kernel - use new td_ucred in numerous places
[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>
a724d72b 83#include <sys/file.h>
984263bc
MD
84#include <sys/filedesc.h>
85#include <sys/kernel.h>
86#include <sys/sysctl.h>
87#include <sys/vnode.h>
88#include <sys/proc.h>
fad57d0e 89#include <sys/nlookup.h>
984263bc
MD
90#include <sys/file.h>
91#include <sys/stat.h>
92#include <sys/filio.h>
93#include <sys/fcntl.h>
94#include <sys/unistd.h>
95#include <sys/resourcevar.h>
96#include <sys/event.h>
dda4b42b 97#include <sys/kern_syscall.h>
1c55bd1c 98#include <sys/kcore.h>
7b124c9f 99#include <sys/kinfo.h>
ea8f324c 100#include <sys/un.h>
984263bc
MD
101
102#include <vm/vm.h>
103#include <vm/vm_extern.h>
104
e43a034f 105#include <sys/thread2.h>
dadab5e9 106#include <sys/file2.h>
85fceac1 107#include <sys/spinlock2.h>
dadab5e9 108
5b287bba
MD
109static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
110static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
111static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
85fceac1 112static int checkfpclosed(struct filedesc *fdp, int fd, struct file *fp);
5b287bba 113static void ffree(struct file *fp);
fa541be6 114
984263bc
MD
115static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
116static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
117 "file desc to leader structures");
118MALLOC_DEFINE(M_FILE, "file", "Open file structure");
119static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
120
121static d_open_t fdopen;
122#define NUMFDESC 64
123
124#define CDEV_MAJOR 22
fef8985e
MD
125static struct dev_ops fildesc_ops = {
126 { "FD", CDEV_MAJOR, 0 },
127 .d_open = fdopen,
984263bc
MD
128};
129
984263bc
MD
130/*
131 * Descriptor management.
132 */
2dd63755
MD
133static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
134static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
135static int nfiles; /* actual number of open files */
984263bc
MD
136extern int cmask;
137
138/*
8247b2f9 139 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
5b287bba
MD
140 *
141 * MPSAFE - must be called with fdp->fd_spin exclusively held
8247b2f9
MD
142 */
143static __inline
144void
5b287bba 145fdfixup_locked(struct filedesc *fdp, int fd)
8247b2f9
MD
146{
147 if (fd < fdp->fd_freefile) {
148 fdp->fd_freefile = fd;
149 }
150 while (fdp->fd_lastfile >= 0 &&
151 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
152 fdp->fd_files[fdp->fd_lastfile].reserved == 0
153 ) {
154 --fdp->fd_lastfile;
155 }
156}
157
158/*
984263bc 159 * System calls on descriptors.
5b287bba
MD
160 *
161 * MPSAFE
984263bc 162 */
984263bc 163int
753fd850 164sys_getdtablesize(struct getdtablesize_args *uap)
984263bc 165{
41c20dac 166 struct proc *p = curproc;
5b287bba 167 struct plimit *limit = p->p_limit;
984263bc 168
5b287bba 169 spin_lock_rd(&limit->p_spin);
c7114eea 170 uap->sysmsg_result =
5b287bba
MD
171 min((int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
172 spin_unlock_rd(&limit->p_spin);
984263bc
MD
173 return (0);
174}
175
176/*
177 * Duplicate a file descriptor to a particular value.
178 *
179 * note: keep in mind that a potential race condition exists when closing
180 * descriptors from a shared descriptor table (via rfork).
5b287bba
MD
181 *
182 * MPSAFE
984263bc 183 */
984263bc 184int
753fd850 185sys_dup2(struct dup2_args *uap)
984263bc 186{
dda4b42b 187 int error;
973c11b9 188 int fd = 0;
dda4b42b 189
973c11b9
MD
190 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
191 uap->sysmsg_fds[0] = fd;
dda4b42b
DRJ
192
193 return (error);
984263bc
MD
194}
195
196/*
197 * Duplicate a file descriptor.
5b287bba
MD
198 *
199 * MPSAFE
984263bc 200 */
984263bc 201int
753fd850 202sys_dup(struct dup_args *uap)
984263bc 203{
dda4b42b 204 int error;
973c11b9 205 int fd = 0;
984263bc 206
973c11b9
MD
207 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
208 uap->sysmsg_fds[0] = fd;
dda4b42b
DRJ
209
210 return (error);
984263bc
MD
211}
212
5b287bba
MD
213/*
214 * MPALMOSTSAFE - acquires mplock for fp operations
215 */
984263bc 216int
87de5057 217kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
984263bc 218{
dadab5e9
MD
219 struct thread *td = curthread;
220 struct proc *p = td->td_proc;
41c20dac 221 struct file *fp;
984263bc 222 struct vnode *vp;
984263bc 223 u_int newmin;
9ba76b73 224 u_int oflags;
3a907475 225 u_int nflags;
dda4b42b 226 int tmp, error, flg = F_POSIX;
984263bc 227
dadab5e9
MD
228 KKASSERT(p);
229
5b287bba
MD
230 /*
231 * Operations on file descriptors that do not require a file pointer.
232 */
dda4b42b 233 switch (cmd) {
984263bc 234 case F_GETFD:
85fceac1
MD
235 error = fgetfdflags(p->p_fd, fd, &tmp);
236 if (error == 0)
237 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
5b287bba 238 return (error);
984263bc
MD
239
240 case F_SETFD:
85fceac1
MD
241 if (dat->fc_cloexec & FD_CLOEXEC)
242 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
243 else
244 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
5b287bba
MD
245 return (error);
246 case F_DUPFD:
247 newmin = dat->fc_fd;
248 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
249 return (error);
250 default:
85fceac1 251 break;
5b287bba 252 }
984263bc 253
5b287bba
MD
254 /*
255 * Operations on file pointers
256 */
257 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
258 return (EBADF);
259
260 get_mplock();
261 switch (cmd) {
984263bc 262 case F_GETFL:
dda4b42b 263 dat->fc_flags = OFLAGS(fp->f_flag);
85fceac1
MD
264 error = 0;
265 break;
984263bc
MD
266
267 case F_SETFL:
3a907475
MD
268 oflags = fp->f_flag;
269 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
270 nflags |= oflags & ~FCNTLFLAGS;
271
f58b505b 272 error = 0;
3a907475
MD
273 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
274 error = EINVAL;
275 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
e8cdd284 276 tmp = nflags & FASYNC;
87baaf0c
MD
277 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
278 cred, NULL);
f58b505b 279 }
3a907475
MD
280 if (error == 0)
281 fp->f_flag = nflags;
85fceac1 282 break;
984263bc
MD
283
284 case F_GETOWN:
87baaf0c
MD
285 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
286 cred, NULL);
85fceac1 287 break;
984263bc
MD
288
289 case F_SETOWN:
87baaf0c
MD
290 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
291 cred, NULL);
85fceac1 292 break;
984263bc
MD
293
294 case F_SETLKW:
295 flg |= F_WAIT;
296 /* Fall into F_SETLK */
297
298 case F_SETLK:
85fceac1
MD
299 if (fp->f_type != DTYPE_VNODE) {
300 error = EBADF;
301 break;
302 }
984263bc
MD
303 vp = (struct vnode *)fp->f_data;
304
305 /*
306 * copyin/lockop may block
307 */
dda4b42b
DRJ
308 if (dat->fc_flock.l_whence == SEEK_CUR)
309 dat->fc_flock.l_start += fp->f_offset;
984263bc 310
dda4b42b 311 switch (dat->fc_flock.l_type) {
984263bc
MD
312 case F_RDLCK:
313 if ((fp->f_flag & FREAD) == 0) {
314 error = EBADF;
315 break;
316 }
317 p->p_leader->p_flag |= P_ADVLOCK;
318 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
dda4b42b 319 &dat->fc_flock, flg);
984263bc
MD
320 break;
321 case F_WRLCK:
322 if ((fp->f_flag & FWRITE) == 0) {
323 error = EBADF;
324 break;
325 }
326 p->p_leader->p_flag |= P_ADVLOCK;
327 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
dda4b42b 328 &dat->fc_flock, flg);
984263bc
MD
329 break;
330 case F_UNLCK:
331 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
dda4b42b 332 &dat->fc_flock, F_POSIX);
984263bc
MD
333 break;
334 default:
335 error = EINVAL;
336 break;
337 }
85fceac1
MD
338
339 /*
340 * It is possible to race a close() on the descriptor while
341 * we were blocked getting the lock. If this occurs the
342 * close might not have caught the lock.
343 */
344 if (checkfpclosed(p->p_fd, fd, fp)) {
dda4b42b
DRJ
345 dat->fc_flock.l_whence = SEEK_SET;
346 dat->fc_flock.l_start = 0;
347 dat->fc_flock.l_len = 0;
348 dat->fc_flock.l_type = F_UNLCK;
984263bc 349 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
dda4b42b 350 F_UNLCK, &dat->fc_flock, F_POSIX);
984263bc 351 }
85fceac1 352 break;
984263bc
MD
353
354 case F_GETLK:
85fceac1
MD
355 if (fp->f_type != DTYPE_VNODE) {
356 error = EBADF;
357 break;
358 }
984263bc
MD
359 vp = (struct vnode *)fp->f_data;
360 /*
361 * copyin/lockop may block
362 */
dda4b42b
DRJ
363 if (dat->fc_flock.l_type != F_RDLCK &&
364 dat->fc_flock.l_type != F_WRLCK &&
365 dat->fc_flock.l_type != F_UNLCK) {
85fceac1
MD
366 error = EINVAL;
367 break;
984263bc 368 }
dda4b42b
DRJ
369 if (dat->fc_flock.l_whence == SEEK_CUR)
370 dat->fc_flock.l_start += fp->f_offset;
984263bc 371 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
dda4b42b 372 &dat->fc_flock, F_POSIX);
85fceac1 373 break;
984263bc 374 default:
85fceac1
MD
375 error = EINVAL;
376 break;
984263bc 377 }
5b287bba
MD
378 rel_mplock();
379
85fceac1
MD
380 fdrop(fp);
381 return (error);
984263bc
MD
382}
383
384/*
dda4b42b 385 * The file control system call.
5b287bba
MD
386 *
387 * MPSAFE
dda4b42b
DRJ
388 */
389int
753fd850 390sys_fcntl(struct fcntl_args *uap)
dda4b42b
DRJ
391{
392 union fcntl_dat dat;
393 int error;
394
395 switch (uap->cmd) {
396 case F_DUPFD:
397 dat.fc_fd = uap->arg;
398 break;
399 case F_SETFD:
400 dat.fc_cloexec = uap->arg;
401 break;
402 case F_SETFL:
403 dat.fc_flags = uap->arg;
404 break;
405 case F_SETOWN:
406 dat.fc_owner = uap->arg;
407 break;
408 case F_SETLKW:
409 case F_SETLK:
410 case F_GETLK:
411 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
5b287bba 412 sizeof(struct flock));
dda4b42b
DRJ
413 if (error)
414 return (error);
415 break;
416 }
417
9910d07b 418 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
dda4b42b
DRJ
419
420 if (error == 0) {
421 switch (uap->cmd) {
422 case F_DUPFD:
423 uap->sysmsg_result = dat.fc_fd;
424 break;
425 case F_GETFD:
426 uap->sysmsg_result = dat.fc_cloexec;
427 break;
428 case F_GETFL:
429 uap->sysmsg_result = dat.fc_flags;
430 break;
431 case F_GETOWN:
432 uap->sysmsg_result = dat.fc_owner;
433 case F_GETLK:
434 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
435 sizeof(struct flock));
436 break;
437 }
438 }
439
440 return (error);
441}
442
443/*
984263bc 444 * Common code for dup, dup2, and fcntl(F_DUPFD).
dda4b42b
DRJ
445 *
446 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
447 * kern_dup() to destructively dup over an existing file descriptor if new
448 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
449 * unused file descriptor that is greater than or equal to new.
5b287bba
MD
450 *
451 * MPSAFE
984263bc 452 */
dda4b42b
DRJ
453int
454kern_dup(enum dup_type type, int old, int new, int *res)
984263bc 455{
dda4b42b
DRJ
456 struct thread *td = curthread;
457 struct proc *p = td->td_proc;
458 struct filedesc *fdp = p->p_fd;
984263bc
MD
459 struct file *fp;
460 struct file *delfp;
259b8ea0 461 int oldflags;
984263bc 462 int holdleaders;
dda4b42b
DRJ
463 int error, newfd;
464
465 /*
466 * Verify that we have a valid descriptor to dup from and
467 * possibly to dup to.
468 */
259b8ea0 469retry:
5b287bba
MD
470 spin_lock_wr(&fdp->fd_spin);
471 if (new < 0 || new > p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
472 new >= maxfilesperproc) {
473 spin_unlock_wr(&fdp->fd_spin);
474 return (EINVAL);
475 }
476 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
477 spin_unlock_wr(&fdp->fd_spin);
dda4b42b 478 return (EBADF);
5b287bba 479 }
dda4b42b
DRJ
480 if (type == DUP_FIXED && old == new) {
481 *res = new;
5b287bba 482 spin_unlock_wr(&fdp->fd_spin);
dda4b42b
DRJ
483 return (0);
484 }
0679adc4 485 fp = fdp->fd_files[old].fp;
259b8ea0 486 oldflags = fdp->fd_files[old].fileflags;
5b287bba 487 fhold(fp); /* MPSAFE - can be called with a spinlock held */
dda4b42b
DRJ
488
489 /*
259b8ea0
MD
490 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
491 * if the requested descriptor is beyond the current table size.
492 *
493 * This can block. Retry if the source descriptor no longer matches
494 * or if our expectation in the expansion case races.
495 *
496 * If we are not expanding or allocating a new decriptor, then reset
497 * the target descriptor to a reserved state so we have a uniform
498 * setup for the next code block.
dda4b42b
DRJ
499 */
500 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
5b287bba 501 spin_unlock_wr(&fdp->fd_spin);
dda4b42b 502 error = fdalloc(p, new, &newfd);
5b287bba 503 spin_lock_wr(&fdp->fd_spin);
dda4b42b 504 if (error) {
5b287bba 505 spin_unlock_wr(&fdp->fd_spin);
9f87144f 506 fdrop(fp);
dda4b42b
DRJ
507 return (error);
508 }
5b287bba
MD
509 /*
510 * Check for ripout
511 */
259b8ea0 512 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
5b287bba
MD
513 fsetfd_locked(fdp, NULL, newfd);
514 spin_unlock_wr(&fdp->fd_spin);
259b8ea0
MD
515 fdrop(fp);
516 goto retry;
517 }
5b287bba
MD
518 /*
519 * Check for expansion race
520 */
259b8ea0 521 if (type != DUP_VARIABLE && new != newfd) {
5b287bba
MD
522 fsetfd_locked(fdp, NULL, newfd);
523 spin_unlock_wr(&fdp->fd_spin);
259b8ea0
MD
524 fdrop(fp);
525 goto retry;
526 }
5b287bba
MD
527 /*
528 * Check for ripout, newfd reused old (this case probably
529 * can't occur).
530 */
259b8ea0 531 if (old == newfd) {
5b287bba
MD
532 fsetfd_locked(fdp, NULL, newfd);
533 spin_unlock_wr(&fdp->fd_spin);
259b8ea0
MD
534 fdrop(fp);
535 goto retry;
536 }
dda4b42b 537 new = newfd;
259b8ea0
MD
538 delfp = NULL;
539 } else {
540 if (fdp->fd_files[new].reserved) {
5b287bba 541 spin_unlock_wr(&fdp->fd_spin);
259b8ea0 542 fdrop(fp);
6ea70f76 543 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
259b8ea0
MD
544 tsleep(fdp, 0, "fdres", hz);
545 goto retry;
546 }
dda4b42b 547
259b8ea0
MD
548 /*
549 * If the target descriptor was never allocated we have
550 * to allocate it. If it was we have to clean out the
5b287bba
MD
551 * old descriptor. delfp inherits the ref from the
552 * descriptor table.
259b8ea0
MD
553 */
554 delfp = fdp->fd_files[new].fp;
555 fdp->fd_files[new].fp = NULL;
556 fdp->fd_files[new].reserved = 1;
557 if (delfp == NULL) {
5b287bba 558 fdreserve_locked(fdp, new, 1);
259b8ea0
MD
559 if (new > fdp->fd_lastfile)
560 fdp->fd_lastfile = new;
dda4b42b 561 }
259b8ea0 562
dda4b42b 563 }
984263bc
MD
564
565 /*
5b287bba
MD
566 * NOTE: still holding an exclusive spinlock
567 */
568
569 /*
259b8ea0
MD
570 * If a descriptor is being overwritten we may hve to tell
571 * fdfree() to sleep to ensure that all relevant process
572 * leaders can be traversed in closef().
984263bc 573 */
984263bc 574 if (delfp != NULL && p->p_fdtol != NULL) {
984263bc
MD
575 fdp->fd_holdleaderscount++;
576 holdleaders = 1;
259b8ea0 577 } else {
984263bc 578 holdleaders = 0;
259b8ea0 579 }
dda4b42b 580 KASSERT(delfp == NULL || type == DUP_FIXED,
259b8ea0 581 ("dup() picked an open file"));
984263bc
MD
582
583 /*
259b8ea0
MD
584 * Duplicate the source descriptor, update lastfile. If the new
585 * descriptor was not allocated and we aren't replacing an existing
586 * descriptor we have to mark the descriptor as being in use.
587 *
588 * The fd_files[] array inherits fp's hold reference.
984263bc 589 */
5b287bba 590 fsetfd_locked(fdp, fp, new);
259b8ea0 591 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
5b287bba
MD
592 spin_unlock_wr(&fdp->fd_spin);
593 fdrop(fp);
dda4b42b 594 *res = new;
984263bc
MD
595
596 /*
597 * If we dup'd over a valid file, we now own the reference to it
598 * and must dispose of it using closef() semantics (as if a
599 * close() were performed on it).
600 */
601 if (delfp) {
b8477cda 602 closef(delfp, p);
984263bc 603 if (holdleaders) {
5b287bba 604 spin_lock_wr(&fdp->fd_spin);
984263bc
MD
605 fdp->fd_holdleaderscount--;
606 if (fdp->fd_holdleaderscount == 0 &&
607 fdp->fd_holdleaderswakeup != 0) {
608 fdp->fd_holdleaderswakeup = 0;
5b287bba 609 spin_unlock_wr(&fdp->fd_spin);
984263bc 610 wakeup(&fdp->fd_holdleaderscount);
5b287bba
MD
611 } else {
612 spin_unlock_wr(&fdp->fd_spin);
984263bc
MD
613 }
614 }
615 }
616 return (0);
617}
618
619/*
620 * If sigio is on the list associated with a process or process group,
621 * disable signalling from the device, remove sigio from the list and
622 * free sigio.
623 */
624void
7bf8660a 625funsetown(struct sigio *sigio)
984263bc 626{
984263bc
MD
627 if (sigio == NULL)
628 return;
e43a034f 629 crit_enter();
984263bc 630 *(sigio->sio_myref) = NULL;
e43a034f 631 crit_exit();
984263bc
MD
632 if (sigio->sio_pgid < 0) {
633 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
634 sigio, sio_pgsigio);
635 } else /* if ((*sigiop)->sio_pgid > 0) */ {
636 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
637 sigio, sio_pgsigio);
638 }
639 crfree(sigio->sio_ucred);
efda3bd0 640 kfree(sigio, M_SIGIO);
984263bc
MD
641}
642
643/* Free a list of sigio structures. */
644void
7bf8660a 645funsetownlst(struct sigiolst *sigiolst)
984263bc
MD
646{
647 struct sigio *sigio;
648
649 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
650 funsetown(sigio);
651}
652
653/*
654 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
655 *
656 * After permission checking, add a sigio structure to the sigio list for
657 * the process or process group.
658 */
659int
7bf8660a 660fsetown(pid_t pgid, struct sigio **sigiop)
984263bc
MD
661{
662 struct proc *proc;
663 struct pgrp *pgrp;
664 struct sigio *sigio;
984263bc
MD
665
666 if (pgid == 0) {
667 funsetown(*sigiop);
668 return (0);
669 }
670 if (pgid > 0) {
671 proc = pfind(pgid);
672 if (proc == NULL)
673 return (ESRCH);
674
675 /*
676 * Policy - Don't allow a process to FSETOWN a process
677 * in another session.
678 *
679 * Remove this test to allow maximum flexibility or
680 * restrict FSETOWN to the current process or process
681 * group for maximum safety.
682 */
683 if (proc->p_session != curproc->p_session)
684 return (EPERM);
685
686 pgrp = NULL;
687 } else /* if (pgid < 0) */ {
688 pgrp = pgfind(-pgid);
689 if (pgrp == NULL)
690 return (ESRCH);
691
692 /*
693 * Policy - Don't allow a process to FSETOWN a process
694 * in another session.
695 *
696 * Remove this test to allow maximum flexibility or
697 * restrict FSETOWN to the current process or process
698 * group for maximum safety.
699 */
700 if (pgrp->pg_session != curproc->p_session)
701 return (EPERM);
702
703 proc = NULL;
704 }
705 funsetown(*sigiop);
efda3bd0 706 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
984263bc
MD
707 if (pgid > 0) {
708 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
709 sigio->sio_proc = proc;
710 } else {
711 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
712 sigio->sio_pgrp = pgrp;
713 }
714 sigio->sio_pgid = pgid;
9910d07b 715 sigio->sio_ucred = crhold(curthread->td_ucred);
984263bc 716 /* It would be convenient if p_ruid was in ucred. */
9910d07b 717 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
984263bc 718 sigio->sio_myref = sigiop;
e43a034f 719 crit_enter();
984263bc 720 *sigiop = sigio;
e43a034f 721 crit_exit();
984263bc
MD
722 return (0);
723}
724
725/*
726 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
727 */
728pid_t
7bf8660a 729fgetown(struct sigio *sigio)
984263bc
MD
730{
731 return (sigio != NULL ? sigio->sio_pgid : 0);
732}
733
734/*
4336d5df 735 * Close many file descriptors.
5b287bba
MD
736 *
737 * MPSAFE
4336d5df 738 */
4336d5df 739int
753fd850 740sys_closefrom(struct closefrom_args *uap)
4336d5df
JS
741{
742 return(kern_closefrom(uap->fd));
743}
744
5b287bba
MD
745/*
746 * Close all file descriptors greater then or equal to fd
747 *
748 * MPSAFE
749 */
4336d5df
JS
750int
751kern_closefrom(int fd)
752{
753 struct thread *td = curthread;
754 struct proc *p = td->td_proc;
755 struct filedesc *fdp;
4336d5df
JS
756
757 KKASSERT(p);
758 fdp = p->p_fd;
759
5b287bba
MD
760 if (fd < 0)
761 return (EINVAL);
5e8cfac8 762
259b8ea0
MD
763 /*
764 * NOTE: This function will skip unassociated descriptors and
765 * reserved descriptors that have not yet been assigned.
766 * fd_lastfile can change as a side effect of kern_close().
767 */
5b287bba 768 spin_lock_wr(&fdp->fd_spin);
259b8ea0
MD
769 while (fd <= fdp->fd_lastfile) {
770 if (fdp->fd_files[fd].fp != NULL) {
5b287bba
MD
771 spin_unlock_wr(&fdp->fd_spin);
772 /* ok if this races another close */
259b8ea0
MD
773 if (kern_close(fd) == EINTR)
774 return (EINTR);
5b287bba 775 spin_lock_wr(&fdp->fd_spin);
259b8ea0
MD
776 }
777 ++fd;
778 }
5b287bba 779 spin_unlock_wr(&fdp->fd_spin);
4336d5df
JS
780 return (0);
781}
782
783/*
984263bc 784 * Close a file descriptor.
5b287bba
MD
785 *
786 * MPSAFE
984263bc 787 */
984263bc 788int
753fd850 789sys_close(struct close_args *uap)
984263bc 790{
12693083
MD
791 return(kern_close(uap->fd));
792}
793
5b287bba
MD
794/*
795 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
796 */
12693083
MD
797int
798kern_close(int fd)
799{
dadab5e9
MD
800 struct thread *td = curthread;
801 struct proc *p = td->td_proc;
802 struct filedesc *fdp;
41c20dac 803 struct file *fp;
984263bc
MD
804 int error;
805 int holdleaders;
806
dadab5e9
MD
807 KKASSERT(p);
808 fdp = p->p_fd;
809
5b287bba
MD
810 spin_lock_wr(&fdp->fd_spin);
811 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
812 spin_unlock_wr(&fdp->fd_spin);
984263bc 813 return (EBADF);
5b287bba 814 }
984263bc
MD
815 holdleaders = 0;
816 if (p->p_fdtol != NULL) {
817 /*
818 * Ask fdfree() to sleep to ensure that all relevant
819 * process leaders can be traversed in closef().
820 */
821 fdp->fd_holdleaderscount++;
822 holdleaders = 1;
823 }
824
825 /*
826 * we now hold the fp reference that used to be owned by the descriptor
827 * array.
828 */
5b287bba
MD
829 spin_unlock_wr(&fdp->fd_spin);
830 if (fd < fdp->fd_knlistsize) {
831 get_mplock();
7749886d
MD
832 if (fd < fdp->fd_knlistsize)
833 knote_fdclose(p, fd);
5b287bba
MD
834 rel_mplock();
835 }
b8477cda 836 error = closef(fp, p);
984263bc 837 if (holdleaders) {
5b287bba 838 spin_lock_wr(&fdp->fd_spin);
984263bc
MD
839 fdp->fd_holdleaderscount--;
840 if (fdp->fd_holdleaderscount == 0 &&
841 fdp->fd_holdleaderswakeup != 0) {
842 fdp->fd_holdleaderswakeup = 0;
5b287bba 843 spin_unlock_wr(&fdp->fd_spin);
984263bc 844 wakeup(&fdp->fd_holdleaderscount);
5b287bba
MD
845 } else {
846 spin_unlock_wr(&fdp->fd_spin);
984263bc
MD
847 }
848 }
849 return (error);
850}
851
004d2de5
MD
852/*
853 * shutdown_args(int fd, int how)
854 */
855int
856kern_shutdown(int fd, int how)
857{
858 struct thread *td = curthread;
859 struct proc *p = td->td_proc;
004d2de5
MD
860 struct file *fp;
861 int error;
862
863 KKASSERT(p);
864
5b287bba 865 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
004d2de5 866 return (EBADF);
87de5057 867 error = fo_shutdown(fp, how);
9f87144f 868 fdrop(fp);
004d2de5
MD
869
870 return (error);
871}
872
3919ced0
MD
873/*
874 * MPALMOSTSAFE
875 */
004d2de5 876int
753fd850 877sys_shutdown(struct shutdown_args *uap)
004d2de5
MD
878{
879 int error;
880
3919ced0 881 get_mplock();
004d2de5 882 error = kern_shutdown(uap->s, uap->how);
3919ced0 883 rel_mplock();
004d2de5
MD
884
885 return (error);
886}
887
1ee6e3c6
MD
888/*
889 * MPSAFE
890 */
984263bc 891int
8f6f8622 892kern_fstat(int fd, struct stat *ub)
984263bc 893{
dadab5e9
MD
894 struct thread *td = curthread;
895 struct proc *p = td->td_proc;
41c20dac 896 struct file *fp;
984263bc
MD
897 int error;
898
dadab5e9 899 KKASSERT(p);
8f6f8622 900
5b287bba 901 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
984263bc 902 return (EBADF);
9910d07b 903 error = fo_stat(fp, ub, td->td_ucred);
9f87144f 904 fdrop(fp);
8f6f8622 905
984263bc
MD
906 return (error);
907}
984263bc
MD
908
909/*
910 * Return status information about a file descriptor.
1ee6e3c6
MD
911 *
912 * MPSAFE
984263bc 913 */
984263bc 914int
753fd850 915sys_fstat(struct fstat_args *uap)
984263bc 916{
8f6f8622 917 struct stat st;
984263bc
MD
918 int error;
919
8f6f8622
DRJ
920 error = kern_fstat(uap->fd, &st);
921
984263bc 922 if (error == 0)
8f6f8622 923 error = copyout(&st, uap->sb, sizeof(st));
984263bc
MD
924 return (error);
925}
926
927/*
984263bc 928 * Return pathconf information about a file descriptor.
3919ced0
MD
929 *
930 * MPALMOSTSAFE
984263bc 931 */
984263bc 932int
753fd850 933sys_fpathconf(struct fpathconf_args *uap)
984263bc 934{
dadab5e9
MD
935 struct thread *td = curthread;
936 struct proc *p = td->td_proc;
984263bc
MD
937 struct file *fp;
938 struct vnode *vp;
939 int error = 0;
940
5b287bba
MD
941 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
942 return (EBADF);
984263bc
MD
943
944 switch (fp->f_type) {
945 case DTYPE_PIPE:
946 case DTYPE_SOCKET:
947 if (uap->name != _PC_PIPE_BUF) {
948 error = EINVAL;
949 } else {
c7114eea 950 uap->sysmsg_result = PIPE_BUF;
984263bc
MD
951 error = 0;
952 }
953 break;
954 case DTYPE_FIFO:
955 case DTYPE_VNODE:
956 vp = (struct vnode *)fp->f_data;
3919ced0 957 get_mplock();
d557216f 958 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
3919ced0 959 rel_mplock();
984263bc
MD
960 break;
961 default:
962 error = EOPNOTSUPP;
963 break;
964 }
9f87144f 965 fdrop(fp);
984263bc
MD
966 return(error);
967}
968
984263bc
MD
969static int fdexpand;
970SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, "");
971
5b287bba
MD
972/*
973 * Grow the file table so it can hold through descriptor (want).
974 *
975 * The fdp's spinlock must be held exclusively on entry and may be held
976 * exclusively on return. The spinlock may be cycled by the routine.
977 *
978 * MPSAFE
979 */
69908319 980static void
5b287bba 981fdgrow_locked(struct filedesc *fdp, int want)
69908319 982{
0679adc4
MD
983 struct fdnode *newfiles;
984 struct fdnode *oldfiles;
69908319
JH
985 int nf, extra;
986
987 nf = fdp->fd_nfiles;
988 do {
989 /* nf has to be of the form 2^n - 1 */
990 nf = 2 * nf + 1;
991 } while (nf <= want);
992
5b287bba 993 spin_unlock_wr(&fdp->fd_spin);
efda3bd0 994 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
5b287bba 995 spin_lock_wr(&fdp->fd_spin);
69908319
JH
996
997 /*
5b287bba
MD
998 * We could have raced another extend while we were not holding
999 * the spinlock.
69908319
JH
1000 */
1001 if (fdp->fd_nfiles >= nf) {
5b287bba 1002 spin_unlock_wr(&fdp->fd_spin);
efda3bd0 1003 kfree(newfiles, M_FILEDESC);
5b287bba 1004 spin_lock_wr(&fdp->fd_spin);
69908319
JH
1005 return;
1006 }
69908319
JH
1007 /*
1008 * Copy the existing ofile and ofileflags arrays
1009 * and zero the new portion of each array.
1010 */
1011 extra = nf - fdp->fd_nfiles;
0679adc4
MD
1012 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1013 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1014
1015 oldfiles = fdp->fd_files;
1016 fdp->fd_files = newfiles;
69908319 1017 fdp->fd_nfiles = nf;
0679adc4 1018
5b287bba
MD
1019 if (oldfiles != fdp->fd_builtin_files) {
1020 spin_unlock_wr(&fdp->fd_spin);
efda3bd0 1021 kfree(oldfiles, M_FILEDESC);
5b287bba
MD
1022 spin_lock_wr(&fdp->fd_spin);
1023 }
69908319
JH
1024 fdexpand++;
1025}
1026
1027/*
1028 * Number of nodes in right subtree, including the root.
1029 */
1030static __inline int
1031right_subtree_size(int n)
1032{
1033 return (n ^ (n | (n + 1)));
1034}
1035
1036/*
1037 * Bigger ancestor.
1038 */
1039static __inline int
1040right_ancestor(int n)
1041{
1042 return (n | (n + 1));
1043}
1044
1045/*
1046 * Smaller ancestor.
1047 */
1048static __inline int
1049left_ancestor(int n)
1050{
1051 return ((n & (n + 1)) - 1);
1052}
1053
5b287bba
MD
1054/*
1055 * Traverse the in-place binary tree buttom-up adjusting the allocation
1056 * count so scans can determine where free descriptors are located.
1057 *
1058 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1059 */
fa541be6 1060static
69908319 1061void
5b287bba 1062fdreserve_locked(struct filedesc *fdp, int fd, int incr)
69908319
JH
1063{
1064 while (fd >= 0) {
0679adc4
MD
1065 fdp->fd_files[fd].allocated += incr;
1066 KKASSERT(fdp->fd_files[fd].allocated >= 0);
69908319
JH
1067 fd = left_ancestor(fd);
1068 }
1069}
1070
1071/*
259b8ea0
MD
1072 * Reserve a file descriptor for the process. If no error occurs, the
1073 * caller MUST at some point call fsetfd() or assign a file pointer
1074 * or dispose of the reservation.
5b287bba
MD
1075 *
1076 * MPSAFE
69908319 1077 */
984263bc 1078int
dda4b42b 1079fdalloc(struct proc *p, int want, int *result)
984263bc 1080{
41c20dac 1081 struct filedesc *fdp = p->p_fd;
69908319
JH
1082 int fd, rsize, rsum, node, lim;
1083
5b287bba 1084 spin_lock_rd(&p->p_limit->p_spin);
69908319 1085 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
5b287bba 1086 spin_unlock_rd(&p->p_limit->p_spin);
69908319
JH
1087 if (want >= lim)
1088 return (EMFILE);
5b287bba 1089 spin_lock_wr(&fdp->fd_spin);
69908319 1090 if (want >= fdp->fd_nfiles)
5b287bba 1091 fdgrow_locked(fdp, want);
984263bc
MD
1092
1093 /*
1094 * Search for a free descriptor starting at the higher
1095 * of want or fd_freefile. If that fails, consider
1096 * expanding the ofile array.
259b8ea0
MD
1097 *
1098 * NOTE! the 'allocated' field is a cumulative recursive allocation
1099 * count. If we happen to see a value of 0 then we can shortcut
1100 * our search. Otherwise we run through through the tree going
1101 * down branches we know have free descriptor(s) until we hit a
1102 * leaf node. The leaf node will be free but will not necessarily
1103 * have an allocated field of 0.
984263bc 1104 */
69908319
JH
1105retry:
1106 /* move up the tree looking for a subtree with a free node */
1107 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1108 fd = right_ancestor(fd)) {
0679adc4 1109 if (fdp->fd_files[fd].allocated == 0)
69908319 1110 goto found;
984263bc 1111
69908319 1112 rsize = right_subtree_size(fd);
0679adc4 1113 if (fdp->fd_files[fd].allocated == rsize)
69908319 1114 continue; /* right subtree full */
984263bc
MD
1115
1116 /*
69908319
JH
1117 * Free fd is in the right subtree of the tree rooted at fd.
1118 * Call that subtree R. Look for the smallest (leftmost)
1119 * subtree of R with an unallocated fd: continue moving
1120 * down the left branch until encountering a full left
1121 * subtree, then move to the right.
984263bc 1122 */
69908319
JH
1123 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1124 node = fd + rsize;
0679adc4
MD
1125 rsum += fdp->fd_files[node].allocated;
1126 if (fdp->fd_files[fd].allocated == rsum + rsize) {
69908319 1127 fd = node; /* move to the right */
0679adc4 1128 if (fdp->fd_files[node].allocated == 0)
69908319
JH
1129 goto found;
1130 rsum = 0;
1131 }
984263bc 1132 }
69908319 1133 goto found;
984263bc 1134 }
69908319
JH
1135
1136 /*
1137 * No space in current array. Expand?
1138 */
5b287bba
MD
1139 if (fdp->fd_nfiles >= lim) {
1140 spin_unlock_wr(&fdp->fd_spin);
69908319 1141 return (EMFILE);
5b287bba
MD
1142 }
1143 fdgrow_locked(fdp, want);
69908319
JH
1144 goto retry;
1145
1146found:
1147 KKASSERT(fd < fdp->fd_nfiles);
69908319
JH
1148 if (fd > fdp->fd_lastfile)
1149 fdp->fd_lastfile = fd;
1150 if (want <= fdp->fd_freefile)
1151 fdp->fd_freefile = fd;
1152 *result = fd;
0679adc4 1153 KKASSERT(fdp->fd_files[fd].fp == NULL);
259b8ea0
MD
1154 KKASSERT(fdp->fd_files[fd].reserved == 0);
1155 fdp->fd_files[fd].fileflags = 0;
1156 fdp->fd_files[fd].reserved = 1;
5b287bba
MD
1157 fdreserve_locked(fdp, fd, 1);
1158 spin_unlock_wr(&fdp->fd_spin);
984263bc
MD
1159 return (0);
1160}
1161
1162/*
1163 * Check to see whether n user file descriptors
1164 * are available to the process p.
5b287bba
MD
1165 *
1166 * MPSAFE
984263bc
MD
1167 */
1168int
7bf8660a 1169fdavail(struct proc *p, int n)
984263bc 1170{
41c20dac 1171 struct filedesc *fdp = p->p_fd;
0679adc4 1172 struct fdnode *fdnode;
41c20dac 1173 int i, lim, last;
984263bc 1174
5b287bba 1175 spin_lock_rd(&p->p_limit->p_spin);
984263bc 1176 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfilesperproc);
5b287bba 1177 spin_unlock_rd(&p->p_limit->p_spin);
984263bc 1178
5b287bba
MD
1179 spin_lock_rd(&fdp->fd_spin);
1180 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1181 spin_unlock_rd(&fdp->fd_spin);
1182 return (1);
1183 }
984263bc 1184 last = min(fdp->fd_nfiles, lim);
0679adc4
MD
1185 fdnode = &fdp->fd_files[fdp->fd_freefile];
1186 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
5b287bba
MD
1187 if (fdnode->fp == NULL && --n <= 0) {
1188 spin_unlock_rd(&fdp->fd_spin);
984263bc 1189 return (1);
5b287bba 1190 }
984263bc 1191 }
5b287bba 1192 spin_unlock_rd(&fdp->fd_spin);
984263bc
MD
1193 return (0);
1194}
1195
1196/*
b8477cda
MD
1197 * Revoke open descriptors referencing (f_data, f_type)
1198 *
1199 * Any revoke executed within a prison is only able to
1200 * revoke descriptors for processes within that prison.
1201 *
1202 * Returns 0 on success or an error code.
1203 */
1204struct fdrevoke_info {
1205 void *data;
1206 short type;
1207 short unused;
1208 int count;
ea8f324c 1209 int intransit;
b8477cda
MD
1210 struct ucred *cred;
1211 struct file *nfp;
1212};
1213
ea8f324c
MD
1214static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1215static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
b8477cda
MD
1216
1217int
1218fdrevoke(void *f_data, short f_type, struct ucred *cred)
1219{
1220 struct fdrevoke_info info;
1221 int error;
1222
1223 bzero(&info, sizeof(info));
1224 info.data = f_data;
1225 info.type = f_type;
1226 info.cred = cred;
1227 error = falloc(NULL, &info.nfp, NULL);
1228 if (error)
1229 return (error);
ea8f324c
MD
1230
1231 /*
1232 * Scan the file pointer table once. dups do not dup file pointers,
1233 * only descriptors, so there is no leak. Set FREVOKED on the fps
1234 * being revoked.
1235 */
1236 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1237
1238 /*
1239 * If any fps were marked track down the related descriptors
1240 * and close them. Any dup()s at this point will notice
1241 * the FREVOKED already set in the fp and do the right thing.
1242 *
1243 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1244 * socket) bumped the intransit counter and will require a
1245 * scan. Races against fps leaving the socket are closed by
1246 * the socket code checking for FREVOKED.
1247 */
1248 if (info.count)
1249 allproc_scan(fdrevoke_proc_callback, &info);
1250 if (info.intransit)
1251 unp_revoke_gc(info.nfp);
b8477cda 1252 fdrop(info.nfp);
ea8f324c 1253 return(0);
b8477cda
MD
1254}
1255
ea8f324c
MD
1256/*
1257 * Locate matching file pointers directly.
1258 */
b8477cda 1259static int
ea8f324c
MD
1260fdrevoke_check_callback(struct file *fp, void *vinfo)
1261{
1262 struct fdrevoke_info *info = vinfo;
1263
1264 /*
1265 * File pointers already flagged for revokation are skipped.
1266 */
1267 if (fp->f_flag & FREVOKED)
1268 return(0);
1269
1270 /*
1271 * If revoking from a prison file pointers created outside of
1272 * that prison, or file pointers without creds, cannot be revoked.
1273 */
1274 if (info->cred->cr_prison &&
1275 (fp->f_cred == NULL ||
1276 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1277 return(0);
1278 }
1279
1280 /*
1281 * If the file pointer matches then mark it for revocation. The
1282 * flag is currently only used by unp_revoke_gc().
1283 *
1284 * info->count is a heuristic and can race in a SMP environment.
1285 */
1286 if (info->data == fp->f_data && info->type == fp->f_type) {
1287 atomic_set_int(&fp->f_flag, FREVOKED);
1288 info->count += fp->f_count;
1289 if (fp->f_msgcount)
1290 ++info->intransit;
1291 }
1292 return(0);
1293}
1294
1295/*
1296 * Locate matching file pointers via process descriptor tables.
1297 */
1298static int
1299fdrevoke_proc_callback(struct proc *p, void *vinfo)
b8477cda
MD
1300{
1301 struct fdrevoke_info *info = vinfo;
1302 struct filedesc *fdp;
1303 struct file *fp;
1304 int n;
1305
1306 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1307 return(0);
1308 if (info->cred->cr_prison &&
1309 info->cred->cr_prison != p->p_ucred->cr_prison) {
1310 return(0);
1311 }
1312
1313 /*
1314 * If the controlling terminal of the process matches the
1315 * vnode being revoked we clear the controlling terminal.
1316 *
1317 * The normal spec_close() may not catch this because it
1318 * uses curproc instead of p.
1319 */
1320 if (p->p_session && info->type == DTYPE_VNODE &&
1321 info->data == p->p_session->s_ttyvp) {
1322 p->p_session->s_ttyvp = NULL;
1323 vrele(info->data);
1324 }
1325
1326 /*
0a4a9c77 1327 * Softref the fdp to prevent it from being destroyed
b8477cda 1328 */
0a4a9c77
MD
1329 spin_lock_wr(&p->p_spin);
1330 if ((fdp = p->p_fd) == NULL) {
1331 spin_unlock_wr(&p->p_spin);
b8477cda 1332 return(0);
0a4a9c77
MD
1333 }
1334 atomic_add_int(&fdp->fd_softrefs, 1);
1335 spin_unlock_wr(&p->p_spin);
1336
1337 /*
1338 * Locate and close any matching file descriptors.
1339 */
b8477cda
MD
1340 spin_lock_wr(&fdp->fd_spin);
1341 for (n = 0; n < fdp->fd_nfiles; ++n) {
1342 if ((fp = fdp->fd_files[n].fp) == NULL)
1343 continue;
ea8f324c 1344 if (fp->f_flag & FREVOKED) {
b8477cda
MD
1345 fhold(info->nfp);
1346 fdp->fd_files[n].fp = info->nfp;
1347 spin_unlock_wr(&fdp->fd_spin);
1348 closef(fp, p);
1349 spin_lock_wr(&fdp->fd_spin);
ea8f324c 1350 --info->count;
b8477cda
MD
1351 }
1352 }
1353 spin_unlock_wr(&fdp->fd_spin);
0a4a9c77 1354 atomic_subtract_int(&fdp->fd_softrefs, 1);
b8477cda
MD
1355 return(0);
1356}
1357
1358/*
39f91578 1359 * falloc:
259b8ea0
MD
1360 * Create a new open file structure and reserve a file decriptor
1361 * for the process that refers to it.
1362 *
1363 * Root creds are checked using p, or assumed if p is NULL. If
1364 * resultfd is non-NULL then p must also be non-NULL. No file
1365 * descriptor is reserved if resultfd is NULL.
fad57d0e 1366 *
259b8ea0
MD
1367 * A file pointer with a refcount of 1 is returned. Note that the
1368 * file pointer is NOT associated with the descriptor. If falloc
1369 * returns success, fsetfd() MUST be called to either associate the
1370 * file pointer or clear the reservation.
5b287bba 1371 *
2dd63755 1372 * MPSAFE
984263bc
MD
1373 */
1374int
39f91578 1375falloc(struct proc *p, struct file **resultfp, int *resultfd)
984263bc 1376{
c4cb6d8b
HP
1377 static struct timeval lastfail;
1378 static int curfail;
fad57d0e
MD
1379 struct file *fp;
1380 int error;
c4cb6d8b 1381
fad57d0e
MD
1382 fp = NULL;
1383
1384 /*
1385 * Handle filetable full issues and root overfill.
1386 */
60ee93b9 1387 if (nfiles >= maxfiles - maxfilesrootres &&
9b3f1fd5 1388 ((p && p->p_ucred->cr_ruid != 0) || nfiles >= maxfiles)) {
c4cb6d8b 1389 if (ppsratecheck(&lastfail, &curfail, 1)) {
6ea70f76 1390 kprintf("kern.maxfiles limit exceeded by uid %d, please see tuning(7).\n",
9b3f1fd5 1391 (p ? p->p_ucred->cr_ruid : -1));
c4cb6d8b 1392 }
fad57d0e
MD
1393 error = ENFILE;
1394 goto done;
984263bc 1395 }
fad57d0e 1396
984263bc
MD
1397 /*
1398 * Allocate a new file descriptor.
984263bc 1399 */
efda3bd0 1400 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
85fceac1 1401 spin_init(&fp->f_spin);
984263bc 1402 fp->f_count = 1;
984263bc
MD
1403 fp->f_ops = &badfileops;
1404 fp->f_seqcount = 1;
fad57d0e 1405 if (p)
a235f7bb 1406 fp->f_cred = crhold(p->p_ucred);
fad57d0e
MD
1407 else
1408 fp->f_cred = crhold(proc0.p_ucred);
2dd63755
MD
1409 spin_lock_wr(&filehead_spin);
1410 nfiles++;
fad57d0e 1411 LIST_INSERT_HEAD(&filehead, fp, f_list);
2dd63755 1412 spin_unlock_wr(&filehead_spin);
fad57d0e 1413 if (resultfd) {
259b8ea0 1414 if ((error = fdalloc(p, 0, resultfd)) != 0) {
9f87144f 1415 fdrop(fp);
fad57d0e 1416 fp = NULL;
39f91578 1417 }
a235f7bb 1418 } else {
fad57d0e 1419 error = 0;
984263bc 1420 }
fad57d0e
MD
1421done:
1422 *resultfp = fp;
64f33bc8 1423 return (error);
fad57d0e
MD
1424}
1425
1426/*
85fceac1
MD
1427 * MPSAFE
1428 */
1429static
1430int
1431checkfpclosed(struct filedesc *fdp, int fd, struct file *fp)
1432{
1433 int error;
1434
1435 spin_lock_rd(&fdp->fd_spin);
1436 if ((unsigned) fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1437 error = EBADF;
1438 else
1439 error = 0;
1440 spin_unlock_rd(&fdp->fd_spin);
1441 return (error);
1442}
1443
1444/*
259b8ea0
MD
1445 * Associate a file pointer with a previously reserved file descriptor.
1446 * This function always succeeds.
1447 *
1448 * If fp is NULL, the file descriptor is returned to the pool.
fad57d0e 1449 */
fad57d0e 1450
5b287bba
MD
1451/*
1452 * MPSAFE (exclusive spinlock must be held on call)
1453 */
1454static void
1455fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1456{
259b8ea0
MD
1457 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1458 KKASSERT(fdp->fd_files[fd].reserved != 0);
1459 if (fp) {
fad57d0e 1460 fhold(fp);
259b8ea0
MD
1461 fdp->fd_files[fd].fp = fp;
1462 fdp->fd_files[fd].reserved = 0;
d9b2033e
MD
1463 if (fp->f_type == DTYPE_KQUEUE) {
1464 if (fdp->fd_knlistsize < 0)
1465 fdp->fd_knlistsize = 0;
1466 }
259b8ea0
MD
1467 } else {
1468 fdp->fd_files[fd].reserved = 0;
5b287bba
MD
1469 fdreserve_locked(fdp, fd, -1);
1470 fdfixup_locked(fdp, fd);
fad57d0e 1471 }
984263bc
MD
1472}
1473
5b287bba
MD
1474/*
1475 * MPSAFE
1476 */
1477void
1478fsetfd(struct proc *p, struct file *fp, int fd)
1479{
1480 struct filedesc *fdp = p->p_fd;
1481
1482 spin_lock_wr(&fdp->fd_spin);
1483 fsetfd_locked(fdp, fp, fd);
1484 spin_unlock_wr(&fdp->fd_spin);
1485}
1486
1487/*
1488 * MPSAFE (exclusive spinlock must be held on call)
1489 */
fa541be6 1490static
259b8ea0 1491struct file *
5b287bba 1492funsetfd_locked(struct filedesc *fdp, int fd)
69908319 1493{
259b8ea0
MD
1494 struct file *fp;
1495
1496 if ((unsigned)fd >= fdp->fd_nfiles)
1497 return (NULL);
1498 if ((fp = fdp->fd_files[fd].fp) == NULL)
1499 return (NULL);
0679adc4
MD
1500 fdp->fd_files[fd].fp = NULL;
1501 fdp->fd_files[fd].fileflags = 0;
259b8ea0 1502
5b287bba
MD
1503 fdreserve_locked(fdp, fd, -1);
1504 fdfixup_locked(fdp, fd);
259b8ea0 1505 return(fp);
69908319
JH
1506}
1507
85fceac1
MD
1508/*
1509 * MPSAFE
1510 */
1511int
1512fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1513{
1514 int error;
1515
1516 spin_lock_rd(&fdp->fd_spin);
1517 if (((u_int)fd) >= fdp->fd_nfiles) {
1518 error = EBADF;
1519 } else if (fdp->fd_files[fd].fp == NULL) {
1520 error = EBADF;
1521 } else {
1522 *flagsp = fdp->fd_files[fd].fileflags;
1523 error = 0;
1524 }
1525 spin_unlock_rd(&fdp->fd_spin);
1526 return (error);
1527}
1528
1529/*
1530 * MPSAFE
1531 */
228b401d
MD
1532int
1533fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1534{
85fceac1
MD
1535 int error;
1536
1537 spin_lock_wr(&fdp->fd_spin);
1538 if (((u_int)fd) >= fdp->fd_nfiles) {
1539 error = EBADF;
1540 } else if (fdp->fd_files[fd].fp == NULL) {
1541 error = EBADF;
1542 } else {
1543 fdp->fd_files[fd].fileflags |= add_flags;
1544 error = 0;
1545 }
1546 spin_unlock_wr(&fdp->fd_spin);
1547 return (error);
228b401d
MD
1548}
1549
85fceac1
MD
1550/*
1551 * MPSAFE
1552 */
228b401d
MD
1553int
1554fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1555{
85fceac1
MD
1556 int error;
1557
1558 spin_lock_wr(&fdp->fd_spin);
1559 if (((u_int)fd) >= fdp->fd_nfiles) {
1560 error = EBADF;
1561 } else if (fdp->fd_files[fd].fp == NULL) {
1562 error = EBADF;
1563 } else {
1564 fdp->fd_files[fd].fileflags &= ~rem_flags;
1565 error = 0;
1566 }
1567 spin_unlock_wr(&fdp->fd_spin);
1568 return (error);
228b401d
MD
1569}
1570
69908319 1571void
a235f7bb
MD
1572fsetcred(struct file *fp, struct ucred *cr)
1573{
1574 crhold(cr);
1575 crfree(fp->f_cred);
1576 fp->f_cred = cr;
1577}
1578
984263bc
MD
1579/*
1580 * Free a file descriptor.
1581 */
5b287bba 1582static
984263bc 1583void
7bf8660a 1584ffree(struct file *fp)
984263bc
MD
1585{
1586 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
2dd63755 1587 spin_lock_wr(&filehead_spin);
984263bc 1588 LIST_REMOVE(fp, f_list);
2dd63755
MD
1589 nfiles--;
1590 spin_unlock_wr(&filehead_spin);
984263bc 1591 crfree(fp->f_cred);
28623bf9
MD
1592 if (fp->f_nchandle.ncp)
1593 cache_drop(&fp->f_nchandle);
efda3bd0 1594 kfree(fp, M_FILE);
984263bc
MD
1595}
1596
1597/*
228b401d
MD
1598 * called from init_main, initialize filedesc0 for proc0.
1599 */
1600void
1601fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1602{
1603 p0->p_fd = fdp0;
1604 p0->p_fdtol = NULL;
1605 fdp0->fd_refcnt = 1;
1606 fdp0->fd_cmask = cmask;
1607 fdp0->fd_files = fdp0->fd_builtin_files;
1608 fdp0->fd_nfiles = NDFILE;
259b8ea0 1609 fdp0->fd_lastfile = -1;
85fceac1 1610 spin_init(&fdp0->fd_spin);
228b401d
MD
1611}
1612
1613/*
984263bc 1614 * Build a new filedesc structure.
5b287bba
MD
1615 *
1616 * NOT MPSAFE (vref)
984263bc
MD
1617 */
1618struct filedesc *
7bf8660a 1619fdinit(struct proc *p)
984263bc 1620{
0679adc4 1621 struct filedesc *newfdp;
41c20dac 1622 struct filedesc *fdp = p->p_fd;
984263bc 1623
efda3bd0 1624 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
5b287bba 1625 spin_lock_rd(&fdp->fd_spin);
bccde7a3 1626 if (fdp->fd_cdir) {
0679adc4
MD
1627 newfdp->fd_cdir = fdp->fd_cdir;
1628 vref(newfdp->fd_cdir);
28623bf9 1629 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
690a3127 1630 }
bccde7a3
MD
1631
1632 /*
1633 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1634 * proc0, but should unconditionally exist in other processes.
1635 */
1636 if (fdp->fd_rdir) {
0679adc4
MD
1637 newfdp->fd_rdir = fdp->fd_rdir;
1638 vref(newfdp->fd_rdir);
28623bf9 1639 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
bccde7a3
MD
1640 }
1641 if (fdp->fd_jdir) {
0679adc4
MD
1642 newfdp->fd_jdir = fdp->fd_jdir;
1643 vref(newfdp->fd_jdir);
28623bf9 1644 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
690a3127 1645 }
5b287bba 1646 spin_unlock_rd(&fdp->fd_spin);
690a3127 1647
984263bc 1648 /* Create the file descriptor table. */
0679adc4
MD
1649 newfdp->fd_refcnt = 1;
1650 newfdp->fd_cmask = cmask;
1651 newfdp->fd_files = newfdp->fd_builtin_files;
1652 newfdp->fd_nfiles = NDFILE;
1653 newfdp->fd_knlistsize = -1;
259b8ea0 1654 newfdp->fd_lastfile = -1;
85fceac1 1655 spin_init(&newfdp->fd_spin);
0679adc4
MD
1656
1657 return (newfdp);
984263bc
MD
1658}
1659
1660/*
1661 * Share a filedesc structure.
5b287bba
MD
1662 *
1663 * MPSAFE
984263bc
MD
1664 */
1665struct filedesc *
7bf8660a 1666fdshare(struct proc *p)
984263bc 1667{
5b287bba
MD
1668 struct filedesc *fdp;
1669
1670 fdp = p->p_fd;
1671 spin_lock_wr(&fdp->fd_spin);
1672 fdp->fd_refcnt++;
1673 spin_unlock_wr(&fdp->fd_spin);
1674 return (fdp);
984263bc
MD
1675}
1676
1677/*
1678 * Copy a filedesc structure.
5b287bba
MD
1679 *
1680 * MPSAFE
984263bc
MD
1681 */
1682struct filedesc *
7bf8660a 1683fdcopy(struct proc *p)
984263bc 1684{
5b287bba
MD
1685 struct filedesc *fdp = p->p_fd;
1686 struct filedesc *newfdp;
0679adc4 1687 struct fdnode *fdnode;
41c20dac 1688 int i;
5b287bba 1689 int ni;
984263bc 1690
5b287bba
MD
1691 /*
1692 * Certain daemons might not have file descriptors.
1693 */
984263bc
MD
1694 if (fdp == NULL)
1695 return (NULL);
1696
5b287bba
MD
1697 /*
1698 * Allocate the new filedesc and fd_files[] array. This can race
1699 * with operations by other threads on the fdp so we have to be
1700 * careful.
1701 */
efda3bd0 1702 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
5b287bba
MD
1703again:
1704 spin_lock_rd(&fdp->fd_spin);
1705 if (fdp->fd_lastfile < NDFILE) {
1706 newfdp->fd_files = newfdp->fd_builtin_files;
1707 i = NDFILE;
1708 } else {
1709 /*
1710 * We have to allocate (N^2-1) entries for our in-place
1711 * binary tree. Allow the table to shrink.
1712 */
1713 i = fdp->fd_nfiles;
1714 ni = (i - 1) / 2;
1715 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1716 i = ni;
1717 ni = (i - 1) / 2;
1718 }
1719 spin_unlock_rd(&fdp->fd_spin);
77652cad 1720 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
5b287bba
MD
1721 M_FILEDESC, M_WAITOK | M_ZERO);
1722
1723 /*
1724 * Check for race, retry
1725 */
1726 spin_lock_rd(&fdp->fd_spin);
1727 if (i <= fdp->fd_lastfile) {
1728 spin_unlock_rd(&fdp->fd_spin);
efda3bd0 1729 kfree(newfdp->fd_files, M_FILEDESC);
5b287bba
MD
1730 goto again;
1731 }
1732 }
1733
1734 /*
1735 * Dup the remaining fields. vref() and cache_hold() can be
1736 * safely called while holding the read spinlock on fdp.
1737 *
1738 * The read spinlock on fdp is still being held.
1739 *
1740 * NOTE: vref and cache_hold calls for the case where the vnode
1741 * or cache entry already has at least one ref may be called
1742 * while holding spin locks.
1743 */
1744 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
597aea93 1745 vref(newfdp->fd_cdir);
28623bf9 1746 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
690a3127 1747 }
6bdbb368
DR
1748 /*
1749 * We must check for fd_rdir here, at least for now because
1750 * the init process is created before we have access to the
1751 * rootvode to take a reference to it.
1752 */
5b287bba 1753 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
597aea93 1754 vref(newfdp->fd_rdir);
28623bf9 1755 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
690a3127 1756 }
5b287bba 1757 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
597aea93 1758 vref(newfdp->fd_jdir);
28623bf9 1759 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
690a3127 1760 }
984263bc 1761 newfdp->fd_refcnt = 1;
984263bc 1762 newfdp->fd_nfiles = i;
5b287bba
MD
1763 newfdp->fd_lastfile = fdp->fd_lastfile;
1764 newfdp->fd_freefile = fdp->fd_freefile;
1765 newfdp->fd_cmask = fdp->fd_cmask;
1766 newfdp->fd_knlist = NULL;
1767 newfdp->fd_knlistsize = -1;
1768 newfdp->fd_knhash = NULL;
1769 newfdp->fd_knhashmask = 0;
1770 spin_init(&newfdp->fd_spin);
984263bc
MD
1771
1772 /*
5b287bba
MD
1773 * Copy the descriptor table through (i). This also copies the
1774 * allocation state. Then go through and ref the file pointers
1775 * and clean up any KQ descriptors.
1776 *
259b8ea0
MD
1777 * kq descriptors cannot be copied. Since we haven't ref'd the
1778 * copied files yet we can ignore the return value from funsetfd().
5b287bba
MD
1779 *
1780 * The read spinlock on fdp is still being held.
984263bc 1781 */
5b287bba
MD
1782 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1783 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
259b8ea0
MD
1784 fdnode = &newfdp->fd_files[i];
1785 if (fdnode->reserved) {
5b287bba 1786 fdreserve_locked(newfdp, i, -1);
259b8ea0 1787 fdnode->reserved = 0;
5b287bba
MD
1788 fdfixup_locked(newfdp, i);
1789 } else if (fdnode->fp) {
1790 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1791 (void)funsetfd_locked(newfdp, i);
1792 } else {
1793 fhold(fdnode->fp);
1794 }
259b8ea0 1795 }
984263bc 1796 }
5b287bba 1797 spin_unlock_rd(&fdp->fd_spin);
984263bc
MD
1798 return (newfdp);
1799}
1800
1801/*
1802 * Release a filedesc structure.
5b287bba
MD
1803 *
1804 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
984263bc
MD
1805 */
1806void
0a4a9c77 1807fdfree(struct proc *p, struct filedesc *repl)
984263bc 1808{
0a4a9c77 1809 struct filedesc *fdp;
0679adc4 1810 struct fdnode *fdnode;
41c20dac 1811 int i;
984263bc
MD
1812 struct filedesc_to_leader *fdtol;
1813 struct file *fp;
1814 struct vnode *vp;
1815 struct flock lf;
1816
0a4a9c77 1817 /*
0a4a9c77
MD
1818 * Certain daemons might not have file descriptors.
1819 */
464e801e
MD
1820 fdp = p->p_fd;
1821 if (fdp == NULL) {
1822 p->p_fd = repl;
984263bc 1823 return;
464e801e 1824 }
984263bc 1825
5b287bba 1826 /*
0a4a9c77 1827 * Severe messing around to follow.
5b287bba
MD
1828 */
1829 spin_lock_wr(&fdp->fd_spin);
1830
984263bc
MD
1831 /* Check for special need to clear POSIX style locks */
1832 fdtol = p->p_fdtol;
1833 if (fdtol != NULL) {
1834 KASSERT(fdtol->fdl_refcount > 0,
1835 ("filedesc_to_refcount botch: fdl_refcount=%d",
1836 fdtol->fdl_refcount));
1837 if (fdtol->fdl_refcount == 1 &&
1838 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
259b8ea0
MD
1839 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1840 fdnode = &fdp->fd_files[i];
0679adc4 1841 if (fdnode->fp == NULL ||
259b8ea0 1842 fdnode->fp->f_type != DTYPE_VNODE) {
984263bc 1843 continue;
259b8ea0 1844 }
0679adc4 1845 fp = fdnode->fp;
984263bc 1846 fhold(fp);
5b287bba
MD
1847 spin_unlock_wr(&fdp->fd_spin);
1848
984263bc
MD
1849 lf.l_whence = SEEK_SET;
1850 lf.l_start = 0;
1851 lf.l_len = 0;
1852 lf.l_type = F_UNLCK;
1853 vp = (struct vnode *)fp->f_data;
1854 (void) VOP_ADVLOCK(vp,
1855 (caddr_t)p->p_leader,
1856 F_UNLCK,
1857 &lf,
1858 F_POSIX);
9f87144f 1859 fdrop(fp);
5b287bba 1860 spin_lock_wr(&fdp->fd_spin);
984263bc
MD
1861 }
1862 }
1863 retry:
1864 if (fdtol->fdl_refcount == 1) {
1865 if (fdp->fd_holdleaderscount > 0 &&
1866 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1867 /*
1868 * close() or do_dup() has cleared a reference
1869 * in a shared file descriptor table.
1870 */
1871 fdp->fd_holdleaderswakeup = 1;
e590ee86 1872 ssleep(&fdp->fd_holdleaderscount,
7749886d 1873 &fdp->fd_spin, 0, "fdlhold", 0);
984263bc
MD
1874 goto retry;
1875 }
1876 if (fdtol->fdl_holdcount > 0) {
1877 /*
1878 * Ensure that fdtol->fdl_leader
1879 * remains valid in closef().
1880 */
1881 fdtol->fdl_wakeup = 1;
e590ee86 1882 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
984263bc
MD
1883 goto retry;
1884 }
1885 }
1886 fdtol->fdl_refcount--;
1887 if (fdtol->fdl_refcount == 0 &&
1888 fdtol->fdl_holdcount == 0) {
1889 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1890 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
5b287bba 1891 } else {
984263bc 1892 fdtol = NULL;
5b287bba 1893 }
984263bc 1894 p->p_fdtol = NULL;
5b287bba
MD
1895 if (fdtol != NULL) {
1896 spin_unlock_wr(&fdp->fd_spin);
efda3bd0 1897 kfree(fdtol, M_FILEDESC_TO_LEADER);
5b287bba
MD
1898 spin_lock_wr(&fdp->fd_spin);
1899 }
984263bc 1900 }
5b287bba
MD
1901 if (--fdp->fd_refcnt > 0) {
1902 spin_unlock_wr(&fdp->fd_spin);
464e801e
MD
1903 spin_lock_wr(&p->p_spin);
1904 p->p_fd = repl;
1905 spin_unlock_wr(&p->p_spin);
984263bc 1906 return;
5b287bba 1907 }
464e801e
MD
1908
1909 /*
1910 * Even though we are the last reference to the structure allproc
1911 * scans may still reference the structure. Maintain proper
1912 * locks until we can replace p->p_fd.
1913 *
1914 * Also note that kqueue's closef still needs to reference the
1915 * fdp via p->p_fd, so we have to close the descriptors before
1916 * we replace p->p_fd.
1917 */
1918 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1919 if (fdp->fd_files[i].fp) {
1920 fp = funsetfd_locked(fdp, i);
1921 if (fp) {
1922 spin_unlock_wr(&fdp->fd_spin);
1923 closef(fp, p);
1924 spin_lock_wr(&fdp->fd_spin);
1925 }
1926 }
1927 }
5b287bba
MD
1928 spin_unlock_wr(&fdp->fd_spin);
1929
984263bc 1930 /*
464e801e
MD
1931 * Interlock against an allproc scan operations (typically frevoke).
1932 */
1933 spin_lock_wr(&p->p_spin);
1934 p->p_fd = repl;
1935 spin_unlock_wr(&p->p_spin);
1936
1937 /*
0a4a9c77
MD
1938 * Wait for any softrefs to go away. This race rarely occurs so
1939 * we can use a non-critical-path style poll/sleep loop. The
1940 * race only occurs against allproc scans.
1941 *
1942 * No new softrefs can occur with the fdp disconnected from the
1943 * process.
1944 */
1945 if (fdp->fd_softrefs) {
1946 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
1947 while (fdp->fd_softrefs)
1948 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
1949 }
1950
0679adc4 1951 if (fdp->fd_files != fdp->fd_builtin_files)
efda3bd0 1952 kfree(fdp->fd_files, M_FILEDESC);
690a3127 1953 if (fdp->fd_cdir) {
28623bf9 1954 cache_drop(&fdp->fd_ncdir);
984263bc 1955 vrele(fdp->fd_cdir);
690a3127 1956 }
bccde7a3 1957 if (fdp->fd_rdir) {
28623bf9 1958 cache_drop(&fdp->fd_nrdir);
bccde7a3
MD
1959 vrele(fdp->fd_rdir);
1960 }
690a3127 1961 if (fdp->fd_jdir) {
28623bf9 1962 cache_drop(&fdp->fd_njdir);
984263bc 1963 vrele(fdp->fd_jdir);
690a3127 1964 }
984263bc 1965 if (fdp->fd_knlist)
efda3bd0 1966 kfree(fdp->fd_knlist, M_KQUEUE);
984263bc 1967 if (fdp->fd_knhash)
efda3bd0
MD
1968 kfree(fdp->fd_knhash, M_KQUEUE);
1969 kfree(fdp, M_FILEDESC);
984263bc
MD
1970}
1971
1972/*
fa541be6 1973 * Retrieve and reference the file pointer associated with a descriptor.
85fceac1
MD
1974 *
1975 * MPSAFE
fa541be6
MD
1976 */
1977struct file *
1978holdfp(struct filedesc *fdp, int fd, int flag)
1979{
1980 struct file* fp;
1981
85fceac1
MD
1982 spin_lock_rd(&fdp->fd_spin);
1983 if (((u_int)fd) >= fdp->fd_nfiles) {
1984 fp = NULL;
1985 goto done;
1986 }
fa541be6 1987 if ((fp = fdp->fd_files[fd].fp) == NULL)
85fceac1
MD
1988 goto done;
1989 if ((fp->f_flag & flag) == 0 && flag != -1) {
1990 fp = NULL;
1991 goto done;
1992 }
fa541be6 1993 fhold(fp);
85fceac1
MD
1994done:
1995 spin_unlock_rd(&fdp->fd_spin);
fa541be6
MD
1996 return (fp);
1997}
1998
1999/*
2000 * holdsock() - load the struct file pointer associated
2001 * with a socket into *fpp. If an error occurs, non-zero
2002 * will be returned and *fpp will be set to NULL.
5b287bba
MD
2003 *
2004 * MPSAFE
fa541be6
MD
2005 */
2006int
5b287bba 2007holdsock(struct filedesc *fdp, int fd, struct file **fpp)
fa541be6
MD
2008{
2009 struct file *fp;
5b287bba 2010 int error;
fa541be6 2011
5b287bba
MD
2012 spin_lock_rd(&fdp->fd_spin);
2013 if ((unsigned)fd >= fdp->fd_nfiles) {
2014 error = EBADF;
2015 fp = NULL;
2016 goto done;
2017 }
2018 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2019 error = EBADF;
2020 goto done;
2021 }
2022 if (fp->f_type != DTYPE_SOCKET) {
2023 error = ENOTSOCK;
2024 goto done;
2025 }
fa541be6 2026 fhold(fp);
5b287bba
MD
2027 error = 0;
2028done:
2029 spin_unlock_rd(&fdp->fd_spin);
fa541be6
MD
2030 *fpp = fp;
2031 return (error);
2032}
2033
2034/*
5b287bba
MD
2035 * Convert a user file descriptor to a held file pointer.
2036 *
2037 * MPSAFE
228b401d
MD
2038 */
2039int
5b287bba 2040holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
228b401d
MD
2041{
2042 struct file *fp;
5b287bba
MD
2043 int error;
2044
2045 spin_lock_rd(&fdp->fd_spin);
2046 if ((unsigned)fd >= fdp->fd_nfiles) {
2047 error = EBADF;
2048 fp = NULL;
2049 goto done;
2050 }
2051 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2052 error = EBADF;
2053 goto done;
2054 }
2055 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
8f9f2535 2056 fp = NULL;
5b287bba
MD
2057 error = EINVAL;
2058 goto done;
2059 }
2060 fhold(fp);
2061 error = 0;
2062done:
2063 spin_unlock_rd(&fdp->fd_spin);
228b401d 2064 *fpp = fp;
5b287bba 2065 return (error);
228b401d
MD
2066}
2067
2068/*
984263bc
MD
2069 * For setugid programs, we don't want to people to use that setugidness
2070 * to generate error messages which write to a file which otherwise would
2071 * otherwise be off-limits to the process.
2072 *
2073 * This is a gross hack to plug the hole. A better solution would involve
2074 * a special vop or other form of generalized access control mechanism. We
2075 * go ahead and just reject all procfs file systems accesses as dangerous.
2076 *
2077 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2078 * sufficient. We also don't for check setugidness since we know we are.
2079 */
2080static int
2081is_unsafe(struct file *fp)
2082{
2083 if (fp->f_type == DTYPE_VNODE &&
2084 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2085 return (1);
2086 return (0);
2087}
2088
2089/*
2090 * Make this setguid thing safe, if at all possible.
5b287bba
MD
2091 *
2092 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
984263bc
MD
2093 */
2094void
dadab5e9 2095setugidsafety(struct proc *p)
984263bc
MD
2096{
2097 struct filedesc *fdp = p->p_fd;
41c20dac 2098 int i;
984263bc
MD
2099
2100 /* Certain daemons might not have file descriptors. */
2101 if (fdp == NULL)
2102 return;
2103
2104 /*
0679adc4 2105 * note: fdp->fd_files may be reallocated out from under us while
984263bc
MD
2106 * we are blocked in a close. Be careful!
2107 */
2108 for (i = 0; i <= fdp->fd_lastfile; i++) {
2109 if (i > 2)
2110 break;
0679adc4 2111 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
984263bc
MD
2112 struct file *fp;
2113
984263bc
MD
2114 if (i < fdp->fd_knlistsize)
2115 knote_fdclose(p, i);
2116 /*
2117 * NULL-out descriptor prior to close to avoid
2118 * a race while close blocks.
2119 */
5b287bba 2120 if ((fp = funsetfd_locked(fdp, i)) != NULL)
b8477cda 2121 closef(fp, p);
984263bc
MD
2122 }
2123 }
984263bc
MD
2124}
2125
2126/*
2127 * Close any files on exec?
5b287bba
MD
2128 *
2129 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
984263bc
MD
2130 */
2131void
dadab5e9 2132fdcloseexec(struct proc *p)
984263bc
MD
2133{
2134 struct filedesc *fdp = p->p_fd;
41c20dac 2135 int i;
984263bc
MD
2136
2137 /* Certain daemons might not have file descriptors. */
2138 if (fdp == NULL)
2139 return;
2140
2141 /*
0679adc4
MD
2142 * We cannot cache fd_files since operations may block and rip
2143 * them out from under us.
984263bc
MD
2144 */
2145 for (i = 0; i <= fdp->fd_lastfile; i++) {
0679adc4
MD
2146 if (fdp->fd_files[i].fp != NULL &&
2147 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
984263bc
MD
2148 struct file *fp;
2149
984263bc
MD
2150 if (i < fdp->fd_knlistsize)
2151 knote_fdclose(p, i);
2152 /*
2153 * NULL-out descriptor prior to close to avoid
2154 * a race while close blocks.
2155 */
5b287bba 2156 if ((fp = funsetfd_locked(fdp, i)) != NULL)
b8477cda 2157 closef(fp, p);
984263bc
MD
2158 }
2159 }
984263bc
MD
2160}
2161
2162/*
2163 * It is unsafe for set[ug]id processes to be started with file
2164 * descriptors 0..2 closed, as these descriptors are given implicit
2165 * significance in the Standard C library. fdcheckstd() will create a
2166 * descriptor referencing /dev/null for each of stdin, stdout, and
2167 * stderr that is not already open.
5b287bba
MD
2168 *
2169 * NOT MPSAFE - calls falloc, vn_open, etc
984263bc
MD
2170 */
2171int
dadab5e9 2172fdcheckstd(struct proc *p)
984263bc 2173{
fad57d0e 2174 struct nlookupdata nd;
dadab5e9
MD
2175 struct filedesc *fdp;
2176 struct file *fp;
973c11b9 2177 int retval;
259b8ea0 2178 int i, error, flags, devnull;
984263bc 2179
ea8f324c
MD
2180 fdp = p->p_fd;
2181 if (fdp == NULL)
2182 return (0);
2183 devnull = -1;
2184 error = 0;
2185 for (i = 0; i < 3; i++) {
0679adc4 2186 if (fdp->fd_files[i].fp != NULL)
fad57d0e
MD
2187 continue;
2188 if (devnull < 0) {
259b8ea0 2189 if ((error = falloc(p, &fp, &devnull)) != 0)
fad57d0e
MD
2190 break;
2191
2192 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2193 NLC_FOLLOW|NLC_LOCKVP);
2194 flags = FREAD | FWRITE;
2195 if (error == 0)
2196 error = vn_open(&nd, fp, flags, 0);
2197 if (error == 0)
259b8ea0
MD
2198 fsetfd(p, fp, devnull);
2199 else
2200 fsetfd(p, NULL, devnull);
9f87144f 2201 fdrop(fp);
fad57d0e
MD
2202 nlookup_done(&nd);
2203 if (error)
2204 break;
259b8ea0 2205 KKASSERT(i == devnull);
fad57d0e
MD
2206 } else {
2207 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2208 if (error != 0)
2209 break;
2210 }
ea8f324c
MD
2211 }
2212 return (error);
984263bc
MD
2213}
2214
2215/*
2216 * Internal form of close.
2217 * Decrement reference count on file structure.
dadab5e9 2218 * Note: td and/or p may be NULL when closing a file
984263bc 2219 * that was being passed in a message.
5b287bba
MD
2220 *
2221 * MPALMOSTSAFE - acquires mplock for VOP operations
984263bc
MD
2222 */
2223int
b8477cda 2224closef(struct file *fp, struct proc *p)
984263bc
MD
2225{
2226 struct vnode *vp;
2227 struct flock lf;
2228 struct filedesc_to_leader *fdtol;
2229
2230 if (fp == NULL)
2231 return (0);
b8477cda 2232
984263bc
MD
2233 /*
2234 * POSIX record locking dictates that any close releases ALL
2235 * locks owned by this process. This is handled by setting
2236 * a flag in the unlock to free ONLY locks obeying POSIX
2237 * semantics, and not to free BSD-style file locks.
2238 * If the descriptor was in a message, POSIX-style locks
2239 * aren't passed with the descriptor.
2240 */
7749886d
MD
2241 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2242 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2243 ) {
2244 get_mplock();
984263bc
MD
2245 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2246 lf.l_whence = SEEK_SET;
2247 lf.l_start = 0;
2248 lf.l_len = 0;
2249 lf.l_type = F_UNLCK;
2250 vp = (struct vnode *)fp->f_data;
2251 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2252 &lf, F_POSIX);
2253 }
2254 fdtol = p->p_fdtol;
2255 if (fdtol != NULL) {
2256 /*
2257 * Handle special case where file descriptor table
2258 * is shared between multiple process leaders.
2259 */
2260 for (fdtol = fdtol->fdl_next;
2261 fdtol != p->p_fdtol;
2262 fdtol = fdtol->fdl_next) {
2263 if ((fdtol->fdl_leader->p_flag &
2264 P_ADVLOCK) == 0)
2265 continue;
2266 fdtol->fdl_holdcount++;
2267 lf.l_whence = SEEK_SET;
2268 lf.l_start = 0;
2269 lf.l_len = 0;
2270 lf.l_type = F_UNLCK;
2271 vp = (struct vnode *)fp->f_data;
2272 (void) VOP_ADVLOCK(vp,
661fc1a0 2273 (caddr_t)fdtol->fdl_leader,
984263bc
MD
2274 F_UNLCK, &lf, F_POSIX);
2275 fdtol->fdl_holdcount--;
2276 if (fdtol->fdl_holdcount == 0 &&
2277 fdtol->fdl_wakeup != 0) {
2278 fdtol->fdl_wakeup = 0;
2279 wakeup(fdtol);
2280 }
2281 }
2282 }
7749886d 2283 rel_mplock();
984263bc 2284 }
9f87144f 2285 return (fdrop(fp));
984263bc
MD
2286}
2287
85fceac1
MD
2288/*
2289 * MPSAFE
2290 *
2291 * fhold() can only be called if f_count is already at least 1 (i.e. the
2292 * caller of fhold() already has a reference to the file pointer in some
5b287bba
MD
2293 * manner or other).
2294 *
bc1eb676
MD
2295 * f_count is not spin-locked. Instead, atomic ops are used for
2296 * incrementing, decrementing, and handling the 1->0 transition.
85fceac1
MD
2297 */
2298void
2299fhold(struct file *fp)
2300{
2301 atomic_add_int(&fp->f_count, 1);
2302}
2303
2304/*
bc1eb676 2305 * fdrop() - drop a reference to a descriptor
85fceac1 2306 *
5b287bba 2307 * MPALMOSTSAFE - acquires mplock for final close sequence
85fceac1 2308 */
984263bc 2309int
9f87144f 2310fdrop(struct file *fp)
984263bc
MD
2311{
2312 struct flock lf;
2313 struct vnode *vp;
2314 int error;
2315
bc1eb676
MD
2316 /*
2317 * A combined fetch and subtract is needed to properly detect
2318 * 1->0 transitions, otherwise two cpus dropping from a ref
2319 * count of 2 might both try to run the 1->0 code.
2320 */
2321 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
984263bc 2322 return (0);
85fceac1 2323
5b287bba
MD
2324 get_mplock();
2325
85fceac1
MD
2326 /*
2327 * The last reference has gone away, we own the fp structure free
2328 * and clear.
2329 */
984263bc
MD
2330 if (fp->f_count < 0)
2331 panic("fdrop: count < 0");
7749886d
MD
2332 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2333 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2334 ) {
984263bc
MD
2335 lf.l_whence = SEEK_SET;
2336 lf.l_start = 0;
2337 lf.l_len = 0;
2338 lf.l_type = F_UNLCK;
2339 vp = (struct vnode *)fp->f_data;
71c18fe3 2340 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
984263bc
MD
2341 }
2342 if (fp->f_ops != &badfileops)
87de5057 2343 error = fo_close(fp);
984263bc
MD
2344 else
2345 error = 0;
2346 ffree(fp);
5b287bba 2347 rel_mplock();
984263bc
MD
2348 return (error);
2349}
2350
2351/*
2352 * Apply an advisory lock on a file descriptor.
2353 *
2354 * Just attempt to get a record lock of the requested type on
2355 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3919ced0
MD
2356 *
2357 * MPALMOSTSAFE
984263bc 2358 */
984263bc 2359int
753fd850 2360sys_flock(struct flock_args *uap)
984263bc 2361{
41c20dac 2362 struct proc *p = curproc;
41c20dac 2363 struct file *fp;
984263bc
MD
2364 struct vnode *vp;
2365 struct flock lf;
5b287bba 2366 int error;
984263bc 2367
5b287bba 2368 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
984263bc 2369 return (EBADF);
3919ced0 2370 get_mplock();
5b287bba
MD
2371 if (fp->f_type != DTYPE_VNODE) {
2372 error = EOPNOTSUPP;
2373 goto done;
2374 }
984263bc
MD
2375 vp = (struct vnode *)fp->f_data;
2376 lf.l_whence = SEEK_SET;
2377 lf.l_start = 0;
2378 lf.l_len = 0;
2379 if (uap->how & LOCK_UN) {
2380 lf.l_type = F_UNLCK;
2381 fp->f_flag &= ~FHASLOCK;
5b287bba
MD
2382 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2383 goto done;
984263bc
MD
2384 }
2385 if (uap->how & LOCK_EX)
2386 lf.l_type = F_WRLCK;
2387 else if (uap->how & LOCK_SH)
2388 lf.l_type = F_RDLCK;
5b287bba
MD
2389 else {
2390 error = EBADF;
2391 goto done;
2392 }
984263bc
MD
2393 fp->f_flag |= FHASLOCK;
2394 if (uap->how & LOCK_NB)
5b287bba
MD
2395 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2396 else
2397 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2398done:
3919ced0 2399 rel_mplock();
5b287bba
MD
2400 fdrop(fp);
2401 return (error);
984263bc
MD
2402}
2403
2404/*
2405 * File Descriptor pseudo-device driver (/dev/fd/).
2406 *
2407 * Opening minor device N dup()s the file (if any) connected to file
2408 * descriptor N belonging to the calling process. Note that this driver
2409 * consists of only the ``open()'' routine, because all subsequent
2410 * references to this file will be direct to the other driver.
2411 */
984263bc 2412static int
fef8985e 2413fdopen(struct dev_open_args *ap)
984263bc 2414{
fef8985e
MD
2415 thread_t td = curthread;
2416
fbfe4e7d 2417 KKASSERT(td->td_lwp != NULL);
984263bc
MD
2418
2419 /*
fbfe4e7d 2420 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
984263bc
MD
2421 * the file descriptor being sought for duplication. The error
2422 * return ensures that the vnode for this device will be released
2423 * by vn_open. Open will detect this special error and take the
2424 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2425 * will simply report the error.
2426 */
fef8985e 2427 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
984263bc
MD
2428 return (ENODEV);
2429}
2430
2431/*
259b8ea0
MD
2432 * The caller has reserved the file descriptor dfd for us. On success we
2433 * must fsetfd() it. On failure the caller will clean it up.
5b287bba
MD
2434 *
2435 * NOT MPSAFE - isn't getting spinlocks, possibly other things
984263bc
MD
2436 */
2437int
259b8ea0 2438dupfdopen(struct proc *p, int dfd, int sfd, int mode, int error)
984263bc 2439{
259b8ea0 2440 struct filedesc *fdp = p->p_fd;
41c20dac 2441 struct file *wfp;
259b8ea0 2442 struct file *xfp;
ea8f324c 2443 int werror;
984263bc 2444
259b8ea0 2445 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
984263bc 2446 return (EBADF);
984263bc
MD
2447
2448 /*
ea8f324c
MD
2449 * Close a revoke/dup race. Duping a descriptor marked as revoked
2450 * will dup a dummy descriptor instead of the real one.
2451 */
2452 if (wfp->f_flag & FREVOKED) {
2453 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2454 fdrop(wfp);
2455 wfp = NULL;
2456 werror = falloc(NULL, &wfp, NULL);
2457 if (werror)
2458 return (werror);
2459 }
2460
2461 /*
984263bc
MD
2462 * There are two cases of interest here.
2463 *
259b8ea0 2464 * For ENODEV simply dup sfd to file descriptor dfd and return.
984263bc 2465 *
259b8ea0
MD
2466 * For ENXIO steal away the file structure from sfd and store it
2467 * dfd. sfd is effectively closed by this operation.
984263bc
MD
2468 *
2469 * Any other error code is just returned.
2470 */
2471 switch (error) {
2472 case ENODEV:
2473 /*
2474 * Check that the mode the file is being opened for is a
2475 * subset of the mode of the existing descriptor.
2476 */
ea8f324c
MD
2477 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2478 error = EACCES;
2479 break;
2480 }
259b8ea0
MD
2481 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2482 fsetfd(p, wfp, dfd);
2483 error = 0;
2484 break;
984263bc
MD
2485 case ENXIO:
2486 /*
2487 * Steal away the file pointer from dfd, and stuff it into indx.
2488 */
259b8ea0
MD
2489 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2490 fsetfd(p, wfp, dfd);
5b287bba 2491 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL)
259b8ea0 2492 fdrop(xfp);
259b8ea0
MD
2493 error = 0;
2494 break;
984263bc 2495 default:
259b8ea0 2496 break;
984263bc 2497 }
259b8ea0
MD
2498 fdrop(wfp);
2499 return (error);
984263bc
MD
2500}
2501
5b287bba
MD
2502/*
2503 * NOT MPSAFE - I think these refer to a common file descriptor table
2504 * and we need to spinlock that to link fdtol in.
2505 */
984263bc
MD
2506struct filedesc_to_leader *
2507filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2508 struct proc *leader)
2509{
2510 struct filedesc_to_leader *fdtol;
2511
77652cad 2512 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
7bf8660a 2513 M_FILEDESC_TO_LEADER, M_WAITOK);
984263bc
MD
2514 fdtol->fdl_refcount = 1;
2515 fdtol->fdl_holdcount = 0;
2516 fdtol->fdl_wakeup = 0;
2517 fdtol->fdl_leader = leader;
2518 if (old != NULL) {
2519 fdtol->fdl_next = old->fdl_next;
2520 fdtol->fdl_prev = old;
2521 old->fdl_next = fdtol;
2522 fdtol->fdl_next->fdl_prev = fdtol;
2523 } else {
2524 fdtol->fdl_next = fdtol;
2525 fdtol->fdl_prev = fdtol;
2526 }
2527 return fdtol;
2528}
2529
2530/*
2dd63755 2531 * Scan all file pointers in the system. The callback is made with
ea8f324c 2532 * the master list spinlock held exclusively.
2dd63755
MD
2533 *
2534 * MPSAFE
2dd63755
MD
2535 */
2536void
2537allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2538{
2539 struct file *fp;
2540 int res;
2541
2542 spin_lock_wr(&filehead_spin);
2543 LIST_FOREACH(fp, &filehead, f_list) {
2dd63755 2544 res = callback(fp, data);
2dd63755
MD
2545 if (res < 0)
2546 break;
2547 }
2548 spin_unlock_wr(&filehead_spin);
2549}
2550
2551/*
984263bc 2552 * Get file structures.
5b287bba
MD
2553 *
2554 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
984263bc 2555 */
8fa76237
MD
2556
2557struct sysctl_kern_file_info {
2558 int count;
2559 int error;
2560 struct sysctl_req *req;
2561};
2562
2563static int sysctl_kern_file_callback(struct proc *p, void *data);
2564
984263bc
MD
2565static int
2566sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2567{
8fa76237 2568 struct sysctl_kern_file_info info;
984263bc
MD
2569
2570 /*
7b124c9f
JS
2571 * Note: because the number of file descriptors is calculated
2572 * in different ways for sizing vs returning the data,
2573 * there is information leakage from the first loop. However,
2574 * it is of a similar order of magnitude to the leakage from
2575 * global system statistics such as kern.openfiles.
6d132b4d
MD
2576 *
2577 * When just doing a count, note that we cannot just count
2578 * the elements and add f_count via the filehead list because
2579 * threaded processes share their descriptor table and f_count might
2580 * still be '1' in that case.
e38910da
MD
2581 *
2582 * Since the SYSCTL op can block, we must hold the process to
2583 * prevent it being ripped out from under us either in the
2584 * file descriptor loop or in the greater LIST_FOREACH. The
2585 * process may be in varying states of disrepair. If the process
2586 * is in SZOMB we may have caught it just as it is being removed
2587 * from the allproc list, we must skip it in that case to maintain
2588 * an unbroken chain through the allproc list.
984263bc 2589 */
8fa76237
MD
2590 info.count = 0;
2591 info.error = 0;
2592 info.req = req;
2593 allproc_scan(sysctl_kern_file_callback, &info);
6d132b4d
MD
2594
2595 /*
2596 * When just calculating the size, overestimate a bit to try to
2597 * prevent system activity from causing the buffer-fill call
2598 * to fail later on.
2599 */
2600 if (req->oldptr == NULL) {
8fa76237
MD
2601 info.count = (info.count + 16) + (info.count / 10);
2602 info.error = SYSCTL_OUT(req, NULL,
2603 info.count * sizeof(struct kinfo_file));
6d132b4d 2604 }
8fa76237
MD
2605 return (info.error);
2606}
2607
2608static int
2609sysctl_kern_file_callback(struct proc *p, void *data)
2610{
2611 struct sysctl_kern_file_info *info = data;
2612 struct kinfo_file kf;
2613 struct filedesc *fdp;
2614 struct file *fp;
2615 uid_t uid;
2616 int n;
2617
416d05d7 2618 if (p->p_stat == SIDL || p->p_stat == SZOMB)
8fa76237 2619 return(0);
9910d07b 2620 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
8fa76237 2621 return(0);
0a4a9c77
MD
2622
2623 /*
2624 * Softref the fdp to prevent it from being destroyed
2625 */
2626 spin_lock_wr(&p->p_spin);
2627 if ((fdp = p->p_fd) == NULL) {
2628 spin_unlock_wr(&p->p_spin);
8fa76237 2629 return(0);
0a4a9c77
MD
2630 }
2631 atomic_add_int(&fdp->fd_softrefs, 1);
2632 spin_unlock_wr(&p->p_spin);
2633
2634 /*
2635 * The fdp's own spinlock prevents the contents from being
2636 * modified.
2637 */
8fa76237
MD
2638 spin_lock_rd(&fdp->fd_spin);
2639 for (n = 0; n < fdp->fd_nfiles; ++n) {
2640 if ((fp = fdp->fd_files[n].fp) == NULL)
2641 continue;
2642 if (info->req->oldptr == NULL) {
2643 ++info->count;
2644 } else {
2645 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2646 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2647 spin_unlock_rd(&fdp->fd_spin);
2648 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2649 spin_lock_rd(&fdp->fd_spin);
2650 if (info->error)
2651 break;
2652 }
2653 }
2654 spin_unlock_rd(&fdp->fd_spin);
0a4a9c77 2655 atomic_subtract_int(&fdp->fd_softrefs, 1);
8fa76237
MD
2656 if (info->error)
2657 return(-1);
2658 return(0);
984263bc
MD
2659}
2660
2661SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2662 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2663
2664SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2665 &maxfilesperproc, 0, "Maximum files allowed open per process");
2666
2667SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2668 &maxfiles, 0, "Maximum number of files");
2669
60ee93b9
MD
2670SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2671 &maxfilesrootres, 0, "Descriptors reserved for root use");
2672
984263bc
MD
2673SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2674 &nfiles, 0, "System-wide number of open files");
2675
2676static void
2677fildesc_drvinit(void *unused)
2678{
2679 int fd;
2680
e4c9c0c8 2681 for (fd = 0; fd < NUMFDESC; fd++) {
fef8985e 2682 make_dev(&fildesc_ops, fd,
3e82b46c 2683 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
e4c9c0c8 2684 }
cd29885a 2685
fef8985e
MD
2686 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2687 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2688 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
984263bc
MD
2689}
2690
d9b2033e
MD
2691/*
2692 * MPSAFE
2693 */
984263bc 2694struct fileops badfileops = {
b2d248cb
MD
2695 .fo_read = badfo_readwrite,
2696 .fo_write = badfo_readwrite,
2697 .fo_ioctl = badfo_ioctl,
2698 .fo_poll = badfo_poll,
2699 .fo_kqfilter = badfo_kqfilter,
2700 .fo_stat = badfo_stat,
2701 .fo_close = badfo_close,
2702 .fo_shutdown = badfo_shutdown
984263bc
MD
2703};
2704
d9b2033e
MD
2705/*
2706 * MPSAFE
2707 */
a724d72b 2708int
dadab5e9
MD
2709badfo_readwrite(
2710 struct file *fp,
2711 struct uio *uio,
2712 struct ucred *cred,
87de5057 2713 int flags
dadab5e9 2714) {
984263bc
MD
2715 return (EBADF);
2716}
2717
d9b2033e
MD
2718/*
2719 * MPSAFE
2720 */
a724d72b 2721int
87baaf0c
MD
2722badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2723 struct ucred *cred, struct sysmsg *msgv)
984263bc 2724{
984263bc
MD
2725 return (EBADF);
2726}
2727
d9b2033e
MD
2728/*
2729 * MPSAFE
2730 */
a724d72b 2731int
87de5057 2732badfo_poll(struct file *fp, int events, struct ucred *cred)
984263bc 2733{
984263bc
MD
2734 return (0);
2735}
2736
d9b2033e
MD
2737/*
2738 * MPSAFE
2739 */
a724d72b 2740int
dadab5e9 2741badfo_kqfilter(struct file *fp, struct knote *kn)
984263bc 2742{
984263bc
MD
2743 return (0);
2744}
2745
a724d72b 2746int
87de5057 2747badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
984263bc 2748{
984263bc
MD
2749 return (EBADF);
2750}
2751
d9b2033e
MD
2752/*
2753 * MPSAFE
2754 */
a724d72b 2755int
87de5057 2756badfo_close(struct file *fp)
984263bc 2757{
984263bc
MD
2758 return (EBADF);
2759}
2760
d9b2033e
MD
2761/*
2762 * MPSAFE
2763 */
a724d72b 2764int
87de5057 2765badfo_shutdown(struct file *fp, int how)
004d2de5
MD
2766{
2767 return (EBADF);
2768}
2769
d9b2033e
MD
2770/*
2771 * MPSAFE
2772 */
004d2de5 2773int
87de5057 2774nofo_shutdown(struct file *fp, int how)
004d2de5
MD
2775{
2776 return (EOPNOTSUPP);
2777}
2778
984263bc
MD
2779SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2780 fildesc_drvinit,NULL)