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