kernel - Provide descriptions for lwkt.* and debug.* sysctl's
[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 $
984263bc
MD
73 */
74
75#include "opt_compat.h"
76#include <sys/param.h>
77#include <sys/systm.h>
78#include <sys/malloc.h>
79#include <sys/sysproto.h>
80#include <sys/conf.h>
fef8985e 81#include <sys/device.h>
a724d72b 82#include <sys/file.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>
684a93c4 107#include <sys/mplock2.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);
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
1e864525
MD
120static struct krate krate_uidinfo = { .freq = 1 };
121
984263bc
MD
122static d_open_t fdopen;
123#define NUMFDESC 64
124
125#define CDEV_MAJOR 22
fef8985e 126static struct dev_ops fildesc_ops = {
88abd8b5 127 { "FD", 0, 0 },
fef8985e 128 .d_open = fdopen,
984263bc
MD
129};
130
984263bc
MD
131/*
132 * Descriptor management.
133 */
2dd63755
MD
134static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
135static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
136static int nfiles; /* actual number of open files */
984263bc
MD
137extern int cmask;
138
139/*
8247b2f9 140 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
5b287bba
MD
141 *
142 * MPSAFE - must be called with fdp->fd_spin exclusively held
8247b2f9
MD
143 */
144static __inline
145void
5b287bba 146fdfixup_locked(struct filedesc *fdp, int fd)
8247b2f9
MD
147{
148 if (fd < fdp->fd_freefile) {
149 fdp->fd_freefile = fd;
150 }
151 while (fdp->fd_lastfile >= 0 &&
152 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
153 fdp->fd_files[fdp->fd_lastfile].reserved == 0
154 ) {
155 --fdp->fd_lastfile;
156 }
157}
158
159/*
984263bc 160 * System calls on descriptors.
5b287bba
MD
161 *
162 * MPSAFE
984263bc 163 */
984263bc 164int
753fd850 165sys_getdtablesize(struct getdtablesize_args *uap)
984263bc 166{
41c20dac 167 struct proc *p = curproc;
5b287bba 168 struct plimit *limit = p->p_limit;
d37c8f7f 169 int dtsize;
984263bc 170
287a8577 171 spin_lock(&limit->p_spin);
d37c8f7f
MD
172 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
173 dtsize = INT_MAX;
174 else
175 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
287a8577 176 spin_unlock(&limit->p_spin);
a7b16922 177
d37c8f7f
MD
178 if (dtsize > maxfilesperproc)
179 dtsize = maxfilesperproc;
180 if (dtsize < minfilesperproc)
181 dtsize = minfilesperproc;
182 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
183 dtsize = maxfilesperuser;
184 uap->sysmsg_result = dtsize;
984263bc
MD
185 return (0);
186}
187
188/*
189 * Duplicate a file descriptor to a particular value.
190 *
191 * note: keep in mind that a potential race condition exists when closing
192 * descriptors from a shared descriptor table (via rfork).
5b287bba
MD
193 *
194 * MPSAFE
984263bc 195 */
984263bc 196int
753fd850 197sys_dup2(struct dup2_args *uap)
984263bc 198{
dda4b42b 199 int error;
973c11b9 200 int fd = 0;
dda4b42b 201
973c11b9
MD
202 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
203 uap->sysmsg_fds[0] = fd;
dda4b42b
DRJ
204
205 return (error);
984263bc
MD
206}
207
208/*
209 * Duplicate a file descriptor.
5b287bba
MD
210 *
211 * MPSAFE
984263bc 212 */
984263bc 213int
753fd850 214sys_dup(struct dup_args *uap)
984263bc 215{
dda4b42b 216 int error;
973c11b9 217 int fd = 0;
984263bc 218
973c11b9
MD
219 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
220 uap->sysmsg_fds[0] = fd;
dda4b42b
DRJ
221
222 return (error);
984263bc
MD
223}
224
5b287bba
MD
225/*
226 * MPALMOSTSAFE - acquires mplock for fp operations
227 */
984263bc 228int
87de5057 229kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
984263bc 230{
dadab5e9
MD
231 struct thread *td = curthread;
232 struct proc *p = td->td_proc;
41c20dac 233 struct file *fp;
984263bc 234 struct vnode *vp;
984263bc 235 u_int newmin;
9ba76b73 236 u_int oflags;
3a907475 237 u_int nflags;
dda4b42b 238 int tmp, error, flg = F_POSIX;
984263bc 239
dadab5e9
MD
240 KKASSERT(p);
241
5b287bba
MD
242 /*
243 * Operations on file descriptors that do not require a file pointer.
244 */
dda4b42b 245 switch (cmd) {
984263bc 246 case F_GETFD:
85fceac1
MD
247 error = fgetfdflags(p->p_fd, fd, &tmp);
248 if (error == 0)
249 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
5b287bba 250 return (error);
984263bc
MD
251
252 case F_SETFD:
85fceac1
MD
253 if (dat->fc_cloexec & FD_CLOEXEC)
254 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
255 else
256 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
5b287bba
MD
257 return (error);
258 case F_DUPFD:
259 newmin = dat->fc_fd;
260 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
261 return (error);
262 default:
85fceac1 263 break;
5b287bba 264 }
984263bc 265
5b287bba
MD
266 /*
267 * Operations on file pointers
268 */
269 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
270 return (EBADF);
271
272 get_mplock();
273 switch (cmd) {
984263bc 274 case F_GETFL:
dda4b42b 275 dat->fc_flags = OFLAGS(fp->f_flag);
85fceac1
MD
276 error = 0;
277 break;
984263bc
MD
278
279 case F_SETFL:
3a907475
MD
280 oflags = fp->f_flag;
281 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
282 nflags |= oflags & ~FCNTLFLAGS;
283
f58b505b 284 error = 0;
3a907475
MD
285 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
286 error = EINVAL;
287 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
e8cdd284 288 tmp = nflags & FASYNC;
87baaf0c
MD
289 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
290 cred, NULL);
f58b505b 291 }
3a907475
MD
292 if (error == 0)
293 fp->f_flag = nflags;
85fceac1 294 break;
984263bc
MD
295
296 case F_GETOWN:
87baaf0c
MD
297 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
298 cred, NULL);
85fceac1 299 break;
984263bc
MD
300
301 case F_SETOWN:
87baaf0c
MD
302 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
303 cred, NULL);
85fceac1 304 break;
984263bc
MD
305
306 case F_SETLKW:
307 flg |= F_WAIT;
308 /* Fall into F_SETLK */
309
310 case F_SETLK:
85fceac1
MD
311 if (fp->f_type != DTYPE_VNODE) {
312 error = EBADF;
313 break;
314 }
984263bc
MD
315 vp = (struct vnode *)fp->f_data;
316
317 /*
318 * copyin/lockop may block
319 */
dda4b42b
DRJ
320 if (dat->fc_flock.l_whence == SEEK_CUR)
321 dat->fc_flock.l_start += fp->f_offset;
984263bc 322
dda4b42b 323 switch (dat->fc_flock.l_type) {
984263bc
MD
324 case F_RDLCK:
325 if ((fp->f_flag & FREAD) == 0) {
326 error = EBADF;
327 break;
328 }
329 p->p_leader->p_flag |= P_ADVLOCK;
330 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
dda4b42b 331 &dat->fc_flock, flg);
984263bc
MD
332 break;
333 case F_WRLCK:
334 if ((fp->f_flag & FWRITE) == 0) {
335 error = EBADF;
336 break;
337 }
338 p->p_leader->p_flag |= P_ADVLOCK;
339 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
dda4b42b 340 &dat->fc_flock, flg);
984263bc
MD
341 break;
342 case F_UNLCK:
343 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
dda4b42b 344 &dat->fc_flock, F_POSIX);
984263bc
MD
345 break;
346 default:
347 error = EINVAL;
348 break;
349 }
85fceac1
MD
350
351 /*
352 * It is possible to race a close() on the descriptor while
353 * we were blocked getting the lock. If this occurs the
354 * close might not have caught the lock.
355 */
cf9f4e88 356 if (checkfdclosed(p->p_fd, fd, fp)) {
dda4b42b
DRJ
357 dat->fc_flock.l_whence = SEEK_SET;
358 dat->fc_flock.l_start = 0;
359 dat->fc_flock.l_len = 0;
360 dat->fc_flock.l_type = F_UNLCK;
984263bc 361 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
dda4b42b 362 F_UNLCK, &dat->fc_flock, F_POSIX);
984263bc 363 }
85fceac1 364 break;
984263bc
MD
365
366 case F_GETLK:
85fceac1
MD
367 if (fp->f_type != DTYPE_VNODE) {
368 error = EBADF;
369 break;
370 }
984263bc
MD
371 vp = (struct vnode *)fp->f_data;
372 /*
373 * copyin/lockop may block
374 */
dda4b42b
DRJ
375 if (dat->fc_flock.l_type != F_RDLCK &&
376 dat->fc_flock.l_type != F_WRLCK &&
377 dat->fc_flock.l_type != F_UNLCK) {
85fceac1
MD
378 error = EINVAL;
379 break;
984263bc 380 }
dda4b42b
DRJ
381 if (dat->fc_flock.l_whence == SEEK_CUR)
382 dat->fc_flock.l_start += fp->f_offset;
984263bc 383 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
dda4b42b 384 &dat->fc_flock, F_POSIX);
85fceac1 385 break;
984263bc 386 default:
85fceac1
MD
387 error = EINVAL;
388 break;
984263bc 389 }
5b287bba
MD
390 rel_mplock();
391
85fceac1
MD
392 fdrop(fp);
393 return (error);
984263bc
MD
394}
395
396/*
dda4b42b 397 * The file control system call.
5b287bba
MD
398 *
399 * MPSAFE
dda4b42b
DRJ
400 */
401int
753fd850 402sys_fcntl(struct fcntl_args *uap)
dda4b42b
DRJ
403{
404 union fcntl_dat dat;
405 int error;
406
407 switch (uap->cmd) {
408 case F_DUPFD:
409 dat.fc_fd = uap->arg;
410 break;
411 case F_SETFD:
412 dat.fc_cloexec = uap->arg;
413 break;
414 case F_SETFL:
415 dat.fc_flags = uap->arg;
416 break;
417 case F_SETOWN:
418 dat.fc_owner = uap->arg;
419 break;
420 case F_SETLKW:
421 case F_SETLK:
422 case F_GETLK:
423 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
5b287bba 424 sizeof(struct flock));
dda4b42b
DRJ
425 if (error)
426 return (error);
427 break;
428 }
429
9910d07b 430 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
dda4b42b
DRJ
431
432 if (error == 0) {
433 switch (uap->cmd) {
434 case F_DUPFD:
435 uap->sysmsg_result = dat.fc_fd;
436 break;
437 case F_GETFD:
438 uap->sysmsg_result = dat.fc_cloexec;
439 break;
440 case F_GETFL:
441 uap->sysmsg_result = dat.fc_flags;
442 break;
443 case F_GETOWN:
444 uap->sysmsg_result = dat.fc_owner;
445 case F_GETLK:
446 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
447 sizeof(struct flock));
448 break;
449 }
450 }
451
452 return (error);
453}
454
455/*
984263bc 456 * Common code for dup, dup2, and fcntl(F_DUPFD).
dda4b42b
DRJ
457 *
458 * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells
459 * kern_dup() to destructively dup over an existing file descriptor if new
460 * is already open. DUP_VARIABLE tells kern_dup() to find the lowest
461 * unused file descriptor that is greater than or equal to new.
5b287bba
MD
462 *
463 * MPSAFE
984263bc 464 */
dda4b42b
DRJ
465int
466kern_dup(enum dup_type type, int old, int new, int *res)
984263bc 467{
dda4b42b
DRJ
468 struct thread *td = curthread;
469 struct proc *p = td->td_proc;
470 struct filedesc *fdp = p->p_fd;
984263bc
MD
471 struct file *fp;
472 struct file *delfp;
259b8ea0 473 int oldflags;
984263bc 474 int holdleaders;
d37c8f7f 475 int dtsize;
dda4b42b
DRJ
476 int error, newfd;
477
478 /*
479 * Verify that we have a valid descriptor to dup from and
480 * possibly to dup to.
d37c8f7f
MD
481 *
482 * NOTE: maxfilesperuser is not applicable to dup()
dda4b42b 483 */
259b8ea0 484retry:
d37c8f7f
MD
485 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
486 dtsize = INT_MAX;
487 else
488 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
489 if (dtsize > maxfilesperproc)
490 dtsize = maxfilesperproc;
491 if (dtsize < minfilesperproc)
492 dtsize = minfilesperproc;
493
494 if (new < 0 || new > dtsize)
5b287bba 495 return (EINVAL);
d37c8f7f 496
287a8577 497 spin_lock(&fdp->fd_spin);
5b287bba 498 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
287a8577 499 spin_unlock(&fdp->fd_spin);
dda4b42b 500 return (EBADF);
5b287bba 501 }
dda4b42b
DRJ
502 if (type == DUP_FIXED && old == new) {
503 *res = new;
287a8577 504 spin_unlock(&fdp->fd_spin);
dda4b42b
DRJ
505 return (0);
506 }
0679adc4 507 fp = fdp->fd_files[old].fp;
259b8ea0 508 oldflags = fdp->fd_files[old].fileflags;
5b287bba 509 fhold(fp); /* MPSAFE - can be called with a spinlock held */
dda4b42b
DRJ
510
511 /*
259b8ea0
MD
512 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
513 * if the requested descriptor is beyond the current table size.
514 *
515 * This can block. Retry if the source descriptor no longer matches
516 * or if our expectation in the expansion case races.
517 *
518 * If we are not expanding or allocating a new decriptor, then reset
519 * the target descriptor to a reserved state so we have a uniform
520 * setup for the next code block.
dda4b42b
DRJ
521 */
522 if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) {
287a8577 523 spin_unlock(&fdp->fd_spin);
dda4b42b 524 error = fdalloc(p, new, &newfd);
287a8577 525 spin_lock(&fdp->fd_spin);
dda4b42b 526 if (error) {
287a8577 527 spin_unlock(&fdp->fd_spin);
9f87144f 528 fdrop(fp);
dda4b42b
DRJ
529 return (error);
530 }
5b287bba
MD
531 /*
532 * Check for ripout
533 */
259b8ea0 534 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
5b287bba 535 fsetfd_locked(fdp, NULL, newfd);
287a8577 536 spin_unlock(&fdp->fd_spin);
259b8ea0
MD
537 fdrop(fp);
538 goto retry;
539 }
5b287bba
MD
540 /*
541 * Check for expansion race
542 */
259b8ea0 543 if (type != DUP_VARIABLE && new != newfd) {
5b287bba 544 fsetfd_locked(fdp, NULL, newfd);
287a8577 545 spin_unlock(&fdp->fd_spin);
259b8ea0
MD
546 fdrop(fp);
547 goto retry;
548 }
5b287bba
MD
549 /*
550 * Check for ripout, newfd reused old (this case probably
551 * can't occur).
552 */
259b8ea0 553 if (old == newfd) {
5b287bba 554 fsetfd_locked(fdp, NULL, newfd);
287a8577 555 spin_unlock(&fdp->fd_spin);
259b8ea0
MD
556 fdrop(fp);
557 goto retry;
558 }
dda4b42b 559 new = newfd;
259b8ea0
MD
560 delfp = NULL;
561 } else {
562 if (fdp->fd_files[new].reserved) {
287a8577 563 spin_unlock(&fdp->fd_spin);
259b8ea0 564 fdrop(fp);
6ea70f76 565 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
259b8ea0
MD
566 tsleep(fdp, 0, "fdres", hz);
567 goto retry;
568 }
dda4b42b 569
259b8ea0
MD
570 /*
571 * If the target descriptor was never allocated we have
572 * to allocate it. If it was we have to clean out the
5b287bba
MD
573 * old descriptor. delfp inherits the ref from the
574 * descriptor table.
259b8ea0
MD
575 */
576 delfp = fdp->fd_files[new].fp;
577 fdp->fd_files[new].fp = NULL;
578 fdp->fd_files[new].reserved = 1;
579 if (delfp == NULL) {
5b287bba 580 fdreserve_locked(fdp, new, 1);
259b8ea0
MD
581 if (new > fdp->fd_lastfile)
582 fdp->fd_lastfile = new;
dda4b42b 583 }
259b8ea0 584
dda4b42b 585 }
984263bc
MD
586
587 /*
5b287bba
MD
588 * NOTE: still holding an exclusive spinlock
589 */
590
591 /*
259b8ea0
MD
592 * If a descriptor is being overwritten we may hve to tell
593 * fdfree() to sleep to ensure that all relevant process
594 * leaders can be traversed in closef().
984263bc 595 */
984263bc 596 if (delfp != NULL && p->p_fdtol != NULL) {
984263bc
MD
597 fdp->fd_holdleaderscount++;
598 holdleaders = 1;
259b8ea0 599 } else {
984263bc 600 holdleaders = 0;
259b8ea0 601 }
dda4b42b 602 KASSERT(delfp == NULL || type == DUP_FIXED,
259b8ea0 603 ("dup() picked an open file"));
984263bc
MD
604
605 /*
259b8ea0
MD
606 * Duplicate the source descriptor, update lastfile. If the new
607 * descriptor was not allocated and we aren't replacing an existing
608 * descriptor we have to mark the descriptor as being in use.
609 *
610 * The fd_files[] array inherits fp's hold reference.
984263bc 611 */
5b287bba 612 fsetfd_locked(fdp, fp, new);
259b8ea0 613 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
287a8577 614 spin_unlock(&fdp->fd_spin);
5b287bba 615 fdrop(fp);
dda4b42b 616 *res = new;
984263bc
MD
617
618 /*
619 * If we dup'd over a valid file, we now own the reference to it
620 * and must dispose of it using closef() semantics (as if a
621 * close() were performed on it).
622 */
623 if (delfp) {
5a4b22b1 624 if (SLIST_FIRST(&delfp->f_klist))
34e191bd 625 knote_fdclose(delfp, fdp, new);
b8477cda 626 closef(delfp, p);
984263bc 627 if (holdleaders) {
287a8577 628 spin_lock(&fdp->fd_spin);
984263bc
MD
629 fdp->fd_holdleaderscount--;
630 if (fdp->fd_holdleaderscount == 0 &&
631 fdp->fd_holdleaderswakeup != 0) {
632 fdp->fd_holdleaderswakeup = 0;
287a8577 633 spin_unlock(&fdp->fd_spin);
984263bc 634 wakeup(&fdp->fd_holdleaderscount);
5b287bba 635 } else {
287a8577 636 spin_unlock(&fdp->fd_spin);
984263bc
MD
637 }
638 }
639 }
640 return (0);
641}
642
643/*
644 * If sigio is on the list associated with a process or process group,
645 * disable signalling from the device, remove sigio from the list and
646 * free sigio.
647 */
648void
7bf8660a 649funsetown(struct sigio *sigio)
984263bc 650{
984263bc
MD
651 if (sigio == NULL)
652 return;
e43a034f 653 crit_enter();
984263bc 654 *(sigio->sio_myref) = NULL;
e43a034f 655 crit_exit();
984263bc
MD
656 if (sigio->sio_pgid < 0) {
657 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
658 sigio, sio_pgsigio);
659 } else /* if ((*sigiop)->sio_pgid > 0) */ {
660 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
661 sigio, sio_pgsigio);
662 }
663 crfree(sigio->sio_ucred);
efda3bd0 664 kfree(sigio, M_SIGIO);
984263bc
MD
665}
666
667/* Free a list of sigio structures. */
668void
7bf8660a 669funsetownlst(struct sigiolst *sigiolst)
984263bc
MD
670{
671 struct sigio *sigio;
672
673 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
674 funsetown(sigio);
675}
676
677/*
678 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
679 *
680 * After permission checking, add a sigio structure to the sigio list for
681 * the process or process group.
682 */
683int
7bf8660a 684fsetown(pid_t pgid, struct sigio **sigiop)
984263bc
MD
685{
686 struct proc *proc;
687 struct pgrp *pgrp;
688 struct sigio *sigio;
984263bc
MD
689
690 if (pgid == 0) {
691 funsetown(*sigiop);
692 return (0);
693 }
694 if (pgid > 0) {
695 proc = pfind(pgid);
696 if (proc == NULL)
697 return (ESRCH);
698
699 /*
700 * Policy - Don't allow a process to FSETOWN a process
701 * in another session.
702 *
703 * Remove this test to allow maximum flexibility or
704 * restrict FSETOWN to the current process or process
705 * group for maximum safety.
706 */
707 if (proc->p_session != curproc->p_session)
708 return (EPERM);
709
710 pgrp = NULL;
711 } else /* if (pgid < 0) */ {
712 pgrp = pgfind(-pgid);
713 if (pgrp == NULL)
714 return (ESRCH);
715
716 /*
717 * Policy - Don't allow a process to FSETOWN a process
718 * in another session.
719 *
720 * Remove this test to allow maximum flexibility or
721 * restrict FSETOWN to the current process or process
722 * group for maximum safety.
723 */
724 if (pgrp->pg_session != curproc->p_session)
725 return (EPERM);
726
727 proc = NULL;
728 }
729 funsetown(*sigiop);
efda3bd0 730 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
984263bc
MD
731 if (pgid > 0) {
732 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
733 sigio->sio_proc = proc;
734 } else {
735 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
736 sigio->sio_pgrp = pgrp;
737 }
738 sigio->sio_pgid = pgid;
9910d07b 739 sigio->sio_ucred = crhold(curthread->td_ucred);
984263bc 740 /* It would be convenient if p_ruid was in ucred. */
9910d07b 741 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
984263bc 742 sigio->sio_myref = sigiop;
e43a034f 743 crit_enter();
984263bc 744 *sigiop = sigio;
e43a034f 745 crit_exit();
984263bc
MD
746 return (0);
747}
748
749/*
750 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
751 */
752pid_t
7bf8660a 753fgetown(struct sigio *sigio)
984263bc
MD
754{
755 return (sigio != NULL ? sigio->sio_pgid : 0);
756}
757
758/*
4336d5df 759 * Close many file descriptors.
5b287bba
MD
760 *
761 * MPSAFE
4336d5df 762 */
4336d5df 763int
753fd850 764sys_closefrom(struct closefrom_args *uap)
4336d5df
JS
765{
766 return(kern_closefrom(uap->fd));
767}
768
5b287bba
MD
769/*
770 * Close all file descriptors greater then or equal to fd
771 *
772 * MPSAFE
773 */
4336d5df
JS
774int
775kern_closefrom(int fd)
776{
777 struct thread *td = curthread;
778 struct proc *p = td->td_proc;
779 struct filedesc *fdp;
4336d5df
JS
780
781 KKASSERT(p);
782 fdp = p->p_fd;
783
5b287bba
MD
784 if (fd < 0)
785 return (EINVAL);
5e8cfac8 786
259b8ea0
MD
787 /*
788 * NOTE: This function will skip unassociated descriptors and
789 * reserved descriptors that have not yet been assigned.
790 * fd_lastfile can change as a side effect of kern_close().
791 */
287a8577 792 spin_lock(&fdp->fd_spin);
259b8ea0
MD
793 while (fd <= fdp->fd_lastfile) {
794 if (fdp->fd_files[fd].fp != NULL) {
287a8577 795 spin_unlock(&fdp->fd_spin);
5b287bba 796 /* ok if this races another close */
259b8ea0
MD
797 if (kern_close(fd) == EINTR)
798 return (EINTR);
287a8577 799 spin_lock(&fdp->fd_spin);
259b8ea0
MD
800 }
801 ++fd;
802 }
287a8577 803 spin_unlock(&fdp->fd_spin);
4336d5df
JS
804 return (0);
805}
806
807/*
984263bc 808 * Close a file descriptor.
5b287bba
MD
809 *
810 * MPSAFE
984263bc 811 */
984263bc 812int
753fd850 813sys_close(struct close_args *uap)
984263bc 814{
12693083
MD
815 return(kern_close(uap->fd));
816}
817
5b287bba
MD
818/*
819 * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls
820 */
12693083
MD
821int
822kern_close(int fd)
823{
dadab5e9
MD
824 struct thread *td = curthread;
825 struct proc *p = td->td_proc;
826 struct filedesc *fdp;
41c20dac 827 struct file *fp;
984263bc
MD
828 int error;
829 int holdleaders;
830
dadab5e9
MD
831 KKASSERT(p);
832 fdp = p->p_fd;
833
287a8577 834 spin_lock(&fdp->fd_spin);
5b287bba 835 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
287a8577 836 spin_unlock(&fdp->fd_spin);
984263bc 837 return (EBADF);
5b287bba 838 }
984263bc
MD
839 holdleaders = 0;
840 if (p->p_fdtol != NULL) {
841 /*
842 * Ask fdfree() to sleep to ensure that all relevant
843 * process leaders can be traversed in closef().
844 */
845 fdp->fd_holdleaderscount++;
846 holdleaders = 1;
847 }
848
849 /*
850 * we now hold the fp reference that used to be owned by the descriptor
851 * array.
852 */
287a8577 853 spin_unlock(&fdp->fd_spin);
5a4b22b1 854 if (SLIST_FIRST(&fp->f_klist))
ccafe911 855 knote_fdclose(fp, fdp, fd);
b8477cda 856 error = closef(fp, p);
984263bc 857 if (holdleaders) {
287a8577 858 spin_lock(&fdp->fd_spin);
984263bc
MD
859 fdp->fd_holdleaderscount--;
860 if (fdp->fd_holdleaderscount == 0 &&
861 fdp->fd_holdleaderswakeup != 0) {
862 fdp->fd_holdleaderswakeup = 0;
287a8577 863 spin_unlock(&fdp->fd_spin);
984263bc 864 wakeup(&fdp->fd_holdleaderscount);
5b287bba 865 } else {
287a8577 866 spin_unlock(&fdp->fd_spin);
984263bc
MD
867 }
868 }
869 return (error);
870}
871
004d2de5
MD
872/*
873 * shutdown_args(int fd, int how)
874 */
875int
876kern_shutdown(int fd, int how)
877{
878 struct thread *td = curthread;
879 struct proc *p = td->td_proc;
004d2de5
MD
880 struct file *fp;
881 int error;
882
883 KKASSERT(p);
884
5b287bba 885 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
004d2de5 886 return (EBADF);
87de5057 887 error = fo_shutdown(fp, how);
9f87144f 888 fdrop(fp);
004d2de5
MD
889
890 return (error);
891}
892
3919ced0
MD
893/*
894 * MPALMOSTSAFE
895 */
004d2de5 896int
753fd850 897sys_shutdown(struct shutdown_args *uap)
004d2de5
MD
898{
899 int error;
900
3919ced0 901 get_mplock();
004d2de5 902 error = kern_shutdown(uap->s, uap->how);
3919ced0 903 rel_mplock();
004d2de5
MD
904
905 return (error);
906}
907
1ee6e3c6
MD
908/*
909 * MPSAFE
910 */
984263bc 911int
8f6f8622 912kern_fstat(int fd, struct stat *ub)
984263bc 913{
dadab5e9
MD
914 struct thread *td = curthread;
915 struct proc *p = td->td_proc;
41c20dac 916 struct file *fp;
984263bc
MD
917 int error;
918
dadab5e9 919 KKASSERT(p);
8f6f8622 920
5b287bba 921 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
984263bc 922 return (EBADF);
9910d07b 923 error = fo_stat(fp, ub, td->td_ucred);
9f87144f 924 fdrop(fp);
8f6f8622 925
984263bc
MD
926 return (error);
927}
984263bc
MD
928
929/*
930 * Return status information about a file descriptor.
1ee6e3c6
MD
931 *
932 * MPSAFE
984263bc 933 */
984263bc 934int
753fd850 935sys_fstat(struct fstat_args *uap)
984263bc 936{
8f6f8622 937 struct stat st;
984263bc
MD
938 int error;
939
8f6f8622
DRJ
940 error = kern_fstat(uap->fd, &st);
941
984263bc 942 if (error == 0)
8f6f8622 943 error = copyout(&st, uap->sb, sizeof(st));
984263bc
MD
944 return (error);
945}
946
947/*
984263bc 948 * Return pathconf information about a file descriptor.
3919ced0
MD
949 *
950 * MPALMOSTSAFE
984263bc 951 */
984263bc 952int
753fd850 953sys_fpathconf(struct fpathconf_args *uap)
984263bc 954{
dadab5e9
MD
955 struct thread *td = curthread;
956 struct proc *p = td->td_proc;
984263bc
MD
957 struct file *fp;
958 struct vnode *vp;
959 int error = 0;
960
5b287bba
MD
961 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
962 return (EBADF);
984263bc
MD
963
964 switch (fp->f_type) {
965 case DTYPE_PIPE:
966 case DTYPE_SOCKET:
967 if (uap->name != _PC_PIPE_BUF) {
968 error = EINVAL;
969 } else {
c7114eea 970 uap->sysmsg_result = PIPE_BUF;
984263bc
MD
971 error = 0;
972 }
973 break;
974 case DTYPE_FIFO:
975 case DTYPE_VNODE:
976 vp = (struct vnode *)fp->f_data;
3919ced0 977 get_mplock();
d557216f 978 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
3919ced0 979 rel_mplock();
984263bc
MD
980 break;
981 default:
982 error = EOPNOTSUPP;
983 break;
984 }
9f87144f 985 fdrop(fp);
984263bc
MD
986 return(error);
987}
988
984263bc 989static int fdexpand;
0c52fa62
SG
990SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0,
991 "Number of times a file table has been expanded");
984263bc 992
5b287bba
MD
993/*
994 * Grow the file table so it can hold through descriptor (want).
995 *
996 * The fdp's spinlock must be held exclusively on entry and may be held
997 * exclusively on return. The spinlock may be cycled by the routine.
998 *
999 * MPSAFE
1000 */
69908319 1001static void
5b287bba 1002fdgrow_locked(struct filedesc *fdp, int want)
69908319 1003{
0679adc4
MD
1004 struct fdnode *newfiles;
1005 struct fdnode *oldfiles;
69908319
JH
1006 int nf, extra;
1007
1008 nf = fdp->fd_nfiles;
1009 do {
1010 /* nf has to be of the form 2^n - 1 */
1011 nf = 2 * nf + 1;
1012 } while (nf <= want);
1013
287a8577 1014 spin_unlock(&fdp->fd_spin);
efda3bd0 1015 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
287a8577 1016 spin_lock(&fdp->fd_spin);
69908319
JH
1017
1018 /*
5b287bba
MD
1019 * We could have raced another extend while we were not holding
1020 * the spinlock.
69908319
JH
1021 */
1022 if (fdp->fd_nfiles >= nf) {
287a8577 1023 spin_unlock(&fdp->fd_spin);
efda3bd0 1024 kfree(newfiles, M_FILEDESC);
287a8577 1025 spin_lock(&fdp->fd_spin);
69908319
JH
1026 return;
1027 }
69908319
JH
1028 /*
1029 * Copy the existing ofile and ofileflags arrays
1030 * and zero the new portion of each array.
1031 */
1032 extra = nf - fdp->fd_nfiles;
0679adc4
MD
1033 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1034 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1035
1036 oldfiles = fdp->fd_files;
1037 fdp->fd_files = newfiles;
69908319 1038 fdp->fd_nfiles = nf;
0679adc4 1039
5b287bba 1040 if (oldfiles != fdp->fd_builtin_files) {
287a8577 1041 spin_unlock(&fdp->fd_spin);
efda3bd0 1042 kfree(oldfiles, M_FILEDESC);
287a8577 1043 spin_lock(&fdp->fd_spin);
5b287bba 1044 }
69908319
JH
1045 fdexpand++;
1046}
1047
1048/*
1049 * Number of nodes in right subtree, including the root.
1050 */
1051static __inline int
1052right_subtree_size(int n)
1053{
1054 return (n ^ (n | (n + 1)));
1055}
1056
1057/*
1058 * Bigger ancestor.
1059 */
1060static __inline int
1061right_ancestor(int n)
1062{
1063 return (n | (n + 1));
1064}
1065
1066/*
1067 * Smaller ancestor.
1068 */
1069static __inline int
1070left_ancestor(int n)
1071{
1072 return ((n & (n + 1)) - 1);
1073}
1074
5b287bba
MD
1075/*
1076 * Traverse the in-place binary tree buttom-up adjusting the allocation
1077 * count so scans can determine where free descriptors are located.
1078 *
1079 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1080 */
fa541be6 1081static
69908319 1082void
5b287bba 1083fdreserve_locked(struct filedesc *fdp, int fd, int incr)
69908319
JH
1084{
1085 while (fd >= 0) {
0679adc4
MD
1086 fdp->fd_files[fd].allocated += incr;
1087 KKASSERT(fdp->fd_files[fd].allocated >= 0);
69908319
JH
1088 fd = left_ancestor(fd);
1089 }
1090}
1091
1092/*
259b8ea0
MD
1093 * Reserve a file descriptor for the process. If no error occurs, the
1094 * caller MUST at some point call fsetfd() or assign a file pointer
1095 * or dispose of the reservation.
5b287bba
MD
1096 *
1097 * MPSAFE
69908319 1098 */
984263bc 1099int
dda4b42b 1100fdalloc(struct proc *p, int want, int *result)
984263bc 1101{
41c20dac 1102 struct filedesc *fdp = p->p_fd;
d37c8f7f 1103 struct uidinfo *uip;
69908319
JH
1104 int fd, rsize, rsum, node, lim;
1105
d37c8f7f
MD
1106 /*
1107 * Check dtable size limit
1108 */
287a8577 1109 spin_lock(&p->p_limit->p_spin);
d37c8f7f
MD
1110 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1111 lim = INT_MAX;
1112 else
1113 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
287a8577 1114 spin_unlock(&p->p_limit->p_spin);
a7b16922 1115
d37c8f7f
MD
1116 if (lim > maxfilesperproc)
1117 lim = maxfilesperproc;
1118 if (lim < minfilesperproc)
1119 lim = minfilesperproc;
69908319
JH
1120 if (want >= lim)
1121 return (EMFILE);
d37c8f7f
MD
1122
1123 /*
1124 * Check that the user has not run out of descriptors (non-root only).
1125 * As a safety measure the dtable is allowed to have at least
1126 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1127 */
1128 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1129 uip = p->p_ucred->cr_uidinfo;
1130 if (uip->ui_openfiles > maxfilesperuser) {
1e864525 1131 krateprintf(&krate_uidinfo,
d37c8f7f
MD
1132 "Warning: user %d pid %d (%s) ran out of "
1133 "file descriptors (%d/%d)\n",
1134 p->p_ucred->cr_uid, (int)p->p_pid,
1135 p->p_comm,
1136 uip->ui_openfiles, maxfilesperuser);
1137 return(ENFILE);
1138 }
1139 }
1140
1141 /*
1142 * Grow the dtable if necessary
1143 */
287a8577 1144 spin_lock(&fdp->fd_spin);
69908319 1145 if (want >= fdp->fd_nfiles)
5b287bba 1146 fdgrow_locked(fdp, want);
984263bc
MD
1147
1148 /*
1149 * Search for a free descriptor starting at the higher
1150 * of want or fd_freefile. If that fails, consider
1151 * expanding the ofile array.
259b8ea0
MD
1152 *
1153 * NOTE! the 'allocated' field is a cumulative recursive allocation
1154 * count. If we happen to see a value of 0 then we can shortcut
1155 * our search. Otherwise we run through through the tree going
1156 * down branches we know have free descriptor(s) until we hit a
1157 * leaf node. The leaf node will be free but will not necessarily
1158 * have an allocated field of 0.
984263bc 1159 */
69908319
JH
1160retry:
1161 /* move up the tree looking for a subtree with a free node */
1162 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1163 fd = right_ancestor(fd)) {
0679adc4 1164 if (fdp->fd_files[fd].allocated == 0)
69908319 1165 goto found;
984263bc 1166
69908319 1167 rsize = right_subtree_size(fd);
0679adc4 1168 if (fdp->fd_files[fd].allocated == rsize)
69908319 1169 continue; /* right subtree full */
984263bc
MD
1170
1171 /*
69908319
JH
1172 * Free fd is in the right subtree of the tree rooted at fd.
1173 * Call that subtree R. Look for the smallest (leftmost)
1174 * subtree of R with an unallocated fd: continue moving
1175 * down the left branch until encountering a full left
1176 * subtree, then move to the right.
984263bc 1177 */
69908319
JH
1178 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1179 node = fd + rsize;
0679adc4
MD
1180 rsum += fdp->fd_files[node].allocated;
1181 if (fdp->fd_files[fd].allocated == rsum + rsize) {
69908319 1182 fd = node; /* move to the right */
0679adc4 1183 if (fdp->fd_files[node].allocated == 0)
69908319
JH
1184 goto found;
1185 rsum = 0;
1186 }
984263bc 1187 }
69908319 1188 goto found;
984263bc 1189 }
69908319
JH
1190
1191 /*
1192 * No space in current array. Expand?
1193 */
5b287bba 1194 if (fdp->fd_nfiles >= lim) {
287a8577 1195 spin_unlock(&fdp->fd_spin);
69908319 1196 return (EMFILE);
5b287bba
MD
1197 }
1198 fdgrow_locked(fdp, want);
69908319
JH
1199 goto retry;
1200
1201found:
1202 KKASSERT(fd < fdp->fd_nfiles);
69908319
JH
1203 if (fd > fdp->fd_lastfile)
1204 fdp->fd_lastfile = fd;
1205 if (want <= fdp->fd_freefile)
1206 fdp->fd_freefile = fd;
1207 *result = fd;
0679adc4 1208 KKASSERT(fdp->fd_files[fd].fp == NULL);
259b8ea0
MD
1209 KKASSERT(fdp->fd_files[fd].reserved == 0);
1210 fdp->fd_files[fd].fileflags = 0;
1211 fdp->fd_files[fd].reserved = 1;
5b287bba 1212 fdreserve_locked(fdp, fd, 1);
287a8577 1213 spin_unlock(&fdp->fd_spin);
984263bc
MD
1214 return (0);
1215}
1216
1217/*
1218 * Check to see whether n user file descriptors
1219 * are available to the process p.
5b287bba
MD
1220 *
1221 * MPSAFE
984263bc
MD
1222 */
1223int
7bf8660a 1224fdavail(struct proc *p, int n)
984263bc 1225{
41c20dac 1226 struct filedesc *fdp = p->p_fd;
0679adc4 1227 struct fdnode *fdnode;
41c20dac 1228 int i, lim, last;
984263bc 1229
287a8577 1230 spin_lock(&p->p_limit->p_spin);
d37c8f7f
MD
1231 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1232 lim = INT_MAX;
1233 else
1234 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
287a8577 1235 spin_unlock(&p->p_limit->p_spin);
a7b16922 1236
d37c8f7f
MD
1237 if (lim > maxfilesperproc)
1238 lim = maxfilesperproc;
1239 if (lim < minfilesperproc)
1240 lim = minfilesperproc;
984263bc 1241
287a8577 1242 spin_lock(&fdp->fd_spin);
5b287bba 1243 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
287a8577 1244 spin_unlock(&fdp->fd_spin);
5b287bba
MD
1245 return (1);
1246 }
984263bc 1247 last = min(fdp->fd_nfiles, lim);
0679adc4
MD
1248 fdnode = &fdp->fd_files[fdp->fd_freefile];
1249 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
5b287bba 1250 if (fdnode->fp == NULL && --n <= 0) {
287a8577 1251 spin_unlock(&fdp->fd_spin);
984263bc 1252 return (1);
5b287bba 1253 }
984263bc 1254 }
287a8577 1255 spin_unlock(&fdp->fd_spin);
984263bc
MD
1256 return (0);
1257}
1258
1259/*
b8477cda
MD
1260 * Revoke open descriptors referencing (f_data, f_type)
1261 *
1262 * Any revoke executed within a prison is only able to
1263 * revoke descriptors for processes within that prison.
1264 *
1265 * Returns 0 on success or an error code.
1266 */
1267struct fdrevoke_info {
1268 void *data;
1269 short type;
1270 short unused;
1271 int count;
ea8f324c 1272 int intransit;
b8477cda
MD
1273 struct ucred *cred;
1274 struct file *nfp;
1275};
1276
ea8f324c
MD
1277static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1278static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
b8477cda
MD
1279
1280int
1281fdrevoke(void *f_data, short f_type, struct ucred *cred)
1282{
1283 struct fdrevoke_info info;
1284 int error;
1285
1286 bzero(&info, sizeof(info));
1287 info.data = f_data;
1288 info.type = f_type;
1289 info.cred = cred;
1290 error = falloc(NULL, &info.nfp, NULL);
1291 if (error)
1292 return (error);
ea8f324c
MD
1293
1294 /*
1295 * Scan the file pointer table once. dups do not dup file pointers,
1296 * only descriptors, so there is no leak. Set FREVOKED on the fps
1297 * being revoked.
1298 */
1299 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1300
1301 /*
1302 * If any fps were marked track down the related descriptors
1303 * and close them. Any dup()s at this point will notice
1304 * the FREVOKED already set in the fp and do the right thing.
1305 *
1306 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1307 * socket) bumped the intransit counter and will require a
1308 * scan. Races against fps leaving the socket are closed by
1309 * the socket code checking for FREVOKED.
1310 */
1311 if (info.count)
1312 allproc_scan(fdrevoke_proc_callback, &info);
1313 if (info.intransit)
1314 unp_revoke_gc(info.nfp);
b8477cda 1315 fdrop(info.nfp);
ea8f324c 1316 return(0);
b8477cda
MD
1317}
1318
ea8f324c
MD
1319/*
1320 * Locate matching file pointers directly.
2aa0ab56
MD
1321 *
1322 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
ea8f324c 1323 */
b8477cda 1324static int
ea8f324c
MD
1325fdrevoke_check_callback(struct file *fp, void *vinfo)
1326{
1327 struct fdrevoke_info *info = vinfo;
1328
1329 /*
1330 * File pointers already flagged for revokation are skipped.
1331 */
1332 if (fp->f_flag & FREVOKED)
1333 return(0);
1334
1335 /*
1336 * If revoking from a prison file pointers created outside of
1337 * that prison, or file pointers without creds, cannot be revoked.
1338 */
1339 if (info->cred->cr_prison &&
1340 (fp->f_cred == NULL ||
1341 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1342 return(0);
1343 }
1344
1345 /*
1346 * If the file pointer matches then mark it for revocation. The
1347 * flag is currently only used by unp_revoke_gc().
1348 *
1349 * info->count is a heuristic and can race in a SMP environment.
1350 */
1351 if (info->data == fp->f_data && info->type == fp->f_type) {
1352 atomic_set_int(&fp->f_flag, FREVOKED);
1353 info->count += fp->f_count;
1354 if (fp->f_msgcount)
1355 ++info->intransit;
1356 }
1357 return(0);
1358}
1359
1360/*
1361 * Locate matching file pointers via process descriptor tables.
1362 */
1363static int
1364fdrevoke_proc_callback(struct proc *p, void *vinfo)
b8477cda
MD
1365{
1366 struct fdrevoke_info *info = vinfo;
1367 struct filedesc *fdp;
1368 struct file *fp;
1369 int n;
1370
1371 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1372 return(0);
1373 if (info->cred->cr_prison &&
1374 info->cred->cr_prison != p->p_ucred->cr_prison) {
1375 return(0);
1376 }
1377
1378 /*
1379 * If the controlling terminal of the process matches the
1380 * vnode being revoked we clear the controlling terminal.
1381 *
1382 * The normal spec_close() may not catch this because it
1383 * uses curproc instead of p.
1384 */
1385 if (p->p_session && info->type == DTYPE_VNODE &&
1386 info->data == p->p_session->s_ttyvp) {
1387 p->p_session->s_ttyvp = NULL;
1388 vrele(info->data);
1389 }
1390
1391 /*
0a4a9c77 1392 * Softref the fdp to prevent it from being destroyed
b8477cda 1393 */
287a8577 1394 spin_lock(&p->p_spin);
0a4a9c77 1395 if ((fdp = p->p_fd) == NULL) {
287a8577 1396 spin_unlock(&p->p_spin);
b8477cda 1397 return(0);
0a4a9c77
MD
1398 }
1399 atomic_add_int(&fdp->fd_softrefs, 1);
287a8577 1400 spin_unlock(&p->p_spin);
0a4a9c77
MD
1401
1402 /*
1403 * Locate and close any matching file descriptors.
1404 */
287a8577 1405 spin_lock(&fdp->fd_spin);
b8477cda
MD
1406 for (n = 0; n < fdp->fd_nfiles; ++n) {
1407 if ((fp = fdp->fd_files[n].fp) == NULL)
1408 continue;
ea8f324c 1409 if (fp->f_flag & FREVOKED) {
b8477cda
MD
1410 fhold(info->nfp);
1411 fdp->fd_files[n].fp = info->nfp;
287a8577 1412 spin_unlock(&fdp->fd_spin);
ccafe911 1413 knote_fdclose(fp, fdp, n); /* XXX */
b8477cda 1414 closef(fp, p);
287a8577 1415 spin_lock(&fdp->fd_spin);
ea8f324c 1416 --info->count;
b8477cda
MD
1417 }
1418 }
287a8577 1419 spin_unlock(&fdp->fd_spin);
0a4a9c77 1420 atomic_subtract_int(&fdp->fd_softrefs, 1);
b8477cda
MD
1421 return(0);
1422}
1423
1424/*
39f91578 1425 * falloc:
259b8ea0
MD
1426 * Create a new open file structure and reserve a file decriptor
1427 * for the process that refers to it.
1428 *
f3a2d8c4
MD
1429 * Root creds are checked using lp, or assumed if lp is NULL. If
1430 * resultfd is non-NULL then lp must also be non-NULL. No file
1431 * descriptor is reserved (and no process context is needed) if
1432 * resultfd is NULL.
fad57d0e 1433 *
259b8ea0
MD
1434 * A file pointer with a refcount of 1 is returned. Note that the
1435 * file pointer is NOT associated with the descriptor. If falloc
1436 * returns success, fsetfd() MUST be called to either associate the
1437 * file pointer or clear the reservation.
5b287bba 1438 *
2dd63755 1439 * MPSAFE
984263bc
MD
1440 */
1441int
f3a2d8c4 1442falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
984263bc 1443{
c4cb6d8b
HP
1444 static struct timeval lastfail;
1445 static int curfail;
fad57d0e 1446 struct file *fp;
f3a2d8c4 1447 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
fad57d0e 1448 int error;
c4cb6d8b 1449
fad57d0e
MD
1450 fp = NULL;
1451
1452 /*
1453 * Handle filetable full issues and root overfill.
1454 */
60ee93b9 1455 if (nfiles >= maxfiles - maxfilesrootres &&
f3a2d8c4 1456 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
c4cb6d8b 1457 if (ppsratecheck(&lastfail, &curfail, 1)) {
f3a2d8c4
MD
1458 kprintf("kern.maxfiles limit exceeded by uid %d, "
1459 "please see tuning(7).\n",
1460 cred->cr_ruid);
c4cb6d8b 1461 }
fad57d0e
MD
1462 error = ENFILE;
1463 goto done;
984263bc 1464 }
fad57d0e 1465
984263bc
MD
1466 /*
1467 * Allocate a new file descriptor.
984263bc 1468 */
efda3bd0 1469 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
85fceac1 1470 spin_init(&fp->f_spin);
ccafe911 1471 SLIST_INIT(&fp->f_klist);
984263bc 1472 fp->f_count = 1;
984263bc
MD
1473 fp->f_ops = &badfileops;
1474 fp->f_seqcount = 1;
d37c8f7f 1475 fsetcred(fp, cred);
287a8577 1476 spin_lock(&filehead_spin);
2dd63755 1477 nfiles++;
fad57d0e 1478 LIST_INSERT_HEAD(&filehead, fp, f_list);
287a8577 1479 spin_unlock(&filehead_spin);
fad57d0e 1480 if (resultfd) {
f3a2d8c4 1481 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
9f87144f 1482 fdrop(fp);
fad57d0e 1483 fp = NULL;
39f91578 1484 }
a235f7bb 1485 } else {
fad57d0e 1486 error = 0;
984263bc 1487 }
fad57d0e
MD
1488done:
1489 *resultfp = fp;
64f33bc8 1490 return (error);
fad57d0e
MD
1491}
1492
1493/*
cf9f4e88
MD
1494 * Check for races against a file descriptor by determining that the
1495 * file pointer is still associated with the specified file descriptor,
1496 * and a close is not currently in progress.
1497 *
85fceac1
MD
1498 * MPSAFE
1499 */
85fceac1 1500int
cf9f4e88 1501checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
85fceac1
MD
1502{
1503 int error;
1504
287a8577 1505 spin_lock(&fdp->fd_spin);
cf9f4e88 1506 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
85fceac1
MD
1507 error = EBADF;
1508 else
1509 error = 0;
287a8577 1510 spin_unlock(&fdp->fd_spin);
85fceac1
MD
1511 return (error);
1512}
1513
1514/*
259b8ea0
MD
1515 * Associate a file pointer with a previously reserved file descriptor.
1516 * This function always succeeds.
1517 *
1518 * If fp is NULL, the file descriptor is returned to the pool.
fad57d0e 1519 */
fad57d0e 1520
5b287bba
MD
1521/*
1522 * MPSAFE (exclusive spinlock must be held on call)
1523 */
1524static void
1525fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1526{
259b8ea0
MD
1527 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1528 KKASSERT(fdp->fd_files[fd].reserved != 0);
1529 if (fp) {
fad57d0e 1530 fhold(fp);
259b8ea0
MD
1531 fdp->fd_files[fd].fp = fp;
1532 fdp->fd_files[fd].reserved = 0;
1533 } else {
1534 fdp->fd_files[fd].reserved = 0;
5b287bba
MD
1535 fdreserve_locked(fdp, fd, -1);
1536 fdfixup_locked(fdp, fd);
fad57d0e 1537 }
984263bc
MD
1538}
1539
5b287bba
MD
1540/*
1541 * MPSAFE
1542 */
1543void
f3a2d8c4 1544fsetfd(struct filedesc *fdp, struct file *fp, int fd)
5b287bba 1545{
287a8577 1546 spin_lock(&fdp->fd_spin);
5b287bba 1547 fsetfd_locked(fdp, fp, fd);
287a8577 1548 spin_unlock(&fdp->fd_spin);
5b287bba
MD
1549}
1550
1551/*
1552 * MPSAFE (exclusive spinlock must be held on call)
1553 */
fa541be6 1554static
259b8ea0 1555struct file *
5b287bba 1556funsetfd_locked(struct filedesc *fdp, int fd)
69908319 1557{
259b8ea0
MD
1558 struct file *fp;
1559
1560 if ((unsigned)fd >= fdp->fd_nfiles)
1561 return (NULL);
1562 if ((fp = fdp->fd_files[fd].fp) == NULL)
1563 return (NULL);
0679adc4
MD
1564 fdp->fd_files[fd].fp = NULL;
1565 fdp->fd_files[fd].fileflags = 0;
259b8ea0 1566
5b287bba
MD
1567 fdreserve_locked(fdp, fd, -1);
1568 fdfixup_locked(fdp, fd);
259b8ea0 1569 return(fp);
69908319
JH
1570}
1571
85fceac1
MD
1572/*
1573 * MPSAFE
1574 */
1575int
1576fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1577{
1578 int error;
1579
287a8577 1580 spin_lock(&fdp->fd_spin);
85fceac1
MD
1581 if (((u_int)fd) >= fdp->fd_nfiles) {
1582 error = EBADF;
1583 } else if (fdp->fd_files[fd].fp == NULL) {
1584 error = EBADF;
1585 } else {
1586 *flagsp = fdp->fd_files[fd].fileflags;
1587 error = 0;
1588 }
287a8577 1589 spin_unlock(&fdp->fd_spin);
85fceac1
MD
1590 return (error);
1591}
1592
1593/*
1594 * MPSAFE
1595 */
228b401d
MD
1596int
1597fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1598{
85fceac1
MD
1599 int error;
1600
287a8577 1601 spin_lock(&fdp->fd_spin);
85fceac1
MD
1602 if (((u_int)fd) >= fdp->fd_nfiles) {
1603 error = EBADF;
1604 } else if (fdp->fd_files[fd].fp == NULL) {
1605 error = EBADF;
1606 } else {
1607 fdp->fd_files[fd].fileflags |= add_flags;
1608 error = 0;
1609 }
287a8577 1610 spin_unlock(&fdp->fd_spin);
85fceac1 1611 return (error);
228b401d
MD
1612}
1613
85fceac1
MD
1614/*
1615 * MPSAFE
1616 */
228b401d
MD
1617int
1618fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1619{
85fceac1
MD
1620 int error;
1621
287a8577 1622 spin_lock(&fdp->fd_spin);
85fceac1
MD
1623 if (((u_int)fd) >= fdp->fd_nfiles) {
1624 error = EBADF;
1625 } else if (fdp->fd_files[fd].fp == NULL) {
1626 error = EBADF;
1627 } else {
1628 fdp->fd_files[fd].fileflags &= ~rem_flags;
1629 error = 0;
1630 }
287a8577 1631 spin_unlock(&fdp->fd_spin);
85fceac1 1632 return (error);
228b401d
MD
1633}
1634
d37c8f7f
MD
1635/*
1636 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1637 */
69908319 1638void
d37c8f7f 1639fsetcred(struct file *fp, struct ucred *ncr)
a235f7bb 1640{
d37c8f7f
MD
1641 struct ucred *ocr;
1642 struct uidinfo *uip;
1643
1644 ocr = fp->f_cred;
1645 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1646 if (ocr) {
1647 uip = ocr->cr_uidinfo;
1648 atomic_add_int(&uip->ui_openfiles, -1);
1649 }
1650 if (ncr) {
1651 uip = ncr->cr_uidinfo;
1652 atomic_add_int(&uip->ui_openfiles, 1);
1653 }
1654 }
1655 if (ncr)
1656 crhold(ncr);
1657 fp->f_cred = ncr;
1658 if (ocr)
1659 crfree(ocr);
a235f7bb
MD
1660}
1661
984263bc
MD
1662/*
1663 * Free a file descriptor.
1664 */
5b287bba 1665static
984263bc 1666void
7bf8660a 1667ffree(struct file *fp)
984263bc
MD
1668{
1669 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
287a8577 1670 spin_lock(&filehead_spin);
984263bc 1671 LIST_REMOVE(fp, f_list);
2dd63755 1672 nfiles--;
287a8577 1673 spin_unlock(&filehead_spin);
d37c8f7f 1674 fsetcred(fp, NULL);
28623bf9
MD
1675 if (fp->f_nchandle.ncp)
1676 cache_drop(&fp->f_nchandle);
efda3bd0 1677 kfree(fp, M_FILE);
984263bc
MD
1678}
1679
1680/*
228b401d
MD
1681 * called from init_main, initialize filedesc0 for proc0.
1682 */
1683void
1684fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1685{
1686 p0->p_fd = fdp0;
1687 p0->p_fdtol = NULL;
1688 fdp0->fd_refcnt = 1;
1689 fdp0->fd_cmask = cmask;
1690 fdp0->fd_files = fdp0->fd_builtin_files;
1691 fdp0->fd_nfiles = NDFILE;
259b8ea0 1692 fdp0->fd_lastfile = -1;
85fceac1 1693 spin_init(&fdp0->fd_spin);
228b401d
MD
1694}
1695
1696/*
984263bc 1697 * Build a new filedesc structure.
5b287bba
MD
1698 *
1699 * NOT MPSAFE (vref)
984263bc
MD
1700 */
1701struct filedesc *
7bf8660a 1702fdinit(struct proc *p)
984263bc 1703{
0679adc4 1704 struct filedesc *newfdp;
41c20dac 1705 struct filedesc *fdp = p->p_fd;
984263bc 1706
efda3bd0 1707 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
287a8577 1708 spin_lock(&fdp->fd_spin);
bccde7a3 1709 if (fdp->fd_cdir) {
0679adc4
MD
1710 newfdp->fd_cdir = fdp->fd_cdir;
1711 vref(newfdp->fd_cdir);
28623bf9 1712 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
690a3127 1713 }
bccde7a3
MD
1714
1715 /*
1716 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1717 * proc0, but should unconditionally exist in other processes.
1718 */
1719 if (fdp->fd_rdir) {
0679adc4
MD
1720 newfdp->fd_rdir = fdp->fd_rdir;
1721 vref(newfdp->fd_rdir);
28623bf9 1722 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
bccde7a3
MD
1723 }
1724 if (fdp->fd_jdir) {
0679adc4
MD
1725 newfdp->fd_jdir = fdp->fd_jdir;
1726 vref(newfdp->fd_jdir);
28623bf9 1727 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
690a3127 1728 }
287a8577 1729 spin_unlock(&fdp->fd_spin);
690a3127 1730
984263bc 1731 /* Create the file descriptor table. */
0679adc4
MD
1732 newfdp->fd_refcnt = 1;
1733 newfdp->fd_cmask = cmask;
1734 newfdp->fd_files = newfdp->fd_builtin_files;
1735 newfdp->fd_nfiles = NDFILE;
259b8ea0 1736 newfdp->fd_lastfile = -1;
85fceac1 1737 spin_init(&newfdp->fd_spin);
0679adc4
MD
1738
1739 return (newfdp);
984263bc
MD
1740}
1741
1742/*
1743 * Share a filedesc structure.
5b287bba
MD
1744 *
1745 * MPSAFE
984263bc
MD
1746 */
1747struct filedesc *
7bf8660a 1748fdshare(struct proc *p)
984263bc 1749{
5b287bba
MD
1750 struct filedesc *fdp;
1751
1752 fdp = p->p_fd;
287a8577 1753 spin_lock(&fdp->fd_spin);
5b287bba 1754 fdp->fd_refcnt++;
287a8577 1755 spin_unlock(&fdp->fd_spin);
5b287bba 1756 return (fdp);
984263bc
MD
1757}
1758
1759/*
1760 * Copy a filedesc structure.
5b287bba
MD
1761 *
1762 * MPSAFE
984263bc
MD
1763 */
1764struct filedesc *
7bf8660a 1765fdcopy(struct proc *p)
984263bc 1766{
5b287bba
MD
1767 struct filedesc *fdp = p->p_fd;
1768 struct filedesc *newfdp;
0679adc4 1769 struct fdnode *fdnode;
41c20dac 1770 int i;
5b287bba 1771 int ni;
984263bc 1772
5b287bba
MD
1773 /*
1774 * Certain daemons might not have file descriptors.
1775 */
984263bc
MD
1776 if (fdp == NULL)
1777 return (NULL);
1778
5b287bba
MD
1779 /*
1780 * Allocate the new filedesc and fd_files[] array. This can race
1781 * with operations by other threads on the fdp so we have to be
1782 * careful.
1783 */
efda3bd0 1784 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO);
5b287bba 1785again:
287a8577 1786 spin_lock(&fdp->fd_spin);
5b287bba
MD
1787 if (fdp->fd_lastfile < NDFILE) {
1788 newfdp->fd_files = newfdp->fd_builtin_files;
1789 i = NDFILE;
1790 } else {
1791 /*
1792 * We have to allocate (N^2-1) entries for our in-place
1793 * binary tree. Allow the table to shrink.
1794 */
1795 i = fdp->fd_nfiles;
1796 ni = (i - 1) / 2;
1797 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1798 i = ni;
1799 ni = (i - 1) / 2;
1800 }
287a8577 1801 spin_unlock(&fdp->fd_spin);
77652cad 1802 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
5b287bba
MD
1803 M_FILEDESC, M_WAITOK | M_ZERO);
1804
1805 /*
1806 * Check for race, retry
1807 */
287a8577 1808 spin_lock(&fdp->fd_spin);
5b287bba 1809 if (i <= fdp->fd_lastfile) {
287a8577 1810 spin_unlock(&fdp->fd_spin);
efda3bd0 1811 kfree(newfdp->fd_files, M_FILEDESC);
5b287bba
MD
1812 goto again;
1813 }
1814 }
1815
1816 /*
1817 * Dup the remaining fields. vref() and cache_hold() can be
1818 * safely called while holding the read spinlock on fdp.
1819 *
1820 * The read spinlock on fdp is still being held.
1821 *
1822 * NOTE: vref and cache_hold calls for the case where the vnode
1823 * or cache entry already has at least one ref may be called
1824 * while holding spin locks.
1825 */
1826 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
597aea93 1827 vref(newfdp->fd_cdir);
28623bf9 1828 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
690a3127 1829 }
6bdbb368
DR
1830 /*
1831 * We must check for fd_rdir here, at least for now because
1832 * the init process is created before we have access to the
1833 * rootvode to take a reference to it.
1834 */
5b287bba 1835 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
597aea93 1836 vref(newfdp->fd_rdir);
28623bf9 1837 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
690a3127 1838 }
5b287bba 1839 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
597aea93 1840 vref(newfdp->fd_jdir);
28623bf9 1841 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
690a3127 1842 }
984263bc 1843 newfdp->fd_refcnt = 1;
984263bc 1844 newfdp->fd_nfiles = i;
5b287bba
MD
1845 newfdp->fd_lastfile = fdp->fd_lastfile;
1846 newfdp->fd_freefile = fdp->fd_freefile;
1847 newfdp->fd_cmask = fdp->fd_cmask;
5b287bba 1848 spin_init(&newfdp->fd_spin);
984263bc
MD
1849
1850 /*
5b287bba
MD
1851 * Copy the descriptor table through (i). This also copies the
1852 * allocation state. Then go through and ref the file pointers
1853 * and clean up any KQ descriptors.
1854 *
259b8ea0
MD
1855 * kq descriptors cannot be copied. Since we haven't ref'd the
1856 * copied files yet we can ignore the return value from funsetfd().
5b287bba
MD
1857 *
1858 * The read spinlock on fdp is still being held.
984263bc 1859 */
5b287bba
MD
1860 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1861 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
259b8ea0
MD
1862 fdnode = &newfdp->fd_files[i];
1863 if (fdnode->reserved) {
5b287bba 1864 fdreserve_locked(newfdp, i, -1);
259b8ea0 1865 fdnode->reserved = 0;
5b287bba
MD
1866 fdfixup_locked(newfdp, i);
1867 } else if (fdnode->fp) {
1868 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1869 (void)funsetfd_locked(newfdp, i);
1870 } else {
1871 fhold(fdnode->fp);
1872 }
259b8ea0 1873 }
984263bc 1874 }
287a8577 1875 spin_unlock(&fdp->fd_spin);
984263bc
MD
1876 return (newfdp);
1877}
1878
1879/*
1880 * Release a filedesc structure.
5b287bba
MD
1881 *
1882 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
984263bc
MD
1883 */
1884void
0a4a9c77 1885fdfree(struct proc *p, struct filedesc *repl)
984263bc 1886{
0a4a9c77 1887 struct filedesc *fdp;
0679adc4 1888 struct fdnode *fdnode;
41c20dac 1889 int i;
984263bc
MD
1890 struct filedesc_to_leader *fdtol;
1891 struct file *fp;
1892 struct vnode *vp;
1893 struct flock lf;
1894
0a4a9c77 1895 /*
0a4a9c77
MD
1896 * Certain daemons might not have file descriptors.
1897 */
464e801e
MD
1898 fdp = p->p_fd;
1899 if (fdp == NULL) {
1900 p->p_fd = repl;
984263bc 1901 return;
464e801e 1902 }
984263bc 1903
5b287bba 1904 /*
0a4a9c77 1905 * Severe messing around to follow.
5b287bba 1906 */
287a8577 1907 spin_lock(&fdp->fd_spin);
5b287bba 1908
984263bc
MD
1909 /* Check for special need to clear POSIX style locks */
1910 fdtol = p->p_fdtol;
1911 if (fdtol != NULL) {
1912 KASSERT(fdtol->fdl_refcount > 0,
1913 ("filedesc_to_refcount botch: fdl_refcount=%d",
1914 fdtol->fdl_refcount));
1915 if (fdtol->fdl_refcount == 1 &&
1916 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
259b8ea0
MD
1917 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1918 fdnode = &fdp->fd_files[i];
0679adc4 1919 if (fdnode->fp == NULL ||
259b8ea0 1920 fdnode->fp->f_type != DTYPE_VNODE) {
984263bc 1921 continue;
259b8ea0 1922 }
0679adc4 1923 fp = fdnode->fp;
984263bc 1924 fhold(fp);
287a8577 1925 spin_unlock(&fdp->fd_spin);
5b287bba 1926
984263bc
MD
1927 lf.l_whence = SEEK_SET;
1928 lf.l_start = 0;
1929 lf.l_len = 0;
1930 lf.l_type = F_UNLCK;
1931 vp = (struct vnode *)fp->f_data;
1932 (void) VOP_ADVLOCK(vp,
1933 (caddr_t)p->p_leader,
1934 F_UNLCK,
1935 &lf,
1936 F_POSIX);
9f87144f 1937 fdrop(fp);
287a8577 1938 spin_lock(&fdp->fd_spin);
984263bc
MD
1939 }
1940 }
1941 retry:
1942 if (fdtol->fdl_refcount == 1) {
1943 if (fdp->fd_holdleaderscount > 0 &&
1944 (p->p_leader->p_flag & P_ADVLOCK) != 0) {
1945 /*
1946 * close() or do_dup() has cleared a reference
1947 * in a shared file descriptor table.
1948 */
1949 fdp->fd_holdleaderswakeup = 1;
e590ee86 1950 ssleep(&fdp->fd_holdleaderscount,
7749886d 1951 &fdp->fd_spin, 0, "fdlhold", 0);
984263bc
MD
1952 goto retry;
1953 }
1954 if (fdtol->fdl_holdcount > 0) {
1955 /*
1956 * Ensure that fdtol->fdl_leader
1957 * remains valid in closef().
1958 */
1959 fdtol->fdl_wakeup = 1;
e590ee86 1960 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
984263bc
MD
1961 goto retry;
1962 }
1963 }
1964 fdtol->fdl_refcount--;
1965 if (fdtol->fdl_refcount == 0 &&
1966 fdtol->fdl_holdcount == 0) {
1967 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1968 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
5b287bba 1969 } else {
984263bc 1970 fdtol = NULL;
5b287bba 1971 }
984263bc 1972 p->p_fdtol = NULL;
5b287bba 1973 if (fdtol != NULL) {
287a8577 1974 spin_unlock(&fdp->fd_spin);
efda3bd0 1975 kfree(fdtol, M_FILEDESC_TO_LEADER);
287a8577 1976 spin_lock(&fdp->fd_spin);
5b287bba 1977 }
984263bc 1978 }
5b287bba 1979 if (--fdp->fd_refcnt > 0) {
287a8577
AH
1980 spin_unlock(&fdp->fd_spin);
1981 spin_lock(&p->p_spin);
464e801e 1982 p->p_fd = repl;
287a8577 1983 spin_unlock(&p->p_spin);
984263bc 1984 return;
5b287bba 1985 }
464e801e
MD
1986
1987 /*
1988 * Even though we are the last reference to the structure allproc
1989 * scans may still reference the structure. Maintain proper
1990 * locks until we can replace p->p_fd.
1991 *
1992 * Also note that kqueue's closef still needs to reference the
1993 * fdp via p->p_fd, so we have to close the descriptors before
1994 * we replace p->p_fd.
1995 */
1996 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1997 if (fdp->fd_files[i].fp) {
1998 fp = funsetfd_locked(fdp, i);
1999 if (fp) {
287a8577 2000 spin_unlock(&fdp->fd_spin);
5a4b22b1 2001 if (SLIST_FIRST(&fp->f_klist))
34e191bd 2002 knote_fdclose(fp, fdp, i);
464e801e 2003 closef(fp, p);
287a8577 2004 spin_lock(&fdp->fd_spin);
464e801e
MD
2005 }
2006 }
2007 }
287a8577 2008 spin_unlock(&fdp->fd_spin);
5b287bba 2009
984263bc 2010 /*
464e801e
MD
2011 * Interlock against an allproc scan operations (typically frevoke).
2012 */
287a8577 2013 spin_lock(&p->p_spin);
464e801e 2014 p->p_fd = repl;
287a8577 2015 spin_unlock(&p->p_spin);
464e801e
MD
2016
2017 /*
0a4a9c77
MD
2018 * Wait for any softrefs to go away. This race rarely occurs so
2019 * we can use a non-critical-path style poll/sleep loop. The
2020 * race only occurs against allproc scans.
2021 *
2022 * No new softrefs can occur with the fdp disconnected from the
2023 * process.
2024 */
2025 if (fdp->fd_softrefs) {
2026 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2027 while (fdp->fd_softrefs)
2028 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2029 }
2030
0679adc4 2031 if (fdp->fd_files != fdp->fd_builtin_files)
efda3bd0 2032 kfree(fdp->fd_files, M_FILEDESC);
690a3127 2033 if (fdp->fd_cdir) {
28623bf9 2034 cache_drop(&fdp->fd_ncdir);
984263bc 2035 vrele(fdp->fd_cdir);
690a3127 2036 }
bccde7a3 2037 if (fdp->fd_rdir) {
28623bf9 2038 cache_drop(&fdp->fd_nrdir);
bccde7a3
MD
2039 vrele(fdp->fd_rdir);
2040 }
690a3127 2041 if (fdp->fd_jdir) {
28623bf9 2042 cache_drop(&fdp->fd_njdir);
984263bc 2043 vrele(fdp->fd_jdir);
690a3127 2044 }
efda3bd0 2045 kfree(fdp, M_FILEDESC);
984263bc
MD
2046}
2047
2048/*
fa541be6 2049 * Retrieve and reference the file pointer associated with a descriptor.
85fceac1
MD
2050 *
2051 * MPSAFE
fa541be6
MD
2052 */
2053struct file *
2054holdfp(struct filedesc *fdp, int fd, int flag)
2055{
2056 struct file* fp;
2057
287a8577 2058 spin_lock(&fdp->fd_spin);
85fceac1
MD
2059 if (((u_int)fd) >= fdp->fd_nfiles) {
2060 fp = NULL;
2061 goto done;
2062 }
fa541be6 2063 if ((fp = fdp->fd_files[fd].fp) == NULL)
85fceac1
MD
2064 goto done;
2065 if ((fp->f_flag & flag) == 0 && flag != -1) {
2066 fp = NULL;
2067 goto done;
2068 }
fa541be6 2069 fhold(fp);
85fceac1 2070done:
287a8577 2071 spin_unlock(&fdp->fd_spin);
fa541be6
MD
2072 return (fp);
2073}
2074
2075/*
2076 * holdsock() - load the struct file pointer associated
2077 * with a socket into *fpp. If an error occurs, non-zero
2078 * will be returned and *fpp will be set to NULL.
5b287bba
MD
2079 *
2080 * MPSAFE
fa541be6
MD
2081 */
2082int
5b287bba 2083holdsock(struct filedesc *fdp, int fd, struct file **fpp)
fa541be6
MD
2084{
2085 struct file *fp;
5b287bba 2086 int error;
fa541be6 2087
287a8577 2088 spin_lock(&fdp->fd_spin);
5b287bba
MD
2089 if ((unsigned)fd >= fdp->fd_nfiles) {
2090 error = EBADF;
2091 fp = NULL;
2092 goto done;
2093 }
2094 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2095 error = EBADF;
2096 goto done;
2097 }
2098 if (fp->f_type != DTYPE_SOCKET) {
2099 error = ENOTSOCK;
2100 goto done;
2101 }
fa541be6 2102 fhold(fp);
5b287bba
MD
2103 error = 0;
2104done:
287a8577 2105 spin_unlock(&fdp->fd_spin);
fa541be6
MD
2106 *fpp = fp;
2107 return (error);
2108}
2109
2110/*
5b287bba
MD
2111 * Convert a user file descriptor to a held file pointer.
2112 *
2113 * MPSAFE
228b401d
MD
2114 */
2115int
5b287bba 2116holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
228b401d
MD
2117{
2118 struct file *fp;
5b287bba
MD
2119 int error;
2120
287a8577 2121 spin_lock(&fdp->fd_spin);
5b287bba
MD
2122 if ((unsigned)fd >= fdp->fd_nfiles) {
2123 error = EBADF;
2124 fp = NULL;
2125 goto done;
2126 }
2127 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2128 error = EBADF;
2129 goto done;
2130 }
2131 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
8f9f2535 2132 fp = NULL;
5b287bba
MD
2133 error = EINVAL;
2134 goto done;
2135 }
2136 fhold(fp);
2137 error = 0;
2138done:
287a8577 2139 spin_unlock(&fdp->fd_spin);
228b401d 2140 *fpp = fp;
5b287bba 2141 return (error);
228b401d
MD
2142}
2143
2144/*
984263bc
MD
2145 * For setugid programs, we don't want to people to use that setugidness
2146 * to generate error messages which write to a file which otherwise would
2147 * otherwise be off-limits to the process.
2148 *
2149 * This is a gross hack to plug the hole. A better solution would involve
2150 * a special vop or other form of generalized access control mechanism. We
2151 * go ahead and just reject all procfs file systems accesses as dangerous.
2152 *
2153 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2154 * sufficient. We also don't for check setugidness since we know we are.
2155 */
2156static int
2157is_unsafe(struct file *fp)
2158{
2159 if (fp->f_type == DTYPE_VNODE &&
2160 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2161 return (1);
2162 return (0);
2163}
2164
2165/*
2166 * Make this setguid thing safe, if at all possible.
5b287bba
MD
2167 *
2168 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
984263bc
MD
2169 */
2170void
dadab5e9 2171setugidsafety(struct proc *p)
984263bc
MD
2172{
2173 struct filedesc *fdp = p->p_fd;
41c20dac 2174 int i;
984263bc
MD
2175
2176 /* Certain daemons might not have file descriptors. */
2177 if (fdp == NULL)
2178 return;
2179
2180 /*
0679adc4 2181 * note: fdp->fd_files may be reallocated out from under us while
984263bc
MD
2182 * we are blocked in a close. Be careful!
2183 */
2184 for (i = 0; i <= fdp->fd_lastfile; i++) {
2185 if (i > 2)
2186 break;
0679adc4 2187 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
984263bc
MD
2188 struct file *fp;
2189
984263bc
MD
2190 /*
2191 * NULL-out descriptor prior to close to avoid
2192 * a race while close blocks.
2193 */
ccafe911
MD
2194 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2195 knote_fdclose(fp, fdp, i);
b8477cda 2196 closef(fp, p);
ccafe911 2197 }
984263bc
MD
2198 }
2199 }
984263bc
MD
2200}
2201
2202/*
2203 * Close any files on exec?
5b287bba
MD
2204 *
2205 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
984263bc
MD
2206 */
2207void
dadab5e9 2208fdcloseexec(struct proc *p)
984263bc
MD
2209{
2210 struct filedesc *fdp = p->p_fd;
41c20dac 2211 int i;
984263bc
MD
2212
2213 /* Certain daemons might not have file descriptors. */
2214 if (fdp == NULL)
2215 return;
2216
2217 /*
0679adc4
MD
2218 * We cannot cache fd_files since operations may block and rip
2219 * them out from under us.
984263bc
MD
2220 */
2221 for (i = 0; i <= fdp->fd_lastfile; i++) {
0679adc4
MD
2222 if (fdp->fd_files[i].fp != NULL &&
2223 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
984263bc
MD
2224 struct file *fp;
2225
984263bc
MD
2226 /*
2227 * NULL-out descriptor prior to close to avoid
2228 * a race while close blocks.
2229 */
ccafe911
MD
2230 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2231 knote_fdclose(fp, fdp, i);
b8477cda 2232 closef(fp, p);
ccafe911 2233 }
984263bc
MD
2234 }
2235 }
984263bc
MD
2236}
2237
2238/*
2239 * It is unsafe for set[ug]id processes to be started with file
2240 * descriptors 0..2 closed, as these descriptors are given implicit
2241 * significance in the Standard C library. fdcheckstd() will create a
2242 * descriptor referencing /dev/null for each of stdin, stdout, and
2243 * stderr that is not already open.
5b287bba
MD
2244 *
2245 * NOT MPSAFE - calls falloc, vn_open, etc
984263bc
MD
2246 */
2247int
f3a2d8c4 2248fdcheckstd(struct lwp *lp)
984263bc 2249{
fad57d0e 2250 struct nlookupdata nd;
dadab5e9
MD
2251 struct filedesc *fdp;
2252 struct file *fp;
973c11b9 2253 int retval;
259b8ea0 2254 int i, error, flags, devnull;
984263bc 2255
f3a2d8c4 2256 fdp = lp->lwp_proc->p_fd;
ea8f324c
MD
2257 if (fdp == NULL)
2258 return (0);
2259 devnull = -1;
2260 error = 0;
2261 for (i = 0; i < 3; i++) {
0679adc4 2262 if (fdp->fd_files[i].fp != NULL)
fad57d0e
MD
2263 continue;
2264 if (devnull < 0) {
f3a2d8c4 2265 if ((error = falloc(lp, &fp, &devnull)) != 0)
fad57d0e
MD
2266 break;
2267
2268 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2269 NLC_FOLLOW|NLC_LOCKVP);
2270 flags = FREAD | FWRITE;
2271 if (error == 0)
2272 error = vn_open(&nd, fp, flags, 0);
2273 if (error == 0)
f3a2d8c4 2274 fsetfd(fdp, fp, devnull);
259b8ea0 2275 else
f3a2d8c4 2276 fsetfd(fdp, NULL, devnull);
9f87144f 2277 fdrop(fp);
fad57d0e
MD
2278 nlookup_done(&nd);
2279 if (error)
2280 break;
259b8ea0 2281 KKASSERT(i == devnull);
fad57d0e
MD
2282 } else {
2283 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2284 if (error != 0)
2285 break;
2286 }
ea8f324c
MD
2287 }
2288 return (error);
984263bc
MD
2289}
2290
2291/*
2292 * Internal form of close.
2293 * Decrement reference count on file structure.
dadab5e9 2294 * Note: td and/or p may be NULL when closing a file
984263bc 2295 * that was being passed in a message.
5b287bba
MD
2296 *
2297 * MPALMOSTSAFE - acquires mplock for VOP operations
984263bc
MD
2298 */
2299int
b8477cda 2300closef(struct file *fp, struct proc *p)
984263bc
MD
2301{
2302 struct vnode *vp;
2303 struct flock lf;
2304 struct filedesc_to_leader *fdtol;
2305
2306 if (fp == NULL)
2307 return (0);
b8477cda 2308
984263bc
MD
2309 /*
2310 * POSIX record locking dictates that any close releases ALL
2311 * locks owned by this process. This is handled by setting
2312 * a flag in the unlock to free ONLY locks obeying POSIX
2313 * semantics, and not to free BSD-style file locks.
2314 * If the descriptor was in a message, POSIX-style locks
2315 * aren't passed with the descriptor.
2316 */
7749886d
MD
2317 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2318 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2319 ) {
2320 get_mplock();
984263bc
MD
2321 if ((p->p_leader->p_flag & P_ADVLOCK) != 0) {
2322 lf.l_whence = SEEK_SET;
2323 lf.l_start = 0;
2324 lf.l_len = 0;
2325 lf.l_type = F_UNLCK;
2326 vp = (struct vnode *)fp->f_data;
2327 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2328 &lf, F_POSIX);
2329 }
2330 fdtol = p->p_fdtol;
2331 if (fdtol != NULL) {
2332 /*
2333 * Handle special case where file descriptor table
2334 * is shared between multiple process leaders.
2335 */
2336 for (fdtol = fdtol->fdl_next;
2337 fdtol != p->p_fdtol;
2338 fdtol = fdtol->fdl_next) {
2339 if ((fdtol->fdl_leader->p_flag &
2340 P_ADVLOCK) == 0)
2341 continue;
2342 fdtol->fdl_holdcount++;
2343 lf.l_whence = SEEK_SET;
2344 lf.l_start = 0;
2345 lf.l_len = 0;
2346 lf.l_type = F_UNLCK;
2347 vp = (struct vnode *)fp->f_data;
2348 (void) VOP_ADVLOCK(vp,
661fc1a0 2349 (caddr_t)fdtol->fdl_leader,
984263bc
MD
2350 F_UNLCK, &lf, F_POSIX);
2351 fdtol->fdl_holdcount--;
2352 if (fdtol->fdl_holdcount == 0 &&
2353 fdtol->fdl_wakeup != 0) {
2354 fdtol->fdl_wakeup = 0;
2355 wakeup(fdtol);
2356 }
2357 }
2358 }
7749886d 2359 rel_mplock();
984263bc 2360 }
9f87144f 2361 return (fdrop(fp));
984263bc
MD
2362}
2363
85fceac1
MD
2364/*
2365 * MPSAFE
2366 *
2367 * fhold() can only be called if f_count is already at least 1 (i.e. the
2368 * caller of fhold() already has a reference to the file pointer in some
5b287bba
MD
2369 * manner or other).
2370 *
bc1eb676
MD
2371 * f_count is not spin-locked. Instead, atomic ops are used for
2372 * incrementing, decrementing, and handling the 1->0 transition.
85fceac1
MD
2373 */
2374void
2375fhold(struct file *fp)
2376{
2377 atomic_add_int(&fp->f_count, 1);
2378}
2379
2380/*
bc1eb676 2381 * fdrop() - drop a reference to a descriptor
85fceac1 2382 *
5b287bba 2383 * MPALMOSTSAFE - acquires mplock for final close sequence
85fceac1 2384 */
984263bc 2385int
9f87144f 2386fdrop(struct file *fp)
984263bc
MD
2387{
2388 struct flock lf;
2389 struct vnode *vp;
2390 int error;
2391
bc1eb676
MD
2392 /*
2393 * A combined fetch and subtract is needed to properly detect
2394 * 1->0 transitions, otherwise two cpus dropping from a ref
2395 * count of 2 might both try to run the 1->0 code.
2396 */
2397 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
984263bc 2398 return (0);
85fceac1 2399
34e191bd 2400 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
5b287bba
MD
2401 get_mplock();
2402
85fceac1
MD
2403 /*
2404 * The last reference has gone away, we own the fp structure free
2405 * and clear.
2406 */
984263bc
MD
2407 if (fp->f_count < 0)
2408 panic("fdrop: count < 0");
7749886d
MD
2409 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2410 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2411 ) {
984263bc
MD
2412 lf.l_whence = SEEK_SET;
2413 lf.l_start = 0;
2414 lf.l_len = 0;
2415 lf.l_type = F_UNLCK;
2416 vp = (struct vnode *)fp->f_data;
71c18fe3 2417 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
984263bc
MD
2418 }
2419 if (fp->f_ops != &badfileops)
87de5057 2420 error = fo_close(fp);
984263bc
MD
2421 else
2422 error = 0;
2423 ffree(fp);
5b287bba 2424 rel_mplock();
984263bc
MD
2425 return (error);
2426}
2427
2428/*
2429 * Apply an advisory lock on a file descriptor.
2430 *
2431 * Just attempt to get a record lock of the requested type on
2432 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3919ced0
MD
2433 *
2434 * MPALMOSTSAFE
984263bc 2435 */
984263bc 2436int
753fd850 2437sys_flock(struct flock_args *uap)
984263bc 2438{
41c20dac 2439 struct proc *p = curproc;
41c20dac 2440 struct file *fp;
984263bc
MD
2441 struct vnode *vp;
2442 struct flock lf;
5b287bba 2443 int error;
984263bc 2444
5b287bba 2445 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
984263bc 2446 return (EBADF);
3919ced0 2447 get_mplock();
5b287bba
MD
2448 if (fp->f_type != DTYPE_VNODE) {
2449 error = EOPNOTSUPP;
2450 goto done;
2451 }
984263bc
MD
2452 vp = (struct vnode *)fp->f_data;
2453 lf.l_whence = SEEK_SET;
2454 lf.l_start = 0;
2455 lf.l_len = 0;
2456 if (uap->how & LOCK_UN) {
2457 lf.l_type = F_UNLCK;
2458 fp->f_flag &= ~FHASLOCK;
5b287bba
MD
2459 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2460 goto done;
984263bc
MD
2461 }
2462 if (uap->how & LOCK_EX)
2463 lf.l_type = F_WRLCK;
2464 else if (uap->how & LOCK_SH)
2465 lf.l_type = F_RDLCK;
5b287bba
MD
2466 else {
2467 error = EBADF;
2468 goto done;
2469 }
984263bc
MD
2470 fp->f_flag |= FHASLOCK;
2471 if (uap->how & LOCK_NB)
5b287bba
MD
2472 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2473 else
2474 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2475done:
3919ced0 2476 rel_mplock();
5b287bba
MD
2477 fdrop(fp);
2478 return (error);
984263bc
MD
2479}
2480
2481/*
2482 * File Descriptor pseudo-device driver (/dev/fd/).
2483 *
2484 * Opening minor device N dup()s the file (if any) connected to file
2485 * descriptor N belonging to the calling process. Note that this driver
2486 * consists of only the ``open()'' routine, because all subsequent
2487 * references to this file will be direct to the other driver.
2488 */
984263bc 2489static int
fef8985e 2490fdopen(struct dev_open_args *ap)
984263bc 2491{
fef8985e
MD
2492 thread_t td = curthread;
2493
fbfe4e7d 2494 KKASSERT(td->td_lwp != NULL);
984263bc
MD
2495
2496 /*
fbfe4e7d 2497 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
984263bc
MD
2498 * the file descriptor being sought for duplication. The error
2499 * return ensures that the vnode for this device will be released
2500 * by vn_open. Open will detect this special error and take the
2501 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2502 * will simply report the error.
2503 */
fef8985e 2504 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
984263bc
MD
2505 return (ENODEV);
2506}
2507
2508/*
259b8ea0
MD
2509 * The caller has reserved the file descriptor dfd for us. On success we
2510 * must fsetfd() it. On failure the caller will clean it up.
5b287bba 2511 *
aac0aabd 2512 * MPSAFE
984263bc
MD
2513 */
2514int
f3a2d8c4 2515dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
984263bc 2516{
41c20dac 2517 struct file *wfp;
259b8ea0 2518 struct file *xfp;
ea8f324c 2519 int werror;
984263bc 2520
259b8ea0 2521 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
984263bc 2522 return (EBADF);
984263bc
MD
2523
2524 /*
ea8f324c
MD
2525 * Close a revoke/dup race. Duping a descriptor marked as revoked
2526 * will dup a dummy descriptor instead of the real one.
2527 */
2528 if (wfp->f_flag & FREVOKED) {
2529 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2530 fdrop(wfp);
2531 wfp = NULL;
2532 werror = falloc(NULL, &wfp, NULL);
2533 if (werror)
2534 return (werror);
2535 }
2536
2537 /*
984263bc
MD
2538 * There are two cases of interest here.
2539 *
259b8ea0 2540 * For ENODEV simply dup sfd to file descriptor dfd and return.
984263bc 2541 *
259b8ea0
MD
2542 * For ENXIO steal away the file structure from sfd and store it
2543 * dfd. sfd is effectively closed by this operation.
984263bc
MD
2544 *
2545 * Any other error code is just returned.
2546 */
2547 switch (error) {
2548 case ENODEV:
2549 /*
2550 * Check that the mode the file is being opened for is a
2551 * subset of the mode of the existing descriptor.
2552 */
ea8f324c
MD
2553 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2554 error = EACCES;
2555 break;
2556 }
287a8577 2557 spin_lock(&fdp->fd_spin);
259b8ea0 2558 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
aac0aabd 2559 fsetfd_locked(fdp, wfp, dfd);
287a8577 2560 spin_unlock(&fdp->fd_spin);
259b8ea0
MD
2561 error = 0;
2562 break;
984263bc
MD
2563 case ENXIO:
2564 /*
2565 * Steal away the file pointer from dfd, and stuff it into indx.
2566 */
287a8577 2567 spin_lock(&fdp->fd_spin);
259b8ea0 2568 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
f3a2d8c4 2569 fsetfd(fdp, wfp, dfd);
aac0aabd 2570 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
287a8577 2571 spin_unlock(&fdp->fd_spin);
259b8ea0 2572 fdrop(xfp);
aac0aabd 2573 } else {
287a8577 2574 spin_unlock(&fdp->fd_spin);
aac0aabd 2575 }
259b8ea0
MD
2576 error = 0;
2577 break;
984263bc 2578 default:
259b8ea0 2579 break;
984263bc 2580 }
259b8ea0
MD
2581 fdrop(wfp);
2582 return (error);
984263bc
MD
2583}
2584
5b287bba
MD
2585/*
2586 * NOT MPSAFE - I think these refer to a common file descriptor table
2587 * and we need to spinlock that to link fdtol in.
2588 */
984263bc
MD
2589struct filedesc_to_leader *
2590filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2591 struct proc *leader)
2592{
2593 struct filedesc_to_leader *fdtol;
2594
77652cad 2595 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
7bf8660a 2596 M_FILEDESC_TO_LEADER, M_WAITOK);
984263bc
MD
2597 fdtol->fdl_refcount = 1;
2598 fdtol->fdl_holdcount = 0;
2599 fdtol->fdl_wakeup = 0;
2600 fdtol->fdl_leader = leader;
2601 if (old != NULL) {
2602 fdtol->fdl_next = old->fdl_next;
2603 fdtol->fdl_prev = old;
2604 old->fdl_next = fdtol;
2605 fdtol->fdl_next->fdl_prev = fdtol;
2606 } else {
2607 fdtol->fdl_next = fdtol;
2608 fdtol->fdl_prev = fdtol;
2609 }
2610 return fdtol;
2611}
2612
2613/*
2dd63755 2614 * Scan all file pointers in the system. The callback is made with
ea8f324c 2615 * the master list spinlock held exclusively.
2dd63755
MD
2616 *
2617 * MPSAFE
2dd63755
MD
2618 */
2619void
2620allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2621{
2622 struct file *fp;
2623 int res;
2624
287a8577 2625 spin_lock(&filehead_spin);
2dd63755 2626 LIST_FOREACH(fp, &filehead, f_list) {
2dd63755 2627 res = callback(fp, data);
2dd63755
MD
2628 if (res < 0)
2629 break;
2630 }
287a8577 2631 spin_unlock(&filehead_spin);
2dd63755
MD
2632}
2633
2634/*
984263bc 2635 * Get file structures.
5b287bba
MD
2636 *
2637 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
984263bc 2638 */
8fa76237
MD
2639
2640struct sysctl_kern_file_info {
2641 int count;
2642 int error;
2643 struct sysctl_req *req;
2644};
2645
2646static int sysctl_kern_file_callback(struct proc *p, void *data);
2647
984263bc
MD
2648static int
2649sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2650{
8fa76237 2651 struct sysctl_kern_file_info info;
984263bc
MD
2652
2653 /*
7b124c9f
JS
2654 * Note: because the number of file descriptors is calculated
2655 * in different ways for sizing vs returning the data,
2656 * there is information leakage from the first loop. However,
2657 * it is of a similar order of magnitude to the leakage from
2658 * global system statistics such as kern.openfiles.
6d132b4d
MD
2659 *
2660 * When just doing a count, note that we cannot just count
2661 * the elements and add f_count via the filehead list because
2662 * threaded processes share their descriptor table and f_count might
2663 * still be '1' in that case.
e38910da
MD
2664 *
2665 * Since the SYSCTL op can block, we must hold the process to
2666 * prevent it being ripped out from under us either in the
2667 * file descriptor loop or in the greater LIST_FOREACH. The
2668 * process may be in varying states of disrepair. If the process
2669 * is in SZOMB we may have caught it just as it is being removed
2670 * from the allproc list, we must skip it in that case to maintain
2671 * an unbroken chain through the allproc list.
984263bc 2672 */
8fa76237
MD
2673 info.count = 0;
2674 info.error = 0;
2675 info.req = req;
2676 allproc_scan(sysctl_kern_file_callback, &info);
6d132b4d
MD
2677
2678 /*
2679 * When just calculating the size, overestimate a bit to try to
2680 * prevent system activity from causing the buffer-fill call
2681 * to fail later on.
2682 */
2683 if (req->oldptr == NULL) {
8fa76237
MD
2684 info.count = (info.count + 16) + (info.count / 10);
2685 info.error = SYSCTL_OUT(req, NULL,
2686 info.count * sizeof(struct kinfo_file));
6d132b4d 2687 }
8fa76237
MD
2688 return (info.error);
2689}
2690
2691static int
2692sysctl_kern_file_callback(struct proc *p, void *data)
2693{
2694 struct sysctl_kern_file_info *info = data;
2695 struct kinfo_file kf;
2696 struct filedesc *fdp;
2697 struct file *fp;
2698 uid_t uid;
2699 int n;
2700
416d05d7 2701 if (p->p_stat == SIDL || p->p_stat == SZOMB)
8fa76237 2702 return(0);
9910d07b 2703 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
8fa76237 2704 return(0);
0a4a9c77
MD
2705
2706 /*
2707 * Softref the fdp to prevent it from being destroyed
2708 */
287a8577 2709 spin_lock(&p->p_spin);
0a4a9c77 2710 if ((fdp = p->p_fd) == NULL) {
287a8577 2711 spin_unlock(&p->p_spin);
8fa76237 2712 return(0);
0a4a9c77
MD
2713 }
2714 atomic_add_int(&fdp->fd_softrefs, 1);
287a8577 2715 spin_unlock(&p->p_spin);
0a4a9c77
MD
2716
2717 /*
2718 * The fdp's own spinlock prevents the contents from being
2719 * modified.
2720 */
287a8577 2721 spin_lock(&fdp->fd_spin);
8fa76237
MD
2722 for (n = 0; n < fdp->fd_nfiles; ++n) {
2723 if ((fp = fdp->fd_files[n].fp) == NULL)
2724 continue;
2725 if (info->req->oldptr == NULL) {
2726 ++info->count;
2727 } else {
2728 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2729 kcore_make_file(&kf, fp, p->p_pid, uid, n);
287a8577 2730 spin_unlock(&fdp->fd_spin);
8fa76237 2731 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
287a8577 2732 spin_lock(&fdp->fd_spin);
8fa76237
MD
2733 if (info->error)
2734 break;
2735 }
2736 }
287a8577 2737 spin_unlock(&fdp->fd_spin);
0a4a9c77 2738 atomic_subtract_int(&fdp->fd_softrefs, 1);
8fa76237
MD
2739 if (info->error)
2740 return(-1);
2741 return(0);
984263bc
MD
2742}
2743
2744SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2745 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2746
d37c8f7f
MD
2747SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2748 &minfilesperproc, 0, "Minimum files allowed open per process");
984263bc
MD
2749SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2750 &maxfilesperproc, 0, "Maximum files allowed open per process");
d37c8f7f
MD
2751SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2752 &maxfilesperuser, 0, "Maximum files allowed open per user");
984263bc
MD
2753
2754SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2755 &maxfiles, 0, "Maximum number of files");
2756
60ee93b9
MD
2757SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2758 &maxfilesrootres, 0, "Descriptors reserved for root use");
2759
984263bc
MD
2760SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2761 &nfiles, 0, "System-wide number of open files");
2762
2763static void
2764fildesc_drvinit(void *unused)
2765{
2766 int fd;
2767
e4c9c0c8 2768 for (fd = 0; fd < NUMFDESC; fd++) {
fef8985e 2769 make_dev(&fildesc_ops, fd,
3e82b46c 2770 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
e4c9c0c8 2771 }
cd29885a 2772
fef8985e
MD
2773 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2774 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2775 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
984263bc
MD
2776}
2777
d9b2033e
MD
2778/*
2779 * MPSAFE
2780 */
984263bc 2781struct fileops badfileops = {
b2d248cb
MD
2782 .fo_read = badfo_readwrite,
2783 .fo_write = badfo_readwrite,
2784 .fo_ioctl = badfo_ioctl,
2785 .fo_poll = badfo_poll,
2786 .fo_kqfilter = badfo_kqfilter,
2787 .fo_stat = badfo_stat,
2788 .fo_close = badfo_close,
2789 .fo_shutdown = badfo_shutdown
984263bc
MD
2790};
2791
a724d72b 2792int
dadab5e9
MD
2793badfo_readwrite(
2794 struct file *fp,
2795 struct uio *uio,
2796 struct ucred *cred,
87de5057 2797 int flags
dadab5e9 2798) {
984263bc
MD
2799 return (EBADF);
2800}
2801
a724d72b 2802int
87baaf0c
MD
2803badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2804 struct ucred *cred, struct sysmsg *msgv)
984263bc 2805{
984263bc
MD
2806 return (EBADF);
2807}
2808
a724d72b 2809int
87de5057 2810badfo_poll(struct file *fp, int events, struct ucred *cred)
984263bc 2811{
984263bc
MD
2812 return (0);
2813}
2814
d9b2033e 2815/*
304d60e8
MD
2816 * Must return an error to prevent registration, typically
2817 * due to a revoked descriptor (file_filtops assigned).
d9b2033e 2818 */
a724d72b 2819int
dadab5e9 2820badfo_kqfilter(struct file *fp, struct knote *kn)
984263bc 2821{
304d60e8 2822 return (EOPNOTSUPP);
984263bc
MD
2823}
2824
304d60e8
MD
2825/*
2826 * MPSAFE
2827 */
a724d72b 2828int
87de5057 2829badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
984263bc 2830{
984263bc
MD
2831 return (EBADF);
2832}
2833
d9b2033e
MD
2834/*
2835 * MPSAFE
2836 */
a724d72b 2837int
87de5057 2838badfo_close(struct file *fp)
984263bc 2839{
984263bc
MD
2840 return (EBADF);
2841}
2842
d9b2033e
MD
2843/*
2844 * MPSAFE
2845 */
a724d72b 2846int
87de5057 2847badfo_shutdown(struct file *fp, int how)
004d2de5
MD
2848{
2849 return (EBADF);
2850}
2851
d9b2033e
MD
2852/*
2853 * MPSAFE
2854 */
004d2de5 2855int
87de5057 2856nofo_shutdown(struct file *fp, int how)
004d2de5
MD
2857{
2858 return (EOPNOTSUPP);
2859}
2860
984263bc
MD
2861SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2862 fildesc_drvinit,NULL)