda4b42b
DRJ
196 int error;
197
198 error = kern_dup(DUP_FIXED, uap->from, uap->to, uap->sysmsg_fds);
199
200 return (error);
984263bc
MD
201}
202
203/*
204 * Duplicate a file descriptor.
5b287bba
MD
205 *
206 * MPSAFE
984263bc 207 */
984263bc 208int
753fd850 209sys_dup(struct dup_args *uap)
984263bc 210{
dda4b42b 211 int error;
984263bc 212
dda4b42b
DRJ
213 error = kern_dup(DUP_VARIABLE, uap->fd, 0, uap->sysmsg_fds);
214
215 return (error);
984263bc
MD
216}
217
5b287bba
MD
218/*
219 * MPALMOSTSAFE - acquires mplock for fp operations
220 */
984263bc 221int
87de5057 222kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
984263bc 223{
dadab5e9
MD
224 struct thread *td = curthread;
225 struct proc *p = td->td_proc;
41c20dac 226 struct file *fp;
984263bc 227 struct vnode *vp;
984263bc 228 u_int newmin;
9ba76b73 229 u_int oflags;
3a907475 230 u_int nflags;
dda4b42b 231 int tmp, error, flg = F_POSIX;
984263bc 232
dadab5e9
MD
233 KKASSERT(p);
234
5b287bba
MD
235 /*
236 * Operations on file descriptors that do not require a file pointer.
237 */
dda4b42b 238 switch (cmd) {
984263bc 239 case F_GETFD:
85fceac1
MD
240 error = fgetfdflags(p->p_fd, fd, &tmp);
241 if (error == 0)
242 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
5b287bba 243 return (error);
984263bc
MD
244
245 case F_SETFD:
85fceac1
MD
246 if (dat->fc_cloexec & FD_CLOEXEC)
247 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
248 else
249 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
5b287bba
MD
250 return (error);
251 case F_DUPFD:
252 newmin = dat->fc_fd;
253 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
254 return (error);
255 default:
85fceac1 256 break;
5b287bba 257 }
984263bc 258
5b287bba
MD
259 /*
260 * Operations on file pointers
261 */
262 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
263 return (EBADF);
264
265 get_mplock();
266 switch (cmd) {
984263bc 267 case F_GETFL:
dda4b42b 268 dat->fc_flags = OFLAGS(fp->f_flag);
85fceac1
MD
269 error = 0;
270 break;
984263bc
MD
271
272 case F_SETFL:
3a907475
MD
273 oflags = fp->f_flag;
274 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
275 nflags |= oflags & ~FCNTLFLAGS;
276
f58b505b 277 error = 0;
3a907475
MD
278 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
279 error = EINVAL;
280 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
f58b505b
MD
281 tmp = fp->f_flag & FASYNC;
282 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, cred);
283 }
3a907475
MD
284 if (error == 0)
285 fp->f_flag = nflags;
85fceac1 286 break;
984263bc
MD
287
288 case F_GETOWN:
87de5057 289 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, cred);
85fceac1 290 break;
984263bc
MD
291
292 case F_SETOWN:
87de5057 293 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, cred);
85fceac1 294 break;
984263bc
MD
295
296 case F_SETLKW:
297 flg |= F_WAIT;
298 /* Fall into F_SETLK */
299
300 case F_SETLK:
85fceac1
MD
301 if (fp->f_type != DTYPE_VNODE) {
302 error = EBADF;
303 break;
304 }
984263bc
MD
305 vp = (struct vnode *)fp->f_data;
306
307 /*
308 * copyin/lockop may block
309 */
dda4b42b
DRJ
310 if (dat->fc_flock.l_whence == SEEK_CUR)
311 dat->fc_flock.l_start += fp->f_offset;
984263bc 312
dda4b42b 313 switch (dat->fc_flock.l_type) {
984263bc
MD
314 case F_RDLCK:
315 if ((fp->f_flag & FREAD) == 0) {
316 error = EBADF;
317 break;
318 }
319 p->p_leader->p_flag |= P_ADVLOCK;
320 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
dda4b42b 321 &dat->fc_flock, flg);
984263bc
MD
322 break;
323 case F_WRLCK:
324 if ((fp->f_flag & FWRITE) == 0) {
325 error = EBADF;
326 break;
327 }
328 p->p_leader->p_flag |= P_ADVLOCK;
329 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
dda4b42b 330 &dat->fc_flock, flg);
984263bc
MD
331 break;
332 case F_UNLCK:
333 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
dda4b42b 334 &dat->fc_flock, F_POSIX);
984263bc
MD
335 break;
336 default:
337 error = EINVAL;
338 break;
339 }
85fceac1
MD
340
341 /*
342 * It is possible to race a close() on the descriptor while
343 * we were blocked getting the lock. If this occurs the
344 * close might not have caught the lock.
345 */
346 if (checkfpclosed(p->p_fd, fd, fp)) {
dda4b42b
DRJ
347 dat->fc_flock.l_whence = SEEK_SET;
348 dat->fc_flock.l_start = 0;
349 dat->fc_flock.l_len = 0;
350 dat->fc_flock.l_type = F_UNLCK;
984263bc 351 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
dda4b42b 352 F_UNLCK, &dat->fc_flock, F_POSIX);
984263bc 353 }
85fceac1 354 break;
984263bc
MD
355
356 case F_GETLK:
85fceac1
MD
357 if (fp->f_type != DTYPE_VNODE) {
358 error = EBADF;
359 break;
360 }
984263bc
MD
361 vp = (struct vnode *)fp->f_data;
362 /*
363 * copyin/lockop may block
364 */
dda4b42b
DRJ
365 if (dat->fc_flock.l_type != F_RDLCK &&
366 dat->fc_flock.l_type != F_WRLCK &&
367 dat->fc_flock.l_type != F_UNLCK) {
85fceac1
MD
368 error = EINVAL;
369 break;
984263bc 370 }
dda4b42b
DRJ
371 if (dat->fc_flock.l_whence == SEEK_CUR)
372 dat->fc_flock.l_start += fp->f_offset;
984263bc 373 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
dda4b42b 374 &dat->fc_flock, F_POSIX);
85fceac1 375 break;
984263bc 376 default:
85fceac1
MD
377 error = EINVAL;
378 break;
984263bc 379 }
5b287bba
MD
380 rel_mplock();
381
85fceac1
MD
382 fdrop(fp);
383 return (error);
984263bc
MD
384}
385
386/*
dda4b42b 387 * The file control system call.
5b287bba
MD
388 *
389 * MPSAFE
dda4b42b
DRJ
390 */
391int
753fd850 392sys_fcntl(struct fcntl_args *uap)
dda4b42b
DRJ
393{
394 union fcntl_dat dat;
395 int error;
396
397 switch (uap->cmd) {
398 case F_DUPFD:
399 dat.fc_fd = uap->arg;
400 break;
401 case F_SETFD:
402 dat.fc_cloexec = uap->arg;
403 break;
404 case F_SETFL:
405 dat.fc_flags = uap->arg;
406 break;
407 case F_SETOWN:
408 dat.fc_owner = uap->arg;
409 break;
410 case F_SETLKW:
411 case F_SETLK:
412 case F_GETLK:
413 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
5b287bba 414 sizeof(struct flock));
dda4b42b
DRJ
415 if (error)
416 return (error);
417 break;
418 }
419
87de5057 420 error = kern_fcntl(uap->fd, uap->cmd, &dat, curproc->p_ucred);
dda4b42b
DRJ
421
422 if (error == 0) {
423 switch (uap->cmd) {
424 case F_DUPFD:
425 uap->sysmsg_result = dat.fc_fd;
426 break;
427 case F_GETFD:
428 uap->sysmsg_result = dat.fc_cloexec;
429 break;
430 case F_GETFL:
431 uap->sysmsg_result = dat.fc_flags;
432 break;
433 case F_GETOWN:
434 uap->sysmsg_result = dat.fc_owner;
435 case F_GETLK:
436 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
437 sizeof(struct flock));
438 break;
439 }
440 }
441
442 return (error);
443}
444
445/*
984263bc 446 * Common code for dup, dup2, and fcntl(F_DUPFD).
dda4b42b
DRJ
447 *
448 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
449 * kern_dup() to destructively dup over an existing file descriptor if new
450 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
451 * unused file descriptor that is greater than or equal to new.
5b287bba
MD
452 *
453 * MPSAFE
984263bc 454 */
dda4b42b
DRJ
455int
456kern_dup(enum dup_type type, int old, int new, int *res)
984263bc 457{
dda4b42b
DRJ
458 struct thread *td = curthread;
459 struct proc *p = td->td_proc;
460 struct filedesc *fdp = p->p_fd;
984263bc
MD
461 struct file *fp;
462 struct file *delfp;
259b8ea0 463 int oldflags;
984263bc 464 int holdleaders;
dda4b42b
DRJ
465 int error, newfd;
466
467 /*
468 * Verify that we have a valid descriptor to dup from and
469 * possibly to dup to.
470 */
259b8ea0 471retry:
5b287bba
MD
472 spin_lock_wr(&fdp->fd_spin);
473 if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
474 new >= maxfilesperproc) {
475 spin_unlock_wr(&fdp->fd_spin);
476 return (EINVAL);
477 }
478 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
479 spin_unlock_wr(&fdp->fd_spin);
dda4b42b 480 return (EBADF);
5b287bba 481 }
dda4b42b
DRJ
482 if (type == DUP_FIXED && old == new) {
483 *res = new;
5b287bba 484 spin_unlock_wr(&fdp->fd_spin);
dda4b42b
DRJ
485 return (0);
486 }
0679adc4 487 fp = fdp->fd_files[old].fp;
259b8ea0 488 oldflags = fdp->fd_files[old].fileflags;
5b287bba 489 fhold(fp); /* MPSAFE - can be called with a spinlock held */
dda4b42b
DRJ
490
491 /*
259b8ea0
MD
492 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
493 * if the requested descriptor is beyond the current table size.
494 *
495 * This can block. Retry if the source descriptor no longer matches
496 * or if our expectation in the expansion case races.
497 *
498 * If we are not expanding or allocating a new decriptor, then reset
499 * the target descriptor to a reserved state so we have a uniform
500 * setup for the next code block.
dda4b42b
DRJ
501 */
502 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
5b287bba 503 spin_unlock_wr(&fdp->fd_spin);
dda4b42b 504 error = fdalloc(p, new, &newfd);
5b287bba 505 spin_lock_wr(&fdp->fd_spin);
dda4b42b 506 if (error) {
5b287bba 507 spin_unlock_wr(&fdp->fd_spin);
9f87144f 508 fdrop(fp);
dda4b42b
DRJ
509 return (error);
510 }
5b287bba
MD
511 /*
512 * Check for ripout
513 */
259b8ea0 514 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
5b287bba
MD
515 fsetfd_locked(fdp, NULL, newfd);
516 spin_unlock_wr(&fdp->fd_spin);
259b8ea0
MD
517 fdrop(fp);
518 goto retry;
519 }
5b287bba
MD
520 /*
521 * Check for expansion race
522 */
259b8ea0 523 if (type != DUP_VARIABLE && new != newfd) {
5b287bba
MD
524 fsetfd_locked(fdp, NULL, newfd);
525 spin_unlock_wr(&fdp->fd_spin);
259b8ea0
MD
526 fdrop(fp);
527 goto retry;
528 }
5b287bba
MD
529 /*
530 * Check for ripout, newfd reused old (this case probably
531 * can't occur).
532 */
259b8ea0 533 if (old == newfd) {
5b287bba
MD
534 fsetfd_locked(fdp, NULL, newfd);
535 spin_unlock_wr(&fdp->fd_spin);
259b8ea0
MD
536 fdrop(fp);
537 goto retry;
538 }
dda4b42b 539 new = newfd;
259b8ea0
MD
540 delfp = NULL;
541 } else {
542 if (fdp->fd_files[new].reserved) {
5b287bba 543 spin_unlock_wr(&fdp->fd_spin);
259b8ea0 544 fdrop(fp);
6ea70f76 545 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
259b8ea0
MD
546 tsleep(fdp, 0, "fdres", hz);
547 goto retry;
548 }
dda4b42b 549
259b8ea0
MD
550 /*
551 * If the target descriptor was never allocated we have
552 * to allocate it. If it was we have to clean out the
5b287bba
MD
553 * old descriptor. delfp inherits the ref from the
554 * descriptor table.
259b8ea0
MD
555 */
556 delfp = fdp->fd_files[new].fp;
557 fdp->fd_files[new].fp = NULL;
558 fdp->fd_files[new].reserved = 1;
559 if (delfp == NULL) {
5b287bba 560 fdreserve_locked(fdp, new, 1);
259b8ea0
MD
561 if (new > fdp->fd_lastfile)
562 fdp->fd_lastfile = new;
dda4b42b 563 }
259b8ea0 564
dda4b42b 565 }
984263bc
MD
566
567 /*
5b287bba
MD
568 * NOTE: still holding an exclusive spinlock
569 */
570
571 /*
259b8ea0
MD
572 * If a descriptor is being overwritten we may hve to tell
573 * fdfree() to sleep to ensure that all relevant process
574 * leaders can be traversed in closef().
984263bc 575 */
984263bc 576 if (delfp != NULL && p->p_fdtol != NULL) {
984263bc
MD
577 fdp->fd_holdleaderscount++;
578 holdleaders = 1;
259b8ea0 579 } else {
984263bc 580 holdleaders = 0;
259b8ea0 581 }
dda4b42b 582 KASSERT(delfp == NULL || type == DUP_FIXED,
259b8ea0 583 ("dup() picked an open file"));
984263bc
MD
584
585 /*
259b8ea0
MD
586 * Duplicate the source descriptor, update lastfile. If the new
587 * descriptor was not allocated and we aren't replacing an existing
588 * descriptor we have to mark the descriptor as being in use.
589 *
590 * The fd_files[] array inherits fp's hold reference.
984263bc 591 */
5b287bba 592 fsetfd_locked(fdp, fp, new);
259b8ea0 593 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
5b287bba
MD
594 spin_unlock_wr(&fdp->fd_spin);
595 fdrop(fp);
dda4b42b 596 *res = new;
984263bc
MD
597
598 /*
599 * If we dup'd over a valid file, we now own the reference to it
600 * and must dispose of it using closef() semantics (as if a
601 * close() were performed on it).
602 */
603 if (delfp) {
b8477cda 604 closef(delfp, p);
984263bc 605 if (holdleaders) {
5b287bba 606 spin_lock_wr(&fdp->fd_spin);
984263bc
MD
607 fdp->fd_holdleaderscount--;
608 if (fdp->fd_holdleaderscount == 0 &&
609 fdp->fd_holdleaderswakeup != 0) {
610 fdp->fd_holdleaderswakeup = 0;
5b287bba 611 spin_unlock_wr(&fdp->fd_spin);
984263bc 612 wakeup(&fdp->fd_holdleaderscount);
5b287bba
MD
613 } else {
614 spin_unlock_wr(&fdp->fd_spin);
984263bc
MD
615 }
616 }
617 }
618 return (0);
619}
620
621/*
622 * If sigio is on the list associated with a process or process group,
623 * disable signalling from the device, remove sigio from the list and
624 * free sigio.
625 */
626void
7bf8660a 627funsetown(struct sigio *sigio)
984263bc 628{
984263bc
MD
629 if (sigio == NULL)
630 return;
e43a034f 631 crit_enter();
984263bc 632 *(sigio->sio_myref) = NULL;
e43a034f 633 crit_exit();
984263bc
MD
634 if (sigio->sio_pgid < 0) {
635 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
636 sigio, sio_pgsigio);
637 } else /* if ((*sigiop)->sio_pgid > 0) */ {
638 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
639 sigio, sio_pgsigio);
640 }
641 crfree(sigio->sio_ucred);
efda3bd0 642 kfree(sigio, M_SIGIO);
984263bc
MD
643}
644
645/* Free a list of sigio structures. */
646void
7bf8660a 647funsetownlst(struct sigiolst *sigiolst)
984263bc
MD
648{
649 struct sigio *sigio;
650
651 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
652 funsetown(sigio);
653}
654
655/*
656 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
657 *
658 * After permission checking, add a sigio structure to the sigio list for
659 * the process or process group.
660 */
661int
7bf8660a 662fsetown(pid_t pgid, struct sigio **sigiop)
984263bc
MD
663{
664 struct proc *proc;
665 struct pgrp *pgrp;
666 struct sigio *sigio;
984263bc
MD
667
668 if (pgid == 0) {
669 funsetown(*sigiop);
670 return (0);
671 }
672 if (pgid > 0) {
673 proc = pfind(pgid);
674 if (proc == NULL)
675 return (ESRCH);
676
677 /*
678 * Policy - Don't allow a process to FSETOWN a process
679 * in another session.
680 *
681 * Remove this test to allow maximum flexibility or
682 * restrict FSETOWN to the current process or process
683 * group for maximum safety.
684 */
685 if (proc->p_session != curproc->p_session)
686 return (EPERM);
687
688 pgrp = NULL;
689 } else /* if (pgid < 0) */ {
690 pgrp = pgfind(-pgid);
691 if (pgrp == NULL)
692 return (ESRCH);
693
694 /*
695 * Policy - Don't allow a process to FSETOWN a process
696 * in another session.
697 *
698 * Remove this test to allow maximum flexibility or
699 * restrict FSETOWN to the current process or process
700 * group for maximum safety.
701 */
702 if (pgrp->pg_session != curproc->p_session)
703 return (EPERM);
704
705 proc = NULL;
706 }
707 funsetown(*sigiop);
efda3bd0 708 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
984263bc
MD
709 if (pgid > 0) {
710 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
711 sigio->sio_proc = proc;
712 } else {
713 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
714 sigio->sio_pgrp = pgrp;
715 }
716 sigio->sio_pgid = pgid;
e9a372eb 717 sigio->sio_ucred = crhold(curproc->p_ucred);
984263bc 718 /* It would be convenient if p_ruid was in ucred. */
41c20dac 719 sigio->sio_ruid = curproc->p_ucred->cr_ruid;
984263bc 720 sigio->sio_myref = sigiop;
e43a034f 721 crit_enter();
984263bc 722 *sigiop = sigio;
e43a034f 723 crit_exit();
984263bc
MD
724 return (0);
725}
726
727/*
728 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
729 */
730pid_t
7bf8660a 731fgetown(struct sigio *sigio)
984263bc
MD
732{
733 return (sigio != NULL ? sigio->sio_pgid : 0);
734}
735
736/*
4336d5df 737 * Close many file descriptors.
5b287bba
MD
738 *
739 * MPSAFE
4336d5df 740 */
4336d5df 741int
753fd850 742sys_closefrom(struct closefrom_args *uap)
4336d5df
JS
743{
744 return(kern_closefrom(uap->fd));
745}
746
5b287bba
MD
747/*
748 * Close all file descriptors greater then or equal to fd
749 *
750 * MPSAFE
751 */
4336d5df
JS
752int
753kern_closefrom(int fd)
754{
755 struct thread *td = curthread;
756 struct proc *p = td->td_proc;
757 struct filedesc *fdp;
4336d5df
JS
758
759 KKASSERT(p);
760 fdp = p->p_fd;
761
5b287bba
MD
762 if (fd < 0)
763 return (EINVAL);
5e8cfac8 764
259b8ea0
MD
765 /*
766 * NOTE: This function will skip unassociated descriptors and
767 * reserved descriptors that have not yet been assigned.
768 * fd_lastfile can change as a side effect of kern_close().
769 */
5b287bba 770 spin_lock_wr(&fdp->fd_spin);
259b8ea0
MD
771 while (fd <= fdp->fd_lastfile) {
772 if (fdp->fd_files[fd].fp != NULL) {
5b287bba
MD
773 spin_unlock_wr(&fdp->fd_spin);
774 /* ok if this races another close */
259b8ea0
MD
775 if (kern_close(fd) == EINTR)
776 return (EINTR);
5b287bba 777 spin_lock_wr(&fdp->fd_spin);
259b8ea0
MD
778 }
779 ++fd;
780 }
5b287bba 781 spin_unlock_wr(&fdp->fd_spin);
4336d5df
JS
782 return (0);
783}
784
785/*
984263bc 786 * Close a file descriptor.
5b287bba
MD
787 *
788 * MPSAFE
984263bc 789 */
984263bc 790int
753fd850 791sys_close(struct close_args *uap)
984263bc 792{
12693083
MD
793 return(kern_close(uap->fd));
794}
795
5b287bba
MD
796/*
797 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
798 */
12693083
MD
799int
800kern_close(int fd)
801{
dadab5e9
MD
802 struct thread *td = curthread;
803 struct proc *p = td->td_proc;
804 struct filedesc *fdp;
41c20dac 805 struct file *fp;
984263bc
MD
806 int error;
807 int holdleaders;
808
dadab5e9
MD
809 KKASSERT(p);
810 fdp = p->p_fd;
811
5b287bba
MD
812 spin_lock_wr(&fdp->fd_spin);
813 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
814 spin_unlock_wr(&fdp->fd_spin);
984263bc 815 return (EBADF);
5b287bba 816 }
984263bc
MD
817 holdleaders = 0;
818 if (p->p_fdtol != NULL) {
819 /*
820 * Ask fdfree() to sleep to ensure that all relevant
821 * process leaders can be traversed in closef().
822 */
823 fdp->fd_holdleaderscount++;
824 holdleaders = 1;
825 }
826
827 /*
828 * we now hold the fp reference that used to be owned by the descriptor
829 * array.
830 */
5b287bba
MD
831 spin_unlock_wr(&fdp->fd_spin);
832 if (fd < fdp->fd_knlistsize) {
833 get_mplock();
7749886d
MD
834 if (fd < fdp->fd_knlistsize)
835 knote_fdclose(p, fd);
5b287bba
MD
836 rel_mplock();
837 }
b8477cda 838 error = closef(fp, p);
984263bc 839 if (holdleaders) {
5b287bba 840 spin_lock_wr(&fdp->fd_spin);
984263bc
MD
841 fdp->fd_holdleaderscount--;
842 if (fdp->fd_holdleaderscount == 0 &&
843 fdp->fd_holdleaderswakeup != 0) {
844 fdp->fd_holdleaderswakeup = 0;
5b287bba 845 spin_unlock_wr(&fdp->fd_spin);
984263bc 846 wakeup(&fdp->fd_holdleaderscount);
5b287bba
MD
847 } else {
848 spin_unlock_wr(&fdp->fd_spin);
984263bc
MD
849 }
850 }
851 return (error);
852}
853
004d2de5
MD
854/*
855 * shutdown_args(int fd, int how)
856 */
857int
858kern_shutdown(int fd, int how)
859{
860 struct thread *td = curthread;
861 struct proc *p = td->td_proc;
004d2de5
MD
862 struct file *fp;
863 int error;
864
865 KKASSERT(p);
866
5b287bba 867 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
004d2de5 868 return (EBADF);
87de5057 869 error = fo_shutdown(fp, how);
9f87144f 870 fdrop(fp);
004d2de5
MD
871
872 return (error);
873}
874
875int
753fd850 876sys_shutdown(struct shutdown_args *uap)
004d2de5
MD
877{
878 int error;
879
880 error = kern_shutdown(uap->s, uap->how);
881
882 return (error);
883}
884
984263bc 885int
8f6f8622 886kern_fstat(int fd, struct stat *ub)
984263bc 887{
dadab5e9
MD
888 struct thread *td = curthread;
889 struct proc *p = td->td_proc;
41c20dac 890 struct file *fp;
984263bc
MD
891 int error;
892
dadab5e9 893 KKASSERT(p);
8f6f8622 894
5b287bba 895 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
984263bc 896 return (EBADF);
87de5057 897 error = fo_stat(fp, ub, p->p_ucred);
9f87144f 898 fdrop(fp);
8f6f8622 899
984263bc
MD
900 return (error);
901}
984263bc
MD
902
903/*
904 * Return status information about a file descriptor.
905 */
984263bc 906int
753fd850 907sys_fstat(struct fstat_args *uap)
984263bc 908{
8f6f8622 909 struct stat st;
984263bc
MD
910 int error;
911
8f6f8622
DRJ
912 error = kern_fstat(uap->fd, &st);
913
984263bc 914 if (error == 0)
8f6f8622 915 error = copyout(&st, uap->sb, sizeof(st));
984263bc
MD
916 return (error);
917}
918
919/*
984263bc
MD
920 * Return pathconf information about a file descriptor.
921 */
984263bc
MD
922/* ARGSUSED */
923int
753fd850 924sys_fpathconf(struct fpathconf_args *uap)
984263bc 925{
dadab5e9
MD
926 struct thread *td = curthread;
927 struct proc *p = td->td_proc;
984263bc
MD
928 struct file *fp;
929 struct vnode *vp;
930 int error = 0;
931
dadab5e9 932 KKASSERT(p);
984263bc 933
5b287bba
MD
934 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
935 return (EBADF);
984263bc
MD
936
937 switch (fp->f_type) {
938 case DTYPE_PIPE:
939 case DTYPE_SOCKET:
940 if (uap->name != _PC_PIPE_BUF) {
941 error = EINVAL;
942 } else {
c7114eea 943 uap->sysmsg_result = PIPE_BUF;
984263bc
MD
944 error = 0;
945 }
946 break;
947 case DTYPE_FIFO:
948 case DTYPE_VNODE:
949 vp = (struct vnode *)fp->f_data;
c7114eea 950 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
984263bc
MD
951 break;
952 default:
953 error = EOPNOTSUPP;
954 break;
955 }
9f87144f 956 fdrop(fp);
984263bc
MD
957 return(error);
958}
959
984263bc
MD
960static int fdexpand;
961SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
962
5b287bba
MD
963/*
964 * Grow the file table so it can hold through descriptor (want).
965 *
966 * The fdp's spinlock must be held exclusively on entry and may be held
967 * exclusively on return. The spinlock may be cycled by the routine.
968 *
969 * MPSAFE
970 */
69908319 971static void
5b287bba 972fdgrow_locked(struct filedesc *fdp, int want)
69908319 973{
0679adc4
MD
974 struct fdnode *newfiles;
975 struct fdnode *oldfiles;
69908319
JH
976 int nf, extra;
977
978 nf = fdp->fd_nfiles;
979 do {
980 /* nf has to be of the form 2^n - 1 */
981 nf = 2 * nf + 1;
982 } while (nf <= want);
983
5b287bba 984 spin_unlock_wr(&fdp->fd_spin);
efda3bd0 985 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
5b287bba 986 spin_lock_wr(&fdp->fd_spin);
69908319
JH
987
988 /*
5b287bba
MD
989 * We could have raced another extend while we were not holding
990 * the spinlock.
69908319
JH
991 */
992 if (fdp->fd_nfiles >= nf) {
5b287bba 993 spin_unlock_wr(&fdp->fd_spin);
efda3bd0 994 kfree(newfiles, M_FILEDESC);
5b287bba 995 spin_lock_wr(&fdp->fd_spin);
69908319
JH
996 return;
997 }
69908319
JH
998 /*
999 * Copy the existing ofile and ofileflags arrays
1000 * and zero the new portion of each array.
1001 */
1002 extra = nf - fdp->fd_nfiles;
0679adc4
MD
1003 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1004 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1005
1006 oldfiles = fdp->fd_files;
1007 fdp->fd_files = newfiles;
69908319 1008 fdp->fd_nfiles = nf;
0679adc4 1009
5b287bba
MD
1010 if (oldfiles != fdp->fd_builtin_files) {
1011 spin_unlock_wr(&fdp->fd_spin);
efda3bd0 1012 kfree(oldfiles, M_FILEDESC);
5b287bba
MD
1013 spin_lock_wr(&fdp->fd_spin);
1014 }
69908319
JH
1015 fdexpand++;
1016}
1017
1018/*
1019 * Number of nodes in right subtree, including the root.
1020 */
1021static __inline int
1022right_subtree_size(int n)
1023{
1024 return (n ^ (n | (n + 1)));
1025}
1026
1027/*
1028 * Bigger ancestor.
1029 */
1030static __inline int
1031right_ancestor(int n)
1032{
1033 return (n | (n + 1));
1034}
1035
1036/*
1037 * Smaller ancestor.
1038 */
1039static __inline int
1040left_ancestor(int n)
1041{
1042 return ((n & (n + 1)) - 1);
1043}
1044
5b287bba
MD
1045/*
1046 * Traverse the in-place binary tree buttom-up adjusting the allocation
1047 * count so scans can determine where free descriptors are located.
1048 *
1049 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1050 */
fa541be6 1051static
69908319 1052void
5b287bba 1053fdreserve_locked(struct filedesc *fdp, int fd, int incr)
69908319
JH
1054{
1055 while (fd >= 0) {
0679adc4
MD
1056 fdp->fd_files[fd].allocated += incr;
1057 KKASSERT(fdp->fd_files[fd].allocated >= 0);
69908319
JH
1058 fd = left_ancestor(fd);
1059 }
1060}
1061
1062/*
259b8ea0
MD
1063 * Reserve a file descriptor for the process. If no error occurs, the
1064 * caller MUST at some point call fsetfd() or assign a file pointer
1065 * or dispose of the reservation.
5b287bba
MD
1066 *
1067 * MPSAFE
69908319 1068 */
984263bc 1069int
dda4b42b 1070fdalloc(struct proc *p, int want, int *result)
984263bc 1071{
41c20dac 1072 struct filedesc *fdp = p->p_fd;
69908319
JH
1073 int fd, rsize, rsum, node, lim;
1074
5b287bba 1075 spin_lock_rd(&p->p_limit->p_spin);
69908319 1076 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
5b287bba 1077 spin_unlock_rd(&p->p_limit->p_spin);
69908319
JH
1078 if (want >= lim)
1079 return (EMFILE);
5b287bba 1080 spin_lock_wr(&fdp->fd_spin);
69908319 1081 if (want >= fdp->fd_nfiles)
5b287bba 1082 fdgrow_locked(fdp, want);
984263bc
MD
1083
1084 /*
1085 * Search for a free descriptor starting at the higher
1086 * of want or fd_freefile. If that fails, consider
1087 * expanding the ofile array.
259b8ea0
MD
1088 *
1089 * NOTE! the 'allocated' field is a cumulative recursive allocation
1090 * count. If we happen to see a value of 0 then we can shortcut
1091 * our search. Otherwise we run through through the tree going
1092 * down branches we know have free descriptor(s) until we hit a
1093 * leaf node. The leaf node will be free but will not necessarily
1094 * have an allocated field of 0.
984263bc 1095 */
69908319
JH
1096retry:
1097 /* move up the tree looking for a subtree with a free node */
1098 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1099 fd = right_ancestor(fd)) {
0679adc4 1100 if (fdp->fd_files[fd].allocated == 0)
69908319 1101 goto found;
984263bc 1102
69908319 1103 rsize = right_subtree_size(fd);
0679adc4 1104 if (fdp->fd_files[fd].allocated == rsize)
69908319 1105 continue; /* right subtree full */
984263bc
MD
1106
1107 /*
69908319
JH
1108 * Free fd is in the right subtree of the tree rooted at fd.
1109 * Call that subtree R. Look for the smallest (leftmost)
1110 * subtree of R with an unallocated fd: continue moving
1111 * down the left branch until encountering a full left
1112 * subtree, then move to the right.
984263bc 1113 */
69908319
JH
1114 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1115 node = fd + rsize;
0679adc4
MD
1116 rsum += fdp->fd_files[node].allocated;
1117 if (fdp->fd_files[fd].allocated == rsum + rsize) {
69908319 1118 fd = node; /* move to the right */
0679adc4 1119 if (fdp->fd_files[node].allocated == 0)
69908319
JH
1120 goto found;
1121 rsum = 0;
1122 }
984263bc 1123 }
69908319 1124 goto found;
984263bc 1125 }
69908319
JH
1126
1127 /*
1128 * No space in current array. Expand?
1129 */
5b287bba
MD
1130 if (fdp->fd_nfiles >= lim) {
1131 spin_unlock_wr(&fdp->fd_spin);
69908319 1132 return (EMFILE);
5b287bba
MD
1133 }
1134 fdgrow_locked(fdp, want);
69908319
JH
1135 goto retry;
1136
1137found:
1138 KKASSERT(fd < fdp->fd_nfiles);
69908319
JH
1139 if (fd > fdp->fd_lastfile)
1140 fdp->fd_lastfile = fd;
1141 if (want <= fdp->fd_freefile)
1142 fdp->fd_freefile = fd;
1143 *result = fd;
0679adc4 1144 KKASSERT(fdp->fd_files[fd].fp == NULL);
259b8ea0
MD
1145 KKASSERT(fdp->fd_files[fd].reserved == 0);
1146 fdp->fd_files[fd].fileflags = 0;
1147 fdp->fd_files[fd].reserved = 1;
5b287bba
MD
1148 fdreserve_locked(fdp, fd, 1);
1149 spin_unlock_wr(&fdp->fd_spin);
984263bc
MD
1150 return (0);
1151}
1152
1153/*
1154 * Check to see whether n user file descriptors
1155 * are available to the process p.
5b287bba
MD
1156 *
1157 * MPSAFE
984263bc
MD
1158 */
1159int
7bf8660a 1160fdavail(struct proc *p, int n)
984263bc 1161{
41c20dac 1162 struct filedesc *fdp = p->p_fd;
0679adc4 1163 struct fdnode *fdnode;
41c20dac 1164 int i, lim, last;
984263bc 1165
5b287bba 1166 spin_lock_rd(&p->p_limit->p_spin);
984263bc 1167 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
5b287bba 1168 spin_unlock_rd(&p->p_limit->p_spin);
984263bc 1169
5b287bba
MD
1170 spin_lock_rd(&fdp->fd_spin);
1171 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1172 spin_unlock_rd(&fdp->fd_spin);
1173 return (1);
1174 }
984263bc 1175 last = min(fdp->fd_nfiles, lim);
0679adc4
MD
1176 fdnode = &fdp->fd_files[fdp->fd_freefile];
1177 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
5b287bba
MD
1178 if (fdnode->fp == NULL && --n <= 0) {
1179 spin_unlock_rd(&fdp->fd_spin);
984263bc 1180 return (1);
5b287bba 1181 }
984263bc 1182 }
5b287bba 1183 spin_unlock_rd(&fdp->fd_spin);
984263bc
MD
1184 return (0);
1185}
1186
1187/*
b8477cda
MD
1188 * Revoke open descriptors referencing (f_data, f_type)
1189 *
1190 * Any revoke executed within a prison is only able to
1191 * revoke descriptors for processes within that prison.
1192 *
1193 * Returns 0 on success or an error code.
1194 */
1195struct fdrevoke_info {
1196 void *data;
1197 short type;
1198 short unused;
1199 int count;
ea8f324c 1200 int intransit;
b8477cda
MD
1201 struct ucred *cred;
1202 struct file *nfp;
1203};
1204
ea8f324c
MD
1205static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1206static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
b8477cda
MD
1207
1208int
1209fdrevoke(void *f_data, short f_type, struct ucred *cred)
1210{
1211 struct fdrevoke_info info;
1212 int error;
1213
1214 bzero(&info, sizeof(info));
1215 info.data = f_data;
1216 info.type = f_type;
1217 info.cred = cred;
1218 error = falloc(NULL, &info.nfp, NULL);
1219 if (error)
1220 return (error);
ea8f324c
MD
1221
1222 /*
1223 * Scan the file pointer table once. dups do not dup file pointers,
1224 * only descriptors, so there is no leak. Set FREVOKED on the fps
1225 * being revoked.
1226 */
1227 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1228
1229 /*
1230 * If any fps were marked track down the related descriptors
1231 * and close them. Any dup()s at this point will notice
1232 * the FREVOKED already set in the fp and do the right thing.
1233 *
1234 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1235 * socket) bumped the intransit counter and will require a
1236 * scan. Races against fps leaving the socket are closed by
1237 * the socket code checking for FREVOKED.
1238 */
1239 if (info.count)
1240 allproc_scan(fdrevoke_proc_callback, &info);
1241 if (info.intransit)
1242 unp_revoke_gc(info.nfp);
b8477cda 1243 fdrop(info.nfp);
ea8f324c 1244 return(0);
b8477cda
MD
1245}
1246
ea8f324c
MD
1247/*
1248 * Locate matching file pointers directly.
1249 */
b8477cda 1250static int
ea8f324c
MD
1251fdrevoke_check_callback(struct file *fp, void *vinfo)
1252{
1253 struct fdrevoke_info *info = vinfo;
1254
1255 /*
1256 * File pointers already flagged for revokation are skipped.
1257 */
1258 if (fp->f_flag & FREVOKED)
1259 return(0);
1260
1261 /*
1262 * If revoking from a prison file pointers created outside of
1263 * that prison, or file pointers without creds, cannot be revoked.
1264 */
1265 if (info->cred->cr_prison &&
1266 (fp->f_cred == NULL ||
1267 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1268 return(0);
1269 }
1270
1271 /*
1272 * If the file pointer matches then mark it for revocation. The
1273 * flag is currently only used by unp_revoke_gc().
1274 *
1275 * info->count is a heuristic and can race in a SMP environment.
1276 */
1277 if (info->data == fp->f_data && info->type == fp->f_type) {
1278 atomic_set_int(&fp->f_flag, FREVOKED);
1279 info->count += fp->f_count;
1280 if (fp->f_msgcount)
1281 ++info->intransit;
1282 }
1283 return(0);
1284}
1285
1286/*
1287 * Locate matching file pointers via process descriptor tables.
1288 */
1289static int
1290fdrevoke_proc_callback(struct proc *p, void *vinfo)
b8477cda
MD
1291{
1292 struct fdrevoke_info *info = vinfo;
1293 struct filedesc *fdp;
1294 struct file *fp;
1295 int n;
1296
1297 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1298 return(0);
1299 if (info->cred->cr_prison &&
1300 info->cred->cr_prison != p->p_ucred->cr_prison) {
1301 return(0);
1302 }
1303
1304 /*
1305 * If the controlling terminal of the process matches the
1306 * vnode being revoked we clear the controlling terminal.
1307 *
1308 * The normal spec_close() may not catch this because it
1309 * uses curproc instead of p.
1310 */
1311 if (p->p_session && info->type == DTYPE_VNODE &&
1312 info->data == p->p_session->s_ttyvp) {
1313 p->p_session->s_ttyvp = NULL;
1314 vrele(info->data);
1315 }
1316
1317 /*
1318 * Locate and close any matching file descriptors.
1319 */
1320 if ((fdp = p->p_fd) == NULL)
1321 return(0);
1322 spin_lock_wr(&fdp->fd_spin);
1323 for (n = 0; n < fdp->fd_nfiles; ++n) {
1324 if ((fp = fdp->fd_files[n].fp) == NULL)
1325 continue;
ea8f324c 1326 if (fp->f_flag & FREVOKED) {
b8477cda
MD
1327 fhold(info->nfp);
1328 fdp->fd_files[n].fp = info->nfp;
1329 spin_unlock_wr(&fdp->fd_spin);
1330 closef(fp, p);
1331 spin_lock_wr(&fdp->fd_spin);
ea8f324c 1332 --info->count;
b8477cda
MD
1333 }
1334 }
1335 spin_unlock_wr(&fdp->fd_spin);
1336 return(0);
1337}
1338
1339/*
39f91578 1340 * falloc:
259b8ea0
MD
1341 * Create a new open file structure and reserve a file decriptor
1342 * for the process that refers to it.
1343 *
1344 * Root creds are checked using p, or assumed if p is NULL. If
1345 * resultfd is non-NULL then p must also be non-NULL. No file
1346 * descriptor is reserved if resultfd is NULL.
fad57d0e 1347 *
259b8ea0
MD
1348 * A file pointer with a refcount of 1 is returned. Note that the
1349 * file pointer is NOT associated with the descriptor. If falloc
1350 * returns success, fsetfd() MUST be called to either associate the
1351 * file pointer or clear the reservation.
5b287bba 1352 *
2dd63755 1353 * MPSAFE
984263bc
MD
1354 */
1355int
39f91578 1356falloc(struct proc *p, struct file **resultfp, int *resultfd)
984263bc 1357{
c4cb6d8b
HP
1358 static struct timeval lastfail;
1359 static int curfail;
fad57d0e
MD
1360 struct file *fp;
1361 int error;
c4cb6d8b 1362
fad57d0e
MD
1363 fp = NULL;
1364
1365 /*
1366 * Handle filetable full issues and root overfill.
1367 */
60ee93b9 1368 if (nfiles >= maxfiles - maxfilesrootres &&
9b3f1fd5 1369 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) {
c4cb6d8b 1370 if (ppsratecheck(&lastfail, &curfail, 1)) {
6ea70f76 1371 kprintf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n",
9b3f1fd5 1372 (p ? p->p_ucred->cr_ruid : -1));
c4cb6d8b 1373 }
fad57d0e
MD
1374 error = ENFILE;
1375 goto done;
984263bc 1376 }
fad57d0e 1377
984263bc
MD
1378 /*
1379 * Allocate a new file descriptor.
984263bc 1380 */
efda3bd0 1381 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
85fceac1 1382 spin_init(&fp->f_spin);
984263bc 1383 fp->f_count = 1;
984263bc
MD
1384 fp->f_ops = &badfileops;
1385 fp->f_seqcount = 1;
fad57d0e 1386 if (p)
a235f7bb 1387 fp->f_cred = crhold(p->p_ucred);
fad57d0e
MD
1388 else
1389 fp->f_cred = crhold(proc0.p_ucred);
2dd63755
MD
1390 spin_lock_wr(&filehead_spin);
1391 nfiles++;
fad57d0e 1392 LIST_INSERT_HEAD(&filehead, fp, f_list);
2dd63755 1393 spin_unlock_wr(&filehead_spin);
fad57d0e 1394 if (resultfd) {
259b8ea0 1395 if ((error = fdalloc(p, 0, resultfd)) != 0) {
9f87144f 1396 fdrop(fp);
fad57d0e 1397 fp = NULL;
39f91578 1398 }
a235f7bb 1399 } else {
fad57d0e 1400 error = 0;
984263bc 1401 }
fad57d0e
MD
1402done:
1403 *resultfp = fp;
64f33bc8 1404 return (error);
fad57d0e
MD
1405}
1406
1407/*
85fceac1
MD
1408 * MPSAFE
1409 */
1410static
1411int
1412checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1413{
1414 int error;
1415
1416 spin_lock_rd(&fdp->fd_spin);
1417 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1418 error = EBADF;
1419 else
1420 error = 0;
1421 spin_unlock_rd(&fdp->fd_spin);
1422 return (error);
1423}
1424
1425/*
259b8ea0
MD
1426 * Associate a file pointer with a previously reserved file descriptor.
1427 * This function always succeeds.
1428 *
1429 * If fp is NULL, the file descriptor is returned to the pool.
fad57d0e 1430 */
fad57d0e 1431
5b287bba
MD
1432/*
1433 * MPSAFE (exclusive spinlock must be held on call)
1434 */
1435static void
1436fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1437{
259b8ea0
MD
1438 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1439 KKASSERT(fdp->fd_files[fd].reserved != 0);
1440 if (fp) {
fad57d0e 1441 fhold(fp);
259b8ea0
MD
1442 fdp->fd_files[fd].fp = fp;
1443 fdp->fd_files[fd].reserved = 0;
d9b2033e
MD
1444 if (fp->f_type == DTYPE_KQUEUE) {
1445 if (fdp->fd_knlistsize < 0)
1446 fdp->fd_knlistsize = 0;
1447 }
259b8ea0
MD
1448 } else {
1449 fdp->fd_files[fd].reserved = 0;
5b287bba
MD
1450 fdreserve_locked(fdp, fd, -1);
1451 fdfixup_locked(fdp, fd);
fad57d0e 1452 }
984263bc
MD
1453}
1454
5b287bba
MD
1455/*
1456 * MPSAFE
1457 */
1458void
1459fsetfd(struct proc *p, struct file *fp, int fd)
1460{
1461 struct filedesc *fdp = p->p_fd;
1462
1463 spin_lock_wr(&fdp->fd_spin);
1464 fsetfd_locked(fdp, fp, fd);
1465 spin_unlock_wr(&fdp->fd_spin);
1466}
1467
1468/*
1469 * MPSAFE (exclusive spinlock must be held on call)
1470 */
fa541be6 1471static
259b8ea0 1472struct file *
5b287bba 1473funsetfd_locked(struct filedesc *fdp, int fd)
69908319 1474{
259b8ea0
MD
1475 struct file *fp;
1476
1477 if ((unsigned)fd >= fdp->fd_nfiles)
1478 return (NULL);
1479 if ((fp = fdp->fd_files[fd].fp) == NULL)
1480 return (NULL);
0679adc4
MD
1481 fdp->fd_files[fd].fp = NULL;
1482 fdp->fd_files[fd].fileflags = 0;
259b8ea0 1483
5b287bba
MD
1484 fdreserve_locked(fdp, fd, -1);
1485 fdfixup_locked(fdp, fd);
259b8ea0 1486 return(fp);
69908319
JH
1487}
1488
85fceac1
MD
1489/*
1490 * MPSAFE
1491 */
1492int
1493fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1494{
1495 int error;
1496
1497 spin_lock_rd(&fdp->fd_spin);
1498 if (((u_int)fd) >= fdp->fd_nfiles) {
1499 error = EBADF;
1500 } else if (fdp->fd_files[fd].fp == NULL) {
1501 error = EBADF;
1502 } else {
1503 *flagsp = fdp->fd_files[fd].fileflags;
1504 error = 0;
1505 }
1506 spin_unlock_rd(&fdp->fd_spin);
1507 return (error);
1508}
1509
1510/*
1511 * MPSAFE
1512 */
228b401d
MD
1513int
1514fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1515{
85fceac1
MD
1516 int error;
1517
1518 spin_lock_wr(&fdp->fd_spin);
1519 if (((u_int)fd) >= fdp->fd_nfiles) {
1520 error = EBADF;
1521 } else if (fdp->fd_files[fd].fp == NULL) {
1522 error = EBADF;
1523 } else {
1524 fdp->fd_files[fd].fileflags |= add_flags;
1525 error = 0;
1526 }
1527 spin_unlock_wr(&fdp->fd_spin);
1528 return (error);
228b401d
MD
1529}
1530
85fceac1
MD
1531/*
1532 * MPSAFE
1533 */
228b401d
MD
1534int
1535fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1536{
85fceac1
MD
1537 int error;
1538
1539 spin_lock_wr(&fdp->fd_spin);
1540 if (((u_int)fd) >= fdp->fd_nfiles) {
1541 error = EBADF;
1542 } else if (fdp->fd_files[fd].fp == NULL) {
1543 error = EBADF;
1544 } else {
1545 fdp->fd_files[fd].fileflags &= ~rem_flags;
1546 error = 0;
1547 }
1548 spin_unlock_wr(&fdp->fd_spin);
1549 return (error);
228b401d
MD
1550}
1551
69908319 1552void
a235f7bb
MD
1553fsetcred(struct file *fp, struct ucred *cr)
1554{
1555 crhold(cr);
1556 crfree(fp->f_cred);
1557 fp->f_cred = cr;
1558}
1559
984263bc
MD
1560/*
1561 * Free a file descriptor.
1562 */
5b287bba 1563static
984263bc 1564void
7bf8660a 1565ffree(struct file *fp)
984263bc
MD
1566{
1567 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
2dd63755 1568 spin_lock_wr(&filehead_spin);
984263bc 1569 LIST_REMOVE(fp, f_list);
2dd63755
MD
1570 nfiles--;
1571 spin_unlock_wr(&filehead_spin);
984263bc 1572 crfree(fp->f_cred);
28623bf9
MD
1573 if (fp->f_nchandle.ncp)
1574 cache_drop(&fp->f_nchandle);
efda3bd0 1575 kfree(fp, M_FILE);
984263bc
MD
1576}
1577
1578/*
228b401d
MD
1579 * called from init_main, initialize filedesc0 for proc0.
1580 */
1581void
1582fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1583{
1584 p0->p_fd = fdp0;
1585 p0->p_fdtol = NULL;
1586 fdp0->fd_refcnt = 1;
1587 fdp0->fd_cmask = cmask;
1588 fdp0->fd_files = fdp0->fd_builtin_files;
1589 fdp0->fd_nfiles = NDFILE;
259b8ea0 1590 fdp0->fd_lastfile = -1;
85fceac1 1591 spin_init(&fdp0->fd_spin);
228b401d
MD
1592}
1593
1594/*
984263bc 1595 * Build a new filedesc structure.
5b287bba
MD
1596 *
1597 * NOT MPSAFE (vref)
984263bc
MD
1598 */
1599struct filedesc *
7bf8660a 1600fdinit(struct proc *p)
984263bc 1601{
0679adc4 1602 struct filedesc *newfdp;
41c20dac 1603 struct filedesc *fdp = p->p_fd;
984263bc 1604
efda3bd0 1605 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
5b287bba 1606 spin_lock_rd(&fdp->fd_spin);
bccde7a3 1607 if (fdp->fd_cdir) {
0679adc4
MD
1608 newfdp->fd_cdir = fdp->fd_cdir;
1609 vref(newfdp->fd_cdir);
28623bf9 1610 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
690a3127 1611 }
bccde7a3
MD
1612
1613 /*
1614 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1615 * proc0, but should unconditionally exist in other processes.
1616 */
1617 if (fdp->fd_rdir) {
0679adc4
MD
1618 newfdp->fd_rdir = fdp->fd_rdir;
1619 vref(newfdp->fd_rdir);
28623bf9 1620 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
bccde7a3
MD
1621 }
1622 if (fdp->fd_jdir) {
0679adc4
MD
1623 newfdp->fd_jdir = fdp->fd_jdir;
1624 vref(newfdp->fd_jdir);
28623bf9 1625 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
690a3127 1626 }
5b287bba 1627 spin_unlock_rd(&fdp->fd_spin);
690a3127 1628
984263bc 1629 /* Create the file descriptor table. */
0679adc4
MD
1630 newfdp->fd_refcnt = 1;
1631 newfdp->fd_cmask = cmask;
1632 newfdp->fd_files = newfdp->fd_builtin_files;
1633 newfdp->fd_nfiles = NDFILE;
1634 newfdp->fd_knlistsize = -1;
259b8ea0 1635 newfdp->fd_lastfile = -1;
85fceac1 1636 spin_init(&newfdp->fd_spin);
0679adc4
MD
1637
1638 return (newfdp);
984263bc
MD
1639}
1640
1641/*
1642 * Share a filedesc structure.
5b287bba
MD
1643 *
1644 * MPSAFE
984263bc
MD
1645 */
1646struct filedesc *
7bf8660a 1647fdshare(struct proc *p)
984263bc 1648{
5b287bba
MD
1649 struct filedesc *fdp;
1650
1651 fdp = p->p_fd;
1652 spin_lock_wr(&fdp->fd_spin);
1653 fdp->fd_refcnt++;
1654 spin_unlock_wr(&fdp->fd_spin);
1655 return (fdp);
984263bc
MD
1656}
1657
1658/*
1659 * Copy a filedesc structure.
5b287bba
MD
1660 *
1661 * MPSAFE
984263bc
MD
1662 */
1663struct filedesc *
7bf8660a 1664fdcopy(struct proc *p)
984263bc 1665{
5b287bba
MD
1666 struct filedesc *fdp = p->p_fd;
1667 struct filedesc *newfdp;
0679adc4 1668 struct fdnode *fdnode;
41c20dac 1669 int i;
5b287bba 1670 int ni;
984263bc 1671
5b287bba
MD
1672 /*
1673 * Certain daemons might not have file descriptors.
1674 */
984263bc
MD
1675 if (fdp == NULL)
1676 return (NULL);
1677
5b287bba
MD
1678 /*
1679 * Allocate the new filedesc and fd_files[] array. This can race
1680 * with operations by other threads on the fdp so we have to be
1681 * careful.
1682 */
efda3bd0 1683 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
5b287bba
MD
1684again:
1685 spin_lock_rd(&fdp->fd_spin);
1686 if (fdp->fd_lastfile < NDFILE) {
1687 newfdp->fd_files = newfdp->fd_builtin_files;
1688 i = NDFILE;
1689 } else {
1690 /*
1691 * We have to allocate (N^2-1) entries for our in-place
1692 * binary tree. Allow the table to shrink.
1693 */
1694 i = fdp->fd_nfiles;
1695 ni = (i - 1) / 2;
1696 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1697 i = ni;
1698 ni = (i - 1) / 2;
1699 }
1700 spin_unlock_rd(&fdp->fd_spin);
77652cad 1701 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
5b287bba
MD
1702 M_FILEDESC, M_WAITOK | M_ZERO);
1703
1704 /*
1705 * Check for race, retry
1706 */
1707 spin_lock_rd(&fdp->fd_spin);
1708 if (i <= fdp->fd_lastfile) {
1709 spin_unlock_rd(&fdp->fd_spin);
efda3bd0 1710 kfree(newfdp->fd_files, M_FILEDESC);
5b287bba
MD
1711 goto again;
1712 }
1713 }
1714
1715 /*
1716 * Dup the remaining fields. vref() and cache_hold() can be
1717 * safely called while holding the read spinlock on fdp.
1718 *
1719 * The read spinlock on fdp is still being held.
1720 *
1721 * NOTE: vref and cache_hold calls for the case where the vnode
1722 * or cache entry already has at least one ref may be called
1723 * while holding spin locks.
1724 */
1725 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
597aea93 1726 vref(newfdp->fd_cdir);
28623bf9 1727 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
690a3127 1728 }
6bdbb368
DR
1729 /*
1730 * We must check for fd_rdir here, at least for now because
1731 * the init process is created before we have access to the
1732 * rootvode to take a reference to it.
1733 */
5b287bba 1734 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
597aea93 1735 vref(newfdp->fd_rdir);
28623bf9 1736 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
690a3127 1737 }
5b287bba 1738 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
597aea93 1739 vref(newfdp->fd_jdir);
28623bf9 1740 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
690a3127 1741 }
984263bc 1742 newfdp->fd_refcnt = 1;
984263bc 1743 newfdp->fd_nfiles = i;
5b287bba
MD
1744 newfdp->fd_lastfile = fdp->fd_lastfile;
1745 newfdp->fd_freefile = fdp->fd_freefile;
1746 newfdp->fd_cmask = fdp->fd_cmask;
1747 newfdp->fd_knlist = NULL;
1748 newfdp->fd_knlistsize = -1;
1749 newfdp->fd_knhash = NULL;
1750 newfdp->fd_knhashmask = 0;
1751 spin_init(&newfdp->fd_spin);
984263bc
MD
1752
1753 /*
5b287bba
MD
1754 * Copy the descriptor table through (i). This also copies the
1755 * allocation state. Then go through and ref the file pointers
1756 * and clean up any KQ descriptors.
1757 *
259b8ea0
MD
1758 * kq descriptors cannot be copied. Since we haven't ref'd the
1759 * copied files yet we can ignore the return value from funsetfd().
5b287bba
MD
1760 *
1761 * The read spinlock on fdp is still being held.
984263bc 1762 */
5b287bba
MD
1763 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1764 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
259b8ea0
MD
1765 fdnode = &newfdp->fd_files[i];
1766 if (fdnode->reserved) {
5b287bba 1767 fdreserve_locked(newfdp, i, -1);
259b8ea0 1768 fdnode->reserved = 0;
5b287bba
MD
1769 fdfixup_locked(newfdp, i);
1770 } else if (fdnode->fp) {
1771 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1772 (void)funsetfd_locked(newfdp, i);
1773 } else {
1774 fhold(fdnode->fp);
1775 }
259b8ea0 1776 }
984263bc 1777 }
5b287bba 1778 spin_unlock_rd(&fdp->fd_spin);
984263bc
MD
1779 return (newfdp);
1780}
1781
1782/*
1783 * Release a filedesc structure.
5b287bba
MD
1784 *
1785 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
984263bc
MD
1786 */
1787void
41c20dac 1788fdfree(struct proc *p)
984263bc 1789{
41c20dac 1790 struct filedesc *fdp = p->p_fd;
0679adc4 1791 struct fdnode *fdnode;
41c20dac 1792 int i;
984263bc
MD
1793 struct filedesc_to_leader *fdtol;
1794 struct file *fp;
1795 struct vnode *vp;
1796 struct flock lf;
1797
1798 /* Certain daemons might not have file descriptors. */
1799 if (fdp == NULL)
1800 return;
1801
5b287bba
MD
1802 /*
1803 * Severe messing around to follow
1804 */
1805 spin_lock_wr(&fdp->fd_spin);
1806
984263bc
MD
1807 /* Check for special need to clear POSIX style locks */
1808 fdtol = p->p_fdtol;
1809 if (fdtol != NULL) {
1810 KASSERT(fdtol->fdl_refcount > 0,
1811 ("filedesc_to_refcount botch: fdl_refcount=%d",
1812 fdtol->fdl_refcount));
1813 if (fdtol->fdl_refcount == 1 &&
1814 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
259b8ea0
MD
1815 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1816 fdnode = &fdp->fd_files[i];
0679adc4 1817 if (fdnode->fp == NULL ||
259b8ea0 1818 fdnode->fp->f_type != DTYPE_VNODE) {
984263bc 1819 continue;
259b8ea0 1820 }
0679adc4 1821 fp = fdnode->fp;
984263bc 1822 fhold(fp);
5b287bba
MD
1823 spin_unlock_wr(&fdp->fd_spin);
1824
984263bc
MD
1825 lf.l_whence = SEEK_SET;
1826 lf.l_start = 0;
1827 lf.l_len = 0;
1828 lf.l_type = F_UNLCK;
1829 vp = (struct vnode *)fp->f_data;
1830 (void) VOP_ADVLOCK(vp,
1831 (caddr_t)p->p_leader,
1832 F_UNLCK,
1833 &lf,
1834 F_POSIX);
9f87144f 1835 fdrop(fp);
5b287bba 1836 spin_lock_wr(&fdp->fd_spin);
984263bc
MD
1837 }
1838 }
1839 retry:
1840 if (fdtol->fdl_refcount == 1) {
1841 if (fdp->fd_holdleaderscount > 0 &&
1842 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1843 /*
1844 * close() or do_dup() has cleared a reference
1845 * in a shared file descriptor table.
1846 */
1847 fdp->fd_holdleaderswakeup = 1;
7749886d
MD
1848 msleep(&fdp->fd_holdleaderscount,
1849 &fdp->fd_spin, 0, "fdlhold", 0);
984263bc
MD
1850 goto retry;
1851 }
1852 if (fdtol->fdl_holdcount > 0) {
1853 /*
1854 * Ensure that fdtol->fdl_leader
1855 * remains valid in closef().
1856 */
1857 fdtol->fdl_wakeup = 1;
7749886d 1858 msleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
984263bc
MD
1859 goto retry;
1860 }
1861 }
1862 fdtol->fdl_refcount--;
1863 if (fdtol->fdl_refcount == 0 &&
1864 fdtol->fdl_holdcount == 0) {
1865 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1866 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
5b287bba 1867 } else {
984263bc 1868 fdtol = NULL;
5b287bba 1869 }
984263bc 1870 p->p_fdtol = NULL;
5b287bba
MD
1871 if (fdtol != NULL) {
1872 spin_unlock_wr(&fdp->fd_spin);
efda3bd0 1873 kfree(fdtol, M_FILEDESC_TO_LEADER);
5b287bba
MD
1874 spin_lock_wr(&fdp->fd_spin);
1875 }
984263bc 1876 }
5b287bba
MD
1877 if (--fdp->fd_refcnt > 0) {
1878 spin_unlock_wr(&fdp->fd_spin);
984263bc 1879 return;
5b287bba
MD
1880 }
1881 spin_unlock_wr(&fdp->fd_spin);
1882
984263bc
MD
1883 /*
1884 * we are the last reference to the structure, we can
1885 * safely assume it will not change out from under us.
1886 */
0679adc4
MD
1887 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1888 if (fdp->fd_files[i].fp)
b8477cda 1889 closef(fdp->fd_files[i].fp, p);
984263bc 1890 }
0679adc4 1891 if (fdp->fd_files != fdp->fd_builtin_files)
efda3bd0 1892 kfree(fdp->fd_files, M_FILEDESC);
690a3127 1893 if (fdp->fd_cdir) {
28623bf9 1894 cache_drop(&fdp->fd_ncdir);
984263bc 1895 vrele(fdp->fd_cdir);
690a3127 1896 }
bccde7a3 1897 if (fdp->fd_rdir) {
28623bf9 1898 cache_drop(&fdp->fd_nrdir);
bccde7a3
MD
1899 vrele(fdp->fd_rdir);
1900 }
690a3127 1901 if (fdp->fd_jdir) {
28623bf9 1902 cache_drop(&fdp->fd_njdir);
984263bc 1903 vrele(fdp->fd_jdir);
690a3127 1904 }
984263bc 1905 if (fdp->fd_knlist)
efda3bd0 1906 kfree(fdp->fd_knlist, M_KQUEUE);
984263bc 1907 if (fdp->fd_knhash)
efda3bd0
MD
1908 kfree(fdp->fd_knhash, M_KQUEUE);
1909 kfree(fdp, M_FILEDESC);
984263bc
MD
1910}
1911
1912/*
fa541be6 1913 * Retrieve and reference the file pointer associated with a descriptor.
85fceac1
MD
1914 *
1915 * MPSAFE
fa541be6
MD
1916 */
1917struct file *
1918holdfp(struct filedesc *fdp, int fd, int flag)
1919{
1920 struct file* fp;
1921
85fceac1
MD
1922 spin_lock_rd(&fdp->fd_spin);
1923 if (((u_int)fd) >= fdp->fd_nfiles) {
1924 fp = NULL;
1925 goto done;
1926 }
fa541be6 1927 if ((fp = fdp->fd_files[fd].fp) == NULL)
85fceac1
MD
1928 goto done;
1929 if ((fp->f_flag & flag) == 0 && flag != -1) {
1930 fp = NULL;
1931 goto done;
1932 }
fa541be6 1933 fhold(fp);
85fceac1
MD
1934done:
1935 spin_unlock_rd(&fdp->fd_spin);
fa541be6
MD
1936 return (fp);
1937}
1938
1939/*
1940 * holdsock() - load the struct file pointer associated
1941 * with a socket into *fpp. If an error occurs, non-zero
1942 * will be returned and *fpp will be set to NULL.
5b287bba
MD
1943 *
1944 * MPSAFE
fa541be6
MD
1945 */
1946int
5b287bba 1947holdsock(struct filedesc *fdp, int fd, struct file **fpp)
fa541be6
MD
1948{
1949 struct file *fp;
5b287bba 1950 int error;
fa541be6 1951
5b287bba
MD
1952 spin_lock_rd(&fdp->fd_spin);
1953 if ((unsigned)fd >= fdp->fd_nfiles) {
1954 error = EBADF;
1955 fp = NULL;
1956 goto done;
1957 }
1958 if ((fp = fdp->fd_files[fd].fp) == NULL) {
1959 error = EBADF;
1960 goto done;
1961 }
1962 if (fp->f_type != DTYPE_SOCKET) {
1963 error = ENOTSOCK;
1964 goto done;
1965 }
fa541be6 1966 fhold(fp);
5b287bba
MD
1967 error = 0;
1968done:
1969 spin_unlock_rd(&fdp->fd_spin);
fa541be6
MD
1970 *fpp = fp;
1971 return (error);
1972}
1973
1974/*
5b287bba
MD
1975 * Convert a user file descriptor to a held file pointer.
1976 *
1977 * MPSAFE
228b401d
MD
1978 */
1979int
5b287bba 1980holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
228b401d
MD
1981{
1982 struct file *fp;
5b287bba
MD
1983 int error;
1984
1985 spin_lock_rd(&fdp->fd_spin);
1986 if ((unsigned)fd >= fdp->fd_nfiles) {
1987 error = EBADF;
1988 fp = NULL;
1989 goto done;
1990 }
1991 if ((fp = fdp->fd_files[fd].fp) == NULL) {
1992 error = EBADF;
1993 goto done;
1994 }
1995 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
8f9f2535 1996 fp = NULL;
5b287bba
MD
1997 error = EINVAL;
1998 goto done;
1999 }
2000 fhold(fp);
2001 error = 0;
2002done:
2003 spin_unlock_rd(&fdp->fd_spin);
228b401d 2004 *fpp = fp;
5b287bba 2005 return (error);
228b401d
MD
2006}
2007
2008/*
984263bc
MD
2009 * For setugid programs, we don't want to people to use that setugidness
2010 * to generate error messages which write to a file which otherwise would
2011 * otherwise be off-limits to the process.
2012 *
2013 * This is a gross hack to plug the hole. A better solution would involve
2014 * a special vop or other form of generalized access control mechanism. We
2015 * go ahead and just reject all procfs file systems accesses as dangerous.
2016 *
2017 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2018 * sufficient. We also don't for check setugidness since we know we are.
2019 */
2020static int
2021is_unsafe(struct file *fp)
2022{
2023 if (fp->f_type == DTYPE_VNODE &&
2024 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2025 return (1);
2026 return (0);
2027}
2028
2029/*
2030 * Make this setguid thing safe, if at all possible.
5b287bba
MD
2031 *
2032 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
984263bc
MD
2033 */
2034void
dadab5e9 2035setugidsafety(struct proc *p)
984263bc
MD
2036{
2037 struct filedesc *fdp = p->p_fd;
41c20dac 2038 int i;
984263bc
MD
2039
2040 /* Certain daemons might not have file descriptors. */
2041 if (fdp == NULL)
2042 return;
2043
2044 /*
0679adc4 2045 * note: fdp->fd_files may be reallocated out from under us while
984263bc
MD
2046 * we are blocked in a close. Be careful!
2047 */
2048 for (i = 0; i <= fdp->fd_lastfile; i++) {
2049 if (i > 2)
2050 break;
0679adc4 2051 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
984263bc
MD
2052 struct file *fp;
2053
984263bc
MD
2054 if (i < fdp->fd_knlistsize)
2055 knote_fdclose(p, i);
2056 /*
2057 * NULL-out descriptor prior to close to avoid
2058 * a race while close blocks.
2059 */
5b287bba 2060 if ((fp = funsetfd_locked(fdp, i)) != NULL)
b8477cda 2061 closef(fp, p);
984263bc
MD
2062 }
2063 }
984263bc
MD
2064}
2065
2066/*
2067 * Close any files on exec?
5b287bba
MD
2068 *
2069 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
984263bc
MD
2070 */
2071void
dadab5e9 2072fdcloseexec(struct proc *p)
984263bc
MD
2073{
2074 struct filedesc *fdp = p->p_fd;
41c20dac 2075 int i;
984263bc
MD
2076
2077 /* Certain daemons might not have file descriptors. */
2078 if (fdp == NULL)
2079 return;
2080
2081 /*
0679adc4
MD
2082 * We cannot cache fd_files since operations may block and rip
2083 * them out from under us.
984263bc
MD
2084 */
2085 for (i = 0; i <= fdp->fd_lastfile; i++) {
0679adc4
MD
2086 if (fdp->fd_files[i].fp != NULL &&
2087 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
984263bc
MD
2088 struct file *fp;
2089
984263bc
MD
2090 if (i < fdp->fd_knlistsize)
2091 knote_fdclose(p, i);
2092 /*
2093 * NULL-out descriptor prior to close to avoid
2094 * a race while close blocks.
2095 */
5b287bba 2096 if ((fp = funsetfd_locked(fdp, i)) != NULL)
b8477cda 2097 closef(fp, p);
984263bc
MD
2098 }
2099 }
984263bc
MD
2100}
2101
2102/*
2103 * It is unsafe for set[ug]id processes to be started with file
2104 * descriptors 0..2 closed, as these descriptors are given implicit
2105 * significance in the Standard C library. fdcheckstd() will create a
2106 * descriptor referencing /dev/null for each of stdin, stdout, and
2107 * stderr that is not already open.
5b287bba
MD
2108 *
2109 * NOT MPSAFE - calls falloc, vn_open, etc
984263bc
MD
2110 */
2111int
dadab5e9 2112fdcheckstd(struct proc *p)
984263bc 2113{
fad57d0e 2114 struct nlookupdata nd;
dadab5e9
MD
2115 struct filedesc *fdp;
2116 struct file *fp;
2117 register_t retval;
259b8ea0 2118 int i, error, flags, devnull;
984263bc 2119
ea8f324c
MD
2120 fdp = p->p_fd;
2121 if (fdp == NULL)
2122 return (0);
2123 devnull = -1;
2124 error = 0;
2125 for (i = 0; i < 3; i++) {
0679adc4 2126 if (fdp->fd_files[i].fp != NULL)
fad57d0e
MD
2127 continue;
2128 if (devnull < 0) {
259b8ea0 2129 if ((error = falloc(p, &fp, &devnull)) != 0)
fad57d0e
MD
2130 break;
2131
2132 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2133 NLC_FOLLOW|NLC_LOCKVP);
2134 flags = FREAD | FWRITE;
2135 if (error == 0)
2136 error = vn_open(&nd, fp, flags, 0);
2137 if (error == 0)
259b8ea0
MD
2138 fsetfd(p, fp, devnull);
2139 else
2140 fsetfd(p, NULL, devnull);
9f87144f 2141 fdrop(fp);
fad57d0e
MD
2142 nlookup_done(&nd);
2143 if (error)
2144 break;
259b8ea0 2145 KKASSERT(i == devnull);
fad57d0e
MD
2146 } else {
2147 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2148 if (error != 0)
2149 break;
2150 }
ea8f324c
MD
2151 }
2152 return (error);
984263bc
MD
2153}
2154
2155/*
2156 * Internal form of close.
2157 * Decrement reference count on file structure.
dadab5e9 2158 * Note: td and/or p may be NULL when closing a file
984263bc 2159 * that was being passed in a message.
5b287bba
MD
2160 *
2161 * MPALMOSTSAFE - acquires mplock for VOP operations
984263bc
MD
2162 */
2163int
b8477cda 2164closef(struct file *fp, struct proc *p)
984263bc
MD
2165{
2166 struct vnode *vp;
2167 struct flock lf;
2168 struct filedesc_to_leader *fdtol;
2169
2170 if (fp == NULL)
2171 return (0);
b8477cda 2172
984263bc
MD
2173 /*
2174 * POSIX record locking dictates that any close releases ALL
2175 * locks owned by this process. This is handled by setting
2176 * a flag in the unlock to free ONLY locks obeying POSIX
2177 * semantics, and not to free BSD-style file locks.
2178 * If the descriptor was in a message, POSIX-style locks
2179 * aren't passed with the descriptor.
2180 */
7749886d
MD
2181 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2182 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2183 ) {
2184 get_mplock();
984263bc
MD
2185 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2186 lf.l_whence = SEEK_SET;
2187 lf.l_start = 0;
2188 lf.l_len = 0;
2189 lf.l_type = F_UNLCK;
2190 vp = (struct vnode *)fp->f_data;
2191 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2192 &lf, F_POSIX);
2193 }
2194 fdtol = p->p_fdtol;
2195 if (fdtol != NULL) {
2196 /*
2197 * Handle special case where file descriptor table
2198 * is shared between multiple process leaders.
2199 */
2200 for (fdtol = fdtol->fdl_next;
2201 fdtol != p->p_fdtol;
2202 fdtol = fdtol->fdl_next) {
2203 if ((fdtol->fdl_leader->p_flag &
2204 P_ADVLOCK) == 0)
2205 continue;
2206 fdtol->fdl_holdcount++;
2207 lf.l_whence = SEEK_SET;
2208 lf.l_start = 0;
2209 lf.l_len = 0;
2210 lf.l_type = F_UNLCK;
2211 vp = (struct vnode *)fp->f_data;
2212 (void) VOP_ADVLOCK(vp,
661fc1a0 2213 (caddr_t)fdtol->fdl_leader,
984263bc
MD
2214 F_UNLCK, &lf, F_POSIX);
2215 fdtol->fdl_holdcount--;
2216 if (fdtol->fdl_holdcount == 0 &&
2217 fdtol->fdl_wakeup != 0) {
2218 fdtol->fdl_wakeup = 0;
2219 wakeup(fdtol);
2220 }
2221 }
2222 }
7749886d 2223 rel_mplock();
984263bc 2224 }
9f87144f 2225 return (fdrop(fp));
984263bc
MD
2226}
2227
85fceac1
MD
2228/*
2229 * MPSAFE
2230 *
2231 * fhold() can only be called if f_count is already at least 1 (i.e. the
2232 * caller of fhold() already has a reference to the file pointer in some
5b287bba
MD
2233 * manner or other).
2234 *
bc1eb676
MD
2235 * f_count is not spin-locked. Instead, atomic ops are used for
2236 * incrementing, decrementing, and handling the 1->0 transition.
85fceac1
MD
2237 */
2238void
2239fhold(struct file *fp)
2240{
2241 atomic_add_int(&fp->f_count, 1);
2242}
2243
2244/*
bc1eb676 2245 * fdrop() - drop a reference to a descriptor
85fceac1 2246 *
5b287bba 2247 * MPALMOSTSAFE - acquires mplock for final close sequence
85fceac1 2248 */
984263bc 2249int
9f87144f 2250fdrop(struct file *fp)
984263bc
MD
2251{
2252 struct flock lf;
2253 struct vnode *vp;
2254 int error;
2255
bc1eb676
MD
2256 /*
2257 * A combined fetch and subtract is needed to properly detect
2258 * 1->0 transitions, otherwise two cpus dropping from a ref
2259 * count of 2 might both try to run the 1->0 code.
2260 */
2261 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
984263bc 2262 return (0);
85fceac1 2263
5b287bba
MD
2264 get_mplock();
2265
85fceac1
MD
2266 /*
2267 * The last reference has gone away, we own the fp structure free
2268 * and clear.
2269 */
984263bc
MD
2270 if (fp->f_count < 0)
2271 panic("fdrop: count < 0");
7749886d
MD
2272 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2273 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2274 ) {
984263bc
MD
2275 lf.l_whence = SEEK_SET;
2276 lf.l_start = 0;
2277 lf.l_len = 0;
2278 lf.l_type = F_UNLCK;
2279 vp = (struct vnode *)fp->f_data;
71c18fe3 2280 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
984263bc
MD
2281 }
2282 if (fp->f_ops != &badfileops)
87de5057 2283 error = fo_close(fp);
984263bc
MD
2284 else
2285 error = 0;
2286 ffree(fp);
5b287bba 2287 rel_mplock();
984263bc
MD
2288 return (error);
2289}
2290
2291/*
2292 * Apply an advisory lock on a file descriptor.
2293 *
2294 * Just attempt to get a record lock of the requested type on
2295 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2296 */
984263bc 2297int
753fd850 2298sys_flock(struct flock_args *uap)
984263bc 2299{
41c20dac 2300 struct proc *p = curproc;
41c20dac 2301 struct file *fp;
984263bc
MD
2302 struct vnode *vp;
2303 struct flock lf;
5b287bba 2304 int error;
984263bc 2305
5b287bba 2306 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
984263bc 2307 return (EBADF);
5b287bba
MD
2308 if (fp->f_type != DTYPE_VNODE) {
2309 error = EOPNOTSUPP;
2310 goto done;
2311 }
984263bc
MD
2312 vp = (struct vnode *)fp->f_data;
2313 lf.l_whence = SEEK_SET;
2314 lf.l_start = 0;
2315 lf.l_len = 0;
2316 if (uap->how & LOCK_UN) {
2317 lf.l_type = F_UNLCK;
2318 fp->f_flag &= ~FHASLOCK;
5b287bba
MD
2319 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2320 goto done;
984263bc
MD
2321 }
2322 if (uap->how & LOCK_EX)
2323 lf.l_type = F_WRLCK;
2324 else if (uap->how & LOCK_SH)
2325 lf.l_type = F_RDLCK;
5b287bba
MD
2326 else {
2327 error = EBADF;
2328 goto done;
2329 }
984263bc
MD
2330 fp->f_flag |= FHASLOCK;
2331 if (uap->how & LOCK_NB)
5b287bba
MD
2332 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2333 else
2334 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2335done:
2336 fdrop(fp);
2337 return (error);
984263bc
MD
2338}
2339
2340/*
2341 * File Descriptor pseudo-device driver (/dev/fd/).
2342 *
2343 * Opening minor device N dup()s the file (if any) connected to file
2344 * descriptor N belonging to the calling process. Note that this driver
2345 * consists of only the ``open()'' routine, because all subsequent
2346 * references to this file will be direct to the other driver.
2347 */
2348/* ARGSUSED */
2349static int
fef8985e 2350fdopen(struct dev_open_args *ap)
984263bc 2351{
fef8985e
MD
2352 thread_t td = curthread;
2353
fbfe4e7d 2354 KKASSERT(td->td_lwp != NULL);
984263bc
MD
2355
2356 /*
fbfe4e7d 2357 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
984263bc
MD
2358 * the file descriptor being sought for duplication. The error
2359 * return ensures that the vnode for this device will be released
2360 * by vn_open. Open will detect this special error and take the
2361 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2362 * will simply report the error.
2363 */
fef8985e 2364 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
984263bc
MD
2365 return (ENODEV);
2366}
2367
2368/*
259b8ea0
MD
2369 * The caller has reserved the file descriptor dfd for us. On success we
2370 * must fsetfd() it. On failure the caller will clean it up.
5b287bba
MD
2371 *
2372 * NOT MPSAFE - isn't getting spinlocks, possibly other things
984263bc
MD
2373 */
2374int
259b8ea0 2375dupfdopen(struct proc *p, int dfd, int sfd, int mode, int error)
984263bc 2376{
259b8ea0 2377 struct filedesc *fdp = p->p_fd;
41c20dac 2378 struct file *wfp;
259b8ea0 2379 struct file *xfp;
ea8f324c 2380 int werror;
984263bc 2381
259b8ea0 2382 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
984263bc 2383 return (EBADF);
984263bc
MD
2384
2385 /*
ea8f324c
MD
2386 * Close a revoke/dup race. Duping a descriptor marked as revoked
2387 * will dup a dummy descriptor instead of the real one.
2388 */
2389 if (wfp->f_flag & FREVOKED) {
2390 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2391 fdrop(wfp);
2392 wfp = NULL;
2393 werror = falloc(NULL, &wfp, NULL);
2394 if (werror)
2395 return (werror);
2396 }
2397
2398 /*
984263bc
MD
2399 * There are two cases of interest here.
2400 *
259b8ea0 2401 * For ENODEV simply dup sfd to file descriptor dfd and return.
984263bc 2402 *
259b8ea0
MD
2403 * For ENXIO steal away the file structure from sfd and store it
2404 * dfd. sfd is effectively closed by this operation.
984263bc
MD
2405 *
2406 * Any other error code is just returned.
2407 */
2408 switch (error) {
2409 case ENODEV:
2410 /*
2411 * Check that the mode the file is being opened for is a
2412 * subset of the mode of the existing descriptor.
2413 */
ea8f324c
MD
2414 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2415 error = EACCES;
2416 break;
2417 }
259b8ea0
MD
2418 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2419 fsetfd(p, wfp, dfd);
2420 error = 0;
2421 break;
984263bc
MD
2422 case ENXIO:
2423 /*
2424 * Steal away the file pointer from dfd, and stuff it into indx.
2425 */
259b8ea0
MD
2426 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2427 fsetfd(p, wfp, dfd);
5b287bba 2428 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL)
259b8ea0 2429 fdrop(xfp);
259b8ea0
MD
2430 error = 0;
2431 break;
984263bc 2432 default:
259b8ea0 2433 break;
984263bc 2434 }
259b8ea0
MD
2435 fdrop(wfp);
2436 return (error);
984263bc
MD
2437}
2438
5b287bba
MD
2439/*
2440 * NOT MPSAFE - I think these refer to a common file descriptor table
2441 * and we need to spinlock that to link fdtol in.
2442 */
984263bc
MD
2443struct filedesc_to_leader *
2444filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2445 struct proc *leader)
2446{
2447 struct filedesc_to_leader *fdtol;
2448
77652cad 2449 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
7bf8660a 2450 M_FILEDESC_TO_LEADER, M_WAITOK);
984263bc
MD
2451 fdtol->fdl_refcount = 1;
2452 fdtol->fdl_holdcount = 0;
2453 fdtol->fdl_wakeup = 0;
2454 fdtol->fdl_leader = leader;
2455 if (old != NULL) {
2456 fdtol->fdl_next = old->fdl_next;
2457 fdtol->fdl_prev = old;
2458 old->fdl_next = fdtol;
2459 fdtol->fdl_next->fdl_prev = fdtol;
2460 } else {
2461 fdtol->fdl_next = fdtol;
2462 fdtol->fdl_prev = fdtol;
2463 }
2464 return fdtol;
2465}
2466
2467/*
2dd63755 2468 * Scan all file pointers in the system. The callback is made with
ea8f324c 2469 * the master list spinlock held exclusively.
2dd63755
MD
2470 *
2471 * MPSAFE
2dd63755
MD
2472 */
2473void
2474allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2475{
2476 struct file *fp;
2477 int res;
2478
2479 spin_lock_wr(&filehead_spin);
2480 LIST_FOREACH(fp, &filehead, f_list) {
2dd63755 2481 res = callback(fp, data);
2dd63755
MD
2482 if (res < 0)
2483 break;
2484 }
2485 spin_unlock_wr(&filehead_spin);
2486}
2487
2488/*
984263bc 2489 * Get file structures.
5b287bba
MD
2490 *
2491 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
984263bc 2492 */
8fa76237
MD
2493
2494struct sysctl_kern_file_info {
2495 int count;
2496 int error;
2497 struct sysctl_req *req;
2498};
2499
2500static int sysctl_kern_file_callback(struct proc *p, void *data);
2501
984263bc
MD
2502static int
2503sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2504{
8fa76237 2505 struct sysctl_kern_file_info info;
984263bc
MD
2506
2507 /*
7b124c9f
JS
2508 * Note: because the number of file descriptors is calculated
2509 * in different ways for sizing vs returning the data,
2510 * there is information leakage from the first loop. However,
2511 * it is of a similar order of magnitude to the leakage from
2512 * global system statistics such as kern.openfiles.
6d132b4d
MD
2513 *
2514 * When just doing a count, note that we cannot just count
2515 * the elements and add f_count via the filehead list because
2516 * threaded processes share their descriptor table and f_count might
2517 * still be '1' in that case.
e38910da
MD
2518 *
2519 * Since the SYSCTL op can block, we must hold the process to
2520 * prevent it being ripped out from under us either in the
2521 * file descriptor loop or in the greater LIST_FOREACH. The
2522 * process may be in varying states of disrepair. If the process
2523 * is in SZOMB we may have caught it just as it is being removed
2524 * from the allproc list, we must skip it in that case to maintain
2525 * an unbroken chain through the allproc list.
984263bc 2526 */
8fa76237
MD
2527 info.count = 0;
2528 info.error = 0;
2529 info.req = req;
2530 allproc_scan(sysctl_kern_file_callback, &info);
6d132b4d
MD
2531
2532 /*
2533 * When just calculating the size, overestimate a bit to try to
2534 * prevent system activity from causing the buffer-fill call
2535 * to fail later on.
2536 */
2537 if (req->oldptr == NULL) {
8fa76237
MD
2538 info.count = (info.count + 16) + (info.count / 10);
2539 info.error = SYSCTL_OUT(req, NULL,
2540 info.count * sizeof(struct kinfo_file));
6d132b4d 2541 }
8fa76237
MD
2542 return (info.error);
2543}
2544
2545static int
2546sysctl_kern_file_callback(struct proc *p, void *data)
2547{
2548 struct sysctl_kern_file_info *info = data;
2549 struct kinfo_file kf;
2550 struct filedesc *fdp;
2551 struct file *fp;
2552 uid_t uid;
2553 int n;
2554
416d05d7 2555 if (p->p_stat == SIDL || p->p_stat == SZOMB)
8fa76237
MD
2556 return(0);
2557 if (!PRISON_CHECK(info->req->td->td_proc->p_ucred, p->p_ucred) != 0)
2558 return(0);
2559 if ((fdp = p->p_fd) == NULL)
2560 return(0);
2561 spin_lock_rd(&fdp->fd_spin);
2562 for (n = 0; n < fdp->fd_nfiles; ++n) {
2563 if ((fp = fdp->fd_files[n].fp) == NULL)
2564 continue;
2565 if (info->req->oldptr == NULL) {
2566 ++info->count;
2567 } else {
2568 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2569 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2570 spin_unlock_rd(&fdp->fd_spin);
2571 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2572 spin_lock_rd(&fdp->fd_spin);
2573 if (info->error)
2574 break;
2575 }
2576 }
2577 spin_unlock_rd(&fdp->fd_spin);
2578 if (info->error)
2579 return(-1);
2580 return(0);
984263bc
MD
2581}
2582
2583SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2584 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2585
2586SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2587 &maxfilesperproc, 0, "Maximum files allowed open per process");
2588
2589SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2590 &maxfiles, 0, "Maximum number of files");
2591
60ee93b9
MD
2592SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2593 &maxfilesrootres, 0, "Descriptors reserved for root use");
2594
984263bc
MD
2595SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2596 &nfiles, 0, "System-wide number of open files");
2597
2598static void
2599fildesc_drvinit(void *unused)
2600{
2601 int fd;
2602
fef8985e 2603 dev_ops_add(&fildesc_ops, 0, 0);
e4c9c0c8 2604 for (fd = 0; fd < NUMFDESC; fd++) {
fef8985e 2605 make_dev(&fildesc_ops, fd,
984263bc 2606 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
e4c9c0c8 2607 }
fef8985e
MD
2608 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2609 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2610 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
984263bc
MD
2611}
2612
d9b2033e
MD
2613/*
2614 * MPSAFE
2615 */
984263bc 2616struct fileops badfileops = {
b2d248cb
MD
2617 .fo_read = badfo_readwrite,
2618 .fo_write = badfo_readwrite,
2619 .fo_ioctl = badfo_ioctl,
2620 .fo_poll = badfo_poll,
2621 .fo_kqfilter = badfo_kqfilter,
2622 .fo_stat = badfo_stat,
2623 .fo_close = badfo_close,
2624 .fo_shutdown = badfo_shutdown
984263bc
MD
2625};
2626
d9b2033e
MD
2627/*
2628 * MPSAFE
2629 */
984263bc 2630static int
dadab5e9
MD
2631badfo_readwrite(
2632 struct file *fp,
2633 struct uio *uio,
2634 struct ucred *cred,
87de5057 2635 int flags
dadab5e9 2636) {
984263bc
MD
2637 return (EBADF);
2638}
2639
d9b2033e
MD
2640/*
2641 * MPSAFE
2642 */
984263bc 2643static int
87de5057 2644badfo_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *cred)
984263bc 2645{
984263bc
MD
2646 return (EBADF);
2647}
2648
d9b2033e
MD
2649/*
2650 * MPSAFE
2651 */
984263bc 2652static int
87de5057 2653badfo_poll(struct file *fp, int events, struct ucred *cred)
984263bc 2654{
984263bc
MD
2655 return (0);
2656}
2657
d9b2033e
MD
2658/*
2659 * MPSAFE
2660 */
984263bc 2661static int
dadab5e9 2662badfo_kqfilter(struct file *fp, struct knote *kn)
984263bc 2663{
984263bc
MD
2664 return (0);
2665}
2666
2667static int
87de5057 2668badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
984263bc 2669{
984263bc
MD
2670 return (EBADF);
2671}
2672
d9b2033e
MD
2673/*
2674 * MPSAFE
2675 */
984263bc 2676static int
87de5057 2677badfo_close(struct file *fp)
984263bc 2678{
984263bc
MD
2679 return (EBADF);
2680}
2681
d9b2033e
MD
2682/*
2683 * MPSAFE
2684 */
004d2de5 2685static int
87de5057 2686badfo_shutdown(struct file *fp, int how)
004d2de5
MD
2687{
2688 return (EBADF);
2689}
2690
d9b2033e
MD
2691/*
2692 * MPSAFE
2693 */
004d2de5 2694int
87de5057 2695nofo_shutdown(struct file *fp, int how)
004d2de5
MD
2696{
2697 return (EOPNOTSUPP);
2698}
2699
984263bc
MD
2700SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2701 fildesc_drvinit,NULL)