1 /*
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 *
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. Neither the name of the University nor the names of its contributors
52 * may be used to endorse or promote products derived from this software
53 * without specific prior written permission.
54 *
55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65 * SUCH DAMAGE.
66 *
67 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94
68 * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $
69 */
70
71 #include "opt_compat.h"
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/malloc.h>
75 #include <sys/sysproto.h>
76 #include <sys/conf.h>
77 #include <sys/device.h>
78 #include <sys/file.h>
79 #include <sys/filedesc.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/vnode.h>
83 #include <sys/proc.h>
84 #include <sys/nlookup.h>
85 #include <sys/stat.h>
86 #include <sys/filio.h>
87 #include <sys/fcntl.h>
88 #include <sys/unistd.h>
89 #include <sys/resourcevar.h>
90 #include <sys/event.h>
91 #include <sys/kern_syscall.h>
92 #include <sys/kcore.h>
93 #include <sys/kinfo.h>
94 #include <sys/un.h>
95
96 #include <vm/vm.h>
97 #include <vm/vm_extern.h>
98
99 #include <sys/thread2.h>
100 #include <sys/file2.h>
101 #include <sys/spinlock2.h>
102
103 static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd);
104 static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr);
105 static struct file *funsetfd_locked (struct filedesc *fdp, int fd);
106 static void ffree(struct file *fp);
107
108 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
109 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
110 "file desc to leader structures");
111 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
112 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
113
114 static struct krate krate_uidinfo = { .freq = 1 };
115
116 static d_open_t fdopen;
117 #define NUMFDESC 64
118
119 #define CDEV_MAJOR 22
120 static struct dev_ops fildesc_ops = {
121 { "FD", 0, 0 },
122 .d_open = fdopen,
123 };
124
125 /*
126 * Descriptor management.
127 */
128 static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead);
129 static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin);
130 static int nfiles; /* actual number of open files */
131 extern int cmask;
132
133 /*
134 * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared.
135 *
136 * MPSAFE - must be called with fdp->fd_spin exclusively held
137 */
138 static __inline
139 void
140 fdfixup_locked(struct filedesc *fdp, int fd)
141 {
142 if (fd < fdp->fd_freefile) {
143 fdp->fd_freefile = fd;
144 }
145 while (fdp->fd_lastfile >= 0 &&
146 fdp->fd_files[fdp->fd_lastfile].fp == NULL &&
147 fdp->fd_files[fdp->fd_lastfile].reserved == 0
148 ) {
149 --fdp->fd_lastfile;
150 }
151 }
152
153 /*
154 * System calls on descriptors.
155 *
156 * MPSAFE
157 */
158 int
159 sys_getdtablesize(struct getdtablesize_args *uap)
160 {
161 struct proc *p = curproc;
162 struct plimit *limit = p->p_limit;
163 int dtsize;
164
165 spin_lock(&limit->p_spin);
166 if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
167 dtsize = INT_MAX;
168 else
169 dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur;
170 spin_unlock(&limit->p_spin);
171
172 if (dtsize > maxfilesperproc)
173 dtsize = maxfilesperproc;
174 if (dtsize < minfilesperproc)
175 dtsize = minfilesperproc;
176 if (p->p_ucred->cr_uid && dtsize > maxfilesperuser)
177 dtsize = maxfilesperuser;
178 uap->sysmsg_result = dtsize;
179 return (0);
180 }
181
182 /*
183 * Duplicate a file descriptor to a particular value.
184 *
185 * note: keep in mind that a potential race condition exists when closing
186 * descriptors from a shared descriptor table (via rfork).
187 *
188 * MPSAFE
189 */
190 int
191 sys_dup2(struct dup2_args *uap)
192 {
193 int error;
194 int fd = 0;
195
196 error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd);
197 uap->sysmsg_fds[0] = fd;
198
199 return (error);
200 }
201
202 /*
203 * Duplicate a file descriptor.
204 *
205 * MPSAFE
206 */
207 int
208 sys_dup(struct dup_args *uap)
209 {
210 int error;
211 int fd = 0;
212
213 error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd);
214 uap->sysmsg_fds[0] = fd;
215
216 return (error);
217 }
218
219 /*
220 * MPALMOSTSAFE - acquires mplock for fp operations
221 */
222 int
223 kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred)
224 {
225 struct thread *td = curthread;
226 struct proc *p = td->td_proc;
227 struct file *fp;
228 struct vnode *vp;
229 u_int newmin;
230 u_int oflags;
231 u_int nflags;
232 int tmp, error, flg = F_POSIX;
233
234 KKASSERT(p);
235
236 /*
237 * Operations on file descriptors that do not require a file pointer.
238 */
239 switch (cmd) {
240 case F_GETFD:
241 error = fgetfdflags(p->p_fd, fd, &tmp);
242 if (error == 0)
243 dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0;
244 return (error);
245
246 case F_SETFD:
247 if (dat->fc_cloexec & FD_CLOEXEC)
248 error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE);
249 else
250 error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE);
251 return (error);
252 case F_DUPFD:
253 newmin = dat->fc_fd;
254 error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd);
255 return (error);
256 case F_DUP2FD:
257 newmin = dat->fc_fd;
258 error = kern_dup(DUP_FIXED, fd, newmin, &dat->fc_fd);
259 return (error);
260 case F_DUPFD_CLOEXEC:
261 newmin = dat->fc_fd;
262 error = kern_dup(DUP_VARIABLE | DUP_CLOEXEC, fd, newmin,
263 &dat->fc_fd);
264 return (error);
265 case F_DUP2FD_CLOEXEC:
266 newmin = dat->fc_fd;
267 error = kern_dup(DUP_FIXED | DUP_CLOEXEC, fd, newmin,
268 &dat->fc_fd);
269 return (error);
270 default:
271 break;
272 }
273
274 /*
275 * Operations on file pointers
276 */
277 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
278 return (EBADF);
279
280 switch (cmd) {
281 case F_GETFL:
282 dat->fc_flags = OFLAGS(fp->f_flag);
283 error = 0;
284 break;
285
286 case F_SETFL:
287 oflags = fp->f_flag;
288 nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS;
289 nflags |= oflags & ~FCNTLFLAGS;
290
291 error = 0;
292 if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY))
293 error = EINVAL;
294 if (error == 0 && ((nflags ^ oflags) & FASYNC)) {
295 tmp = nflags & FASYNC;
296 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp,
297 cred, NULL);
298 }
299 if (error == 0)
300 fp->f_flag = nflags;
301 break;
302
303 case F_GETOWN:
304 error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner,
305 cred, NULL);
306 break;
307
308 case F_SETOWN:
309 error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner,
310 cred, NULL);
311 break;
312
313 case F_SETLKW:
314 flg |= F_WAIT;
315 /* Fall into F_SETLK */
316
317 case F_SETLK:
318 if (fp->f_type != DTYPE_VNODE) {
319 error = EBADF;
320 break;
321 }
322 vp = (struct vnode *)fp->f_data;
323
324 /*
325 * copyin/lockop may block
326 */
327 if (dat->fc_flock.l_whence == SEEK_CUR)
328 dat->fc_flock.l_start += fp->f_offset;
329
330 switch (dat->fc_flock.l_type) {
331 case F_RDLCK:
332 if ((fp->f_flag & FREAD) == 0) {
333 error = EBADF;
334 break;
335 }
336 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) {
337 lwkt_gettoken(&p->p_leader->p_token);
338 p->p_leader->p_flags |= P_ADVLOCK;
339 lwkt_reltoken(&p->p_leader->p_token);
340 }
341 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
342 &dat->fc_flock, flg);
343 break;
344 case F_WRLCK:
345 if ((fp->f_flag & FWRITE) == 0) {
346 error = EBADF;
347 break;
348 }
349 if ((p->p_leader->p_flags & P_ADVLOCK) == 0) {
350 lwkt_gettoken(&p->p_leader->p_token);
351 p->p_leader->p_flags |= P_ADVLOCK;
352 lwkt_reltoken(&p->p_leader->p_token);
353 }
354 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
355 &dat->fc_flock, flg);
356 break;
357 case F_UNLCK:
358 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
359 &dat->fc_flock, F_POSIX);
360 break;
361 default:
362 error = EINVAL;
363 break;
364 }
365
366 /*
367 * It is possible to race a close() on the descriptor while
368 * we were blocked getting the lock. If this occurs the
369 * close might not have caught the lock.
370 */
371 if (checkfdclosed(p->p_fd, fd, fp)) {
372 dat->fc_flock.l_whence = SEEK_SET;
373 dat->fc_flock.l_start = 0;
374 dat->fc_flock.l_len = 0;
375 dat->fc_flock.l_type = F_UNLCK;
376 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
377 F_UNLCK, &dat->fc_flock, F_POSIX);
378 }
379 break;
380
381 case F_GETLK:
382 if (fp->f_type != DTYPE_VNODE) {
383 error = EBADF;
384 break;
385 }
386 vp = (struct vnode *)fp->f_data;
387 /*
388 * copyin/lockop may block
389 */
390 if (dat->fc_flock.l_type != F_RDLCK &&
391 dat->fc_flock.l_type != F_WRLCK &&
392 dat->fc_flock.l_type != F_UNLCK) {
393 error = EINVAL;
394 break;
395 }
396 if (dat->fc_flock.l_whence == SEEK_CUR)
397 dat->fc_flock.l_start += fp->f_offset;
398 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK,
399 &dat->fc_flock, F_POSIX);
400 break;
401 default:
402 error = EINVAL;
403 break;
404 }
405
406 fdrop(fp);
407 return (error);
408 }
409
410 /*
411 * The file control system call.
412 *
413 * MPSAFE
414 */
415 int
416 sys_fcntl(struct fcntl_args *uap)
417 {
418 union fcntl_dat dat;
419 int error;
420
421 switch (uap->cmd) {
422 case F_DUPFD:
423 case F_DUP2FD:
424 case F_DUPFD_CLOEXEC:
425 case F_DUP2FD_CLOEXEC:
426 dat.fc_fd = uap->arg;
427 break;
428 case F_SETFD:
429 dat.fc_cloexec = uap->arg;
430 break;
431 case F_SETFL:
432 dat.fc_flags = uap->arg;
433 break;
434 case F_SETOWN:
435 dat.fc_owner = uap->arg;
436 break;
437 case F_SETLKW:
438 case F_SETLK:
439 case F_GETLK:
440 error = copyin((caddr_t)uap->arg, &dat.fc_flock,
441 sizeof(struct flock));
442 if (error)
443 return (error);
444 break;
445 }
446
447 error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred);
448
449 if (error == 0) {
450 switch (uap->cmd) {
451 case F_DUPFD:
452 case F_DUP2FD:
453 case F_DUPFD_CLOEXEC:
454 case F_DUP2FD_CLOEXEC:
455 uap->sysmsg_result = dat.fc_fd;
456 break;
457 case F_GETFD:
458 uap->sysmsg_result = dat.fc_cloexec;
459 break;
460 case F_GETFL:
461 uap->sysmsg_result = dat.fc_flags;
462 break;
463 case F_GETOWN:
464 uap->sysmsg_result = dat.fc_owner;
465 break;
466 case F_GETLK:
467 error = copyout(&dat.fc_flock, (caddr_t)uap->arg,
468 sizeof(struct flock));
469 break;
470 }
471 }
472
473 return (error);
474 }
475
476 /*
477 * Common code for dup, dup2, and fcntl(F_DUPFD).
478 *
479 * There are three type flags: DUP_FIXED, DUP_VARIABLE, and DUP_CLOEXEC.
480 * The first two flags are mutually exclusive, and the third is optional.
481 * DUP_FIXED tells kern_dup() to destructively dup over an existing file
482 * descriptor if "new" is already open. DUP_VARIABLE tells kern_dup()
483 * to find the lowest unused file descriptor that is greater than or
484 * equal to "new". DUP_CLOEXEC, which works with either of the first
485 * two flags, sets the close-on-exec flag on the "new" file descriptor.
486 *
487 * MPSAFE
488 */
489 int
490 kern_dup(int flags, int old, int new, int *res)
491 {
492 struct thread *td = curthread;
493 struct proc *p = td->td_proc;
494 struct filedesc *fdp = p->p_fd;
495 struct file *fp;
496 struct file *delfp;
497 int oldflags;
498 int holdleaders;
499 int dtsize;
500 int error, newfd;
501
502 /*
503 * Verify that we have a valid descriptor to dup from and
504 * possibly to dup to.
505 *
506 * NOTE: maxfilesperuser is not applicable to dup()
507 */
508 retry:
509 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
510 dtsize = INT_MAX;
511 else
512 dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
513 if (dtsize > maxfilesperproc)
514 dtsize = maxfilesperproc;
515 if (dtsize < minfilesperproc)
516 dtsize = minfilesperproc;
517
518 if (new < 0 || new > dtsize)
519 return (EINVAL);
520
521 spin_lock(&fdp->fd_spin);
522 if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) {
523 spin_unlock(&fdp->fd_spin);
524 return (EBADF);
525 }
526 if ((flags & DUP_FIXED) && old == new) {
527 *res = new;
528 if (flags & DUP_CLOEXEC)
529 fdp->fd_files[new].fileflags |= UF_EXCLOSE;
530 spin_unlock(&fdp->fd_spin);
531 return (0);
532 }
533 fp = fdp->fd_files[old].fp;
534 oldflags = fdp->fd_files[old].fileflags;
535 fhold(fp); /* MPSAFE - can be called with a spinlock held */
536
537 /*
538 * Allocate a new descriptor if DUP_VARIABLE, or expand the table
539 * if the requested descriptor is beyond the current table size.
540 *
541 * This can block. Retry if the source descriptor no longer matches
542 * or if our expectation in the expansion case races.
543 *
544 * If we are not expanding or allocating a new decriptor, then reset
545 * the target descriptor to a reserved state so we have a uniform
546 * setup for the next code block.
547 */
548 if ((flags & DUP_VARIABLE) || new >= fdp->fd_nfiles) {
549 spin_unlock(&fdp->fd_spin);
550 error = fdalloc(p, new, &newfd);
551 spin_lock(&fdp->fd_spin);
552 if (error) {
553 spin_unlock(&fdp->fd_spin);
554 fdrop(fp);
555 return (error);
556 }
557 /*
558 * Check for ripout
559 */
560 if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) {
561 fsetfd_locked(fdp, NULL, newfd);
562 spin_unlock(&fdp->fd_spin);
563 fdrop(fp);
564 goto retry;
565 }
566 /*
567 * Check for expansion race
568 */
569 if ((flags & DUP_VARIABLE) == 0 && new != newfd) {
570 fsetfd_locked(fdp, NULL, newfd);
571 spin_unlock(&fdp->fd_spin);
572 fdrop(fp);
573 goto retry;
574 }
575 /*
576 * Check for ripout, newfd reused old (this case probably
577 * can't occur).
578 */
579 if (old == newfd) {
580 fsetfd_locked(fdp, NULL, newfd);
581 spin_unlock(&fdp->fd_spin);
582 fdrop(fp);
583 goto retry;
584 }
585 new = newfd;
586 delfp = NULL;
587 } else {
588 if (fdp->fd_files[new].reserved) {
589 spin_unlock(&fdp->fd_spin);
590 fdrop(fp);
591 kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new);
592 tsleep(fdp, 0, "fdres", hz);
593 goto retry;
594 }
595
596 /*
597 * If the target descriptor was never allocated we have
598 * to allocate it. If it was we have to clean out the
599 * old descriptor. delfp inherits the ref from the
600 * descriptor table.
601 */
602 delfp = fdp->fd_files[new].fp;
603 fdp->fd_files[new].fp = NULL;
604 fdp->fd_files[new].reserved = 1;
605 if (delfp == NULL) {
606 fdreserve_locked(fdp, new, 1);
607 if (new > fdp->fd_lastfile)
608 fdp->fd_lastfile = new;
609 }
610
611 }
612
613 /*
614 * NOTE: still holding an exclusive spinlock
615 */
616
617 /*
618 * If a descriptor is being overwritten we may hve to tell
619 * fdfree() to sleep to ensure that all relevant process
620 * leaders can be traversed in closef().
621 */
622 if (delfp != NULL && p->p_fdtol != NULL) {
623 fdp->fd_holdleaderscount++;
624 holdleaders = 1;
625 } else {
626 holdleaders = 0;
627 }
628 KASSERT(delfp == NULL || (flags & DUP_FIXED),
629 ("dup() picked an open file"));
630
631 /*
632 * Duplicate the source descriptor, update lastfile. If the new
633 * descriptor was not allocated and we aren't replacing an existing
634 * descriptor we have to mark the descriptor as being in use.
635 *
636 * The fd_files[] array inherits fp's hold reference.
637 */
638 fsetfd_locked(fdp, fp, new);
639 if ((flags & DUP_CLOEXEC) != 0)
640 fdp->fd_files[new].fileflags = oldflags | UF_EXCLOSE;
641 else
642 fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE;
643 spin_unlock(&fdp->fd_spin);
644 fdrop(fp);
645 *res = new;
646
647 /*
648 * If we dup'd over a valid file, we now own the reference to it
649 * and must dispose of it using closef() semantics (as if a
650 * close() were performed on it).
651 */
652 if (delfp) {
653 if (SLIST_FIRST(&delfp->f_klist))
654 knote_fdclose(delfp, fdp, new);
655 closef(delfp, p);
656 if (holdleaders) {
657 spin_lock(&fdp->fd_spin);
658 fdp->fd_holdleaderscount--;
659 if (fdp->fd_holdleaderscount == 0 &&
660 fdp->fd_holdleaderswakeup != 0) {
661 fdp->fd_holdleaderswakeup = 0;
662 spin_unlock(&fdp->fd_spin);
663 wakeup(&fdp->fd_holdleaderscount);
664 } else {
665 spin_unlock(&fdp->fd_spin);
666 }
667 }
668 }
669 return (0);
670 }
671
672 /*
673 * If sigio is on the list associated with a process or process group,
674 * disable signalling from the device, remove sigio from the list and
675 * free sigio.
676 *
677 * MPSAFE
678 */
679 void
680 funsetown(struct sigio **sigiop)
681 {
682 struct pgrp *pgrp;
683 struct proc *p;
684 struct sigio *sigio;
685
686 if ((sigio = *sigiop) != NULL) {
687 lwkt_gettoken(&sigio_token); /* protect sigio */
688 KKASSERT(sigiop == sigio->sio_myref);
689 sigio = *sigiop;
690 *sigiop = NULL;
691 lwkt_reltoken(&sigio_token);
692 }
693 if (sigio == NULL)
694 return;
695
696 if (sigio->sio_pgid < 0) {
697 pgrp = sigio->sio_pgrp;
698 sigio->sio_pgrp = NULL;
699 lwkt_gettoken(&pgrp->pg_token);
700 SLIST_REMOVE(&pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio);
701 lwkt_reltoken(&pgrp->pg_token);
702 pgrel(pgrp);
703 } else /* if ((*sigiop)->sio_pgid > 0) */ {
704 p = sigio->sio_proc;
705 sigio->sio_proc = NULL;
706 PHOLD(p);
707 lwkt_gettoken(&p->p_token);
708 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
709 lwkt_reltoken(&p->p_token);
710 PRELE(p);
711 }
712 crfree(sigio->sio_ucred);
713 sigio->sio_ucred = NULL;
714 kfree(sigio, M_SIGIO);
715 }
716
717 /*
718 * Free a list of sigio structures. Caller is responsible for ensuring
719 * that the list is MPSAFE.
720 *
721 * MPSAFE
722 */
723 void
724 funsetownlst(struct sigiolst *sigiolst)
725 {
726 struct sigio *sigio;
727
728 while ((sigio = SLIST_FIRST(sigiolst)) != NULL)
729 funsetown(sigio->sio_myref);
730 }
731
732 /*
733 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
734 *
735 * After permission checking, add a sigio structure to the sigio list for
736 * the process or process group.
737 *
738 * MPSAFE
739 */
740 int
741 fsetown(pid_t pgid, struct sigio **sigiop)
742 {
743 struct proc *proc = NULL;
744 struct pgrp *pgrp = NULL;
745 struct sigio *sigio;
746 int error;
747
748 if (pgid == 0) {
749 funsetown(sigiop);
750 return (0);
751 }
752
753 if (pgid > 0) {
754 proc = pfind(pgid);
755 if (proc == NULL) {
756 error = ESRCH;
757 goto done;
758 }
759
760 /*
761 * Policy - Don't allow a process to FSETOWN a process
762 * in another session.
763 *
764 * Remove this test to allow maximum flexibility or
765 * restrict FSETOWN to the current process or process
766 * group for maximum safety.
767 */
768 if (proc->p_session != curproc->p_session) {
769 error = EPERM;
770 goto done;
771 }
772 } else /* if (pgid < 0) */ {
773 pgrp = pgfind(-pgid);
774 if (pgrp == NULL) {
775 error = ESRCH;
776 goto done;
777 }
778
779 /*
780 * Policy - Don't allow a process to FSETOWN a process
781 * in another session.
782 *
783 * Remove this test to allow maximum flexibility or
784 * restrict FSETOWN to the current process or process
785 * group for maximum safety.
786 */
787 if (pgrp->pg_session != curproc->p_session) {
788 error = EPERM;
789 goto done;
790 }
791 }
792 sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK | M_ZERO);
793 if (pgid > 0) {
794 KKASSERT(pgrp == NULL);
795 lwkt_gettoken(&proc->p_token);
796 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
797 sigio->sio_proc = proc;
798 lwkt_reltoken(&proc->p_token);
799 } else {
800 KKASSERT(proc == NULL);
801 lwkt_gettoken(&pgrp->pg_token);
802 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
803 sigio->sio_pgrp = pgrp;
804 lwkt_reltoken(&pgrp->pg_token);
805 pgrp = NULL;
806 }
807 sigio->sio_pgid = pgid;
808 sigio->sio_ucred = crhold(curthread->td_ucred);
809 /* It would be convenient if p_ruid was in ucred. */
810 sigio->sio_ruid = sigio->sio_ucred->cr_ruid;
811 sigio->sio_myref = sigiop;
812
813 lwkt_gettoken(&sigio_token);
814 while (*sigiop)
815 funsetown(sigiop);
816 *sigiop = sigio;
817 lwkt_reltoken(&sigio_token);
818 error = 0;
819 done:
820 if (pgrp)
821 pgrel(pgrp);
822 if (proc)
823 PRELE(proc);
824 return (error);
825 }
826
827 /*
828 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
829 *
830 * MPSAFE
831 */
832 pid_t
833 fgetown(struct sigio **sigiop)
834 {
835 struct sigio *sigio;
836 pid_t own;
837
838 lwkt_gettoken_shared(&sigio_token);
839 sigio = *sigiop;
840 own = (sigio != NULL ? sigio->sio_pgid : 0);
841 lwkt_reltoken(&sigio_token);
842
843 return (own);
844 }
845
846 /*
847 * Close many file descriptors.
848 *
849 * MPSAFE
850 */
851 int
852 sys_closefrom(struct closefrom_args *uap)
853 {
854 return(kern_closefrom(uap->fd));
855 }
856
857 /*
858 * Close all file descriptors greater then or equal to fd
859 *
860 * MPSAFE
861 */
862 int
863 kern_closefrom(int fd)
864 {
865 struct thread *td = curthread;
866 struct proc *p = td->td_proc;
867 struct filedesc *fdp;
868
869 KKASSERT(p);
870 fdp = p->p_fd;
871
872 if (fd < 0)
873 return (EINVAL);
874
875 /*
876 * NOTE: This function will skip unassociated descriptors and
877 * reserved descriptors that have not yet been assigned.
878 * fd_lastfile can change as a side effect of kern_close().
879 */
880 spin_lock(&fdp->fd_spin);
881 while (fd <= fdp->fd_lastfile) {
882 if (fdp->fd_files[fd].fp != NULL) {
883 spin_unlock(&fdp->fd_spin);
884 /* ok if this races another close */
885 if (kern_close(fd) == EINTR)
886 return (EINTR);
887 spin_lock(&fdp->fd_spin);
888 }
889 ++fd;
890 }
891 spin_unlock(&fdp->fd_spin);
892 return (0);
893 }
894
895 /*
896 * Close a file descriptor.
897 *
898 * MPSAFE
899 */
900 int
901 sys_close(struct close_args *uap)
902 {
903 return(kern_close(uap->fd));
904 }
905
906 /*
907 * MPSAFE
908 */
909 int
910 kern_close(int fd)
911 {
912 struct thread *td = curthread;
913 struct proc *p = td->td_proc;
914 struct filedesc *fdp;
915 struct file *fp;
916 int error;
917 int holdleaders;
918
919 KKASSERT(p);
920 fdp = p->p_fd;
921
922 spin_lock(&fdp->fd_spin);
923 if ((fp = funsetfd_locked(fdp, fd)) == NULL) {
924 spin_unlock(&fdp->fd_spin);
925 return (EBADF);
926 }
927 holdleaders = 0;
928 if (p->p_fdtol != NULL) {
929 /*
930 * Ask fdfree() to sleep to ensure that all relevant
931 * process leaders can be traversed in closef().
932 */
933 fdp->fd_holdleaderscount++;
934 holdleaders = 1;
935 }
936
937 /*
938 * we now hold the fp reference that used to be owned by the descriptor
939 * array.
940 */
941 spin_unlock(&fdp->fd_spin);
942 if (SLIST_FIRST(&fp->f_klist))
943 knote_fdclose(fp, fdp, fd);
944 error = closef(fp, p);
945 if (holdleaders) {
946 spin_lock(&fdp->fd_spin);
947 fdp->fd_holdleaderscount--;
948 if (fdp->fd_holdleaderscount == 0 &&
949 fdp->fd_holdleaderswakeup != 0) {
950 fdp->fd_holdleaderswakeup = 0;
951 spin_unlock(&fdp->fd_spin);
952 wakeup(&fdp->fd_holdleaderscount);
953 } else {
954 spin_unlock(&fdp->fd_spin);
955 }
956 }
957 return (error);
958 }
959
960 /*
961 * shutdown_args(int fd, int how)
962 */
963 int
964 kern_shutdown(int fd, int how)
965 {
966 struct thread *td = curthread;
967 struct proc *p = td->td_proc;
968 struct file *fp;
969 int error;
970
971 KKASSERT(p);
972
973 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
974 return (EBADF);
975 error = fo_shutdown(fp, how);
976 fdrop(fp);
977
978 return (error);
979 }
980
981 /*
982 * MPALMOSTSAFE
983 */
984 int
985 sys_shutdown(struct shutdown_args *uap)
986 {
987 int error;
988
989 error = kern_shutdown(uap->s, uap->how);
990
991 return (error);
992 }
993
994 /*
995 * MPSAFE
996 */
997 int
998 kern_fstat(int fd, struct stat *ub)
999 {
1000 struct thread *td = curthread;
1001 struct proc *p = td->td_proc;
1002 struct file *fp;
1003 int error;
1004
1005 KKASSERT(p);
1006
1007 if ((fp = holdfp(p->p_fd, fd, -1)) == NULL)
1008 return (EBADF);
1009 error = fo_stat(fp, ub, td->td_ucred);
1010 fdrop(fp);
1011
1012 return (error);
1013 }
1014
1015 /*
1016 * Return status information about a file descriptor.
1017 *
1018 * MPSAFE
1019 */
1020 int
1021 sys_fstat(struct fstat_args *uap)
1022 {
1023 struct stat st;
1024 int error;
1025
1026 error = kern_fstat(uap->fd, &st);
1027
1028 if (error == 0)
1029 error = copyout(&st, uap->sb, sizeof(st));
1030 return (error);
1031 }
1032
1033 /*
1034 * Return pathconf information about a file descriptor.
1035 *
1036 * MPALMOSTSAFE
1037 */
1038 int
1039 sys_fpathconf(struct fpathconf_args *uap)
1040 {
1041 struct thread *td = curthread;
1042 struct proc *p = td->td_proc;
1043 struct file *fp;
1044 struct vnode *vp;
1045 int error = 0;
1046
1047 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
1048 return (EBADF);
1049
1050 switch (fp->f_type) {
1051 case DTYPE_PIPE:
1052 case DTYPE_SOCKET:
1053 if (uap->name != _PC_PIPE_BUF) {
1054 error = EINVAL;
1055 } else {
1056 uap->sysmsg_result = PIPE_BUF;
1057 error = 0;
1058 }
1059 break;
1060 case DTYPE_FIFO:
1061 case DTYPE_VNODE:
1062 vp = (struct vnode *)fp->f_data;
1063 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
1064 break;
1065 default:
1066 error = EOPNOTSUPP;
1067 break;
1068 }
1069 fdrop(fp);
1070 return(error);
1071 }
1072
1073 static int fdexpand;
1074 SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0,
1075 "Number of times a file table has been expanded");
1076
1077 /*
1078 * Grow the file table so it can hold through descriptor (want).
1079 *
1080 * The fdp's spinlock must be held exclusively on entry and may be held
1081 * exclusively on return. The spinlock may be cycled by the routine.
1082 *
1083 * MPSAFE
1084 */
1085 static void
1086 fdgrow_locked(struct filedesc *fdp, int want)
1087 {
1088 struct fdnode *newfiles;
1089 struct fdnode *oldfiles;
1090 int nf, extra;
1091
1092 nf = fdp->fd_nfiles;
1093 do {
1094 /* nf has to be of the form 2^n - 1 */
1095 nf = 2 * nf + 1;
1096 } while (nf <= want);
1097
1098 spin_unlock(&fdp->fd_spin);
1099 newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK);
1100 spin_lock(&fdp->fd_spin);
1101
1102 /*
1103 * We could have raced another extend while we were not holding
1104 * the spinlock.
1105 */
1106 if (fdp->fd_nfiles >= nf) {
1107 spin_unlock(&fdp->fd_spin);
1108 kfree(newfiles, M_FILEDESC);
1109 spin_lock(&fdp->fd_spin);
1110 return;
1111 }
1112 /*
1113 * Copy the existing ofile and ofileflags arrays
1114 * and zero the new portion of each array.
1115 */
1116 extra = nf - fdp->fd_nfiles;
1117 bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode));
1118 bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode));
1119
1120 oldfiles = fdp->fd_files;
1121 fdp->fd_files = newfiles;
1122 fdp->fd_nfiles = nf;
1123
1124 if (oldfiles != fdp->fd_builtin_files) {
1125 spin_unlock(&fdp->fd_spin);
1126 kfree(oldfiles, M_FILEDESC);
1127 spin_lock(&fdp->fd_spin);
1128 }
1129 fdexpand++;
1130 }
1131
1132 /*
1133 * Number of nodes in right subtree, including the root.
1134 */
1135 static __inline int
1136 right_subtree_size(int n)
1137 {
1138 return (n ^ (n | (n + 1)));
1139 }
1140
1141 /*
1142 * Bigger ancestor.
1143 */
1144 static __inline int
1145 right_ancestor(int n)
1146 {
1147 return (n | (n + 1));
1148 }
1149
1150 /*
1151 * Smaller ancestor.
1152 */
1153 static __inline int
1154 left_ancestor(int n)
1155 {
1156 return ((n & (n + 1)) - 1);
1157 }
1158
1159 /*
1160 * Traverse the in-place binary tree buttom-up adjusting the allocation
1161 * count so scans can determine where free descriptors are located.
1162 *
1163 * MPSAFE - caller must be holding an exclusive spinlock on fdp
1164 */
1165 static
1166 void
1167 fdreserve_locked(struct filedesc *fdp, int fd, int incr)
1168 {
1169 while (fd >= 0) {
1170 fdp->fd_files[fd].allocated += incr;
1171 KKASSERT(fdp->fd_files[fd].allocated >= 0);
1172 fd = left_ancestor(fd);
1173 }
1174 }
1175
1176 /*
1177 * Reserve a file descriptor for the process. If no error occurs, the
1178 * caller MUST at some point call fsetfd() or assign a file pointer
1179 * or dispose of the reservation.
1180 *
1181 * MPSAFE
1182 */
1183 int
1184 fdalloc(struct proc *p, int want, int *result)
1185 {
1186 struct filedesc *fdp = p->p_fd;
1187 struct uidinfo *uip;
1188 int fd, rsize, rsum, node, lim;
1189
1190 /*
1191 * Check dtable size limit
1192 */
1193 spin_lock(&p->p_limit->p_spin);
1194 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1195 lim = INT_MAX;
1196 else
1197 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1198 spin_unlock(&p->p_limit->p_spin);
1199
1200 if (lim > maxfilesperproc)
1201 lim = maxfilesperproc;
1202 if (lim < minfilesperproc)
1203 lim = minfilesperproc;
1204 if (want >= lim)
1205 return (EMFILE);
1206
1207 /*
1208 * Check that the user has not run out of descriptors (non-root only).
1209 * As a safety measure the dtable is allowed to have at least
1210 * minfilesperproc open fds regardless of the maxfilesperuser limit.
1211 */
1212 if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) {
1213 uip = p->p_ucred->cr_uidinfo;
1214 if (uip->ui_openfiles > maxfilesperuser) {
1215 krateprintf(&krate_uidinfo,
1216 "Warning: user %d pid %d (%s) ran out of "
1217 "file descriptors (%d/%d)\n",
1218 p->p_ucred->cr_uid, (int)p->p_pid,
1219 p->p_comm,
1220 uip->ui_openfiles, maxfilesperuser);
1221 return(ENFILE);
1222 }
1223 }
1224
1225 /*
1226 * Grow the dtable if necessary
1227 */
1228 spin_lock(&fdp->fd_spin);
1229 if (want >= fdp->fd_nfiles)
1230 fdgrow_locked(fdp, want);
1231
1232 /*
1233 * Search for a free descriptor starting at the higher
1234 * of want or fd_freefile. If that fails, consider
1235 * expanding the ofile array.
1236 *
1237 * NOTE! the 'allocated' field is a cumulative recursive allocation
1238 * count. If we happen to see a value of 0 then we can shortcut
1239 * our search. Otherwise we run through through the tree going
1240 * down branches we know have free descriptor(s) until we hit a
1241 * leaf node. The leaf node will be free but will not necessarily
1242 * have an allocated field of 0.
1243 */
1244 retry:
1245 /* move up the tree looking for a subtree with a free node */
1246 for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim);
1247 fd = right_ancestor(fd)) {
1248 if (fdp->fd_files[fd].allocated == 0)
1249 goto found;
1250
1251 rsize = right_subtree_size(fd);
1252 if (fdp->fd_files[fd].allocated == rsize)
1253 continue; /* right subtree full */
1254
1255 /*
1256 * Free fd is in the right subtree of the tree rooted at fd.
1257 * Call that subtree R. Look for the smallest (leftmost)
1258 * subtree of R with an unallocated fd: continue moving
1259 * down the left branch until encountering a full left
1260 * subtree, then move to the right.
1261 */
1262 for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) {
1263 node = fd + rsize;
1264 rsum += fdp->fd_files[node].allocated;
1265 if (fdp->fd_files[fd].allocated == rsum + rsize) {
1266 fd = node; /* move to the right */
1267 if (fdp->fd_files[node].allocated == 0)
1268 goto found;
1269 rsum = 0;
1270 }
1271 }
1272 goto found;
1273 }
1274
1275 /*
1276 * No space in current array. Expand?
1277 */
1278 if (fdp->fd_nfiles >= lim) {
1279 spin_unlock(&fdp->fd_spin);
1280 return (EMFILE);
1281 }
1282 fdgrow_locked(fdp, want);
1283 goto retry;
1284
1285 found:
1286 KKASSERT(fd < fdp->fd_nfiles);
1287 if (fd > fdp->fd_lastfile)
1288 fdp->fd_lastfile = fd;
1289 if (want <= fdp->fd_freefile)
1290 fdp->fd_freefile = fd;
1291 *result = fd;
1292 KKASSERT(fdp->fd_files[fd].fp == NULL);
1293 KKASSERT(fdp->fd_files[fd].reserved == 0);
1294 fdp->fd_files[fd].fileflags = 0;
1295 fdp->fd_files[fd].reserved = 1;
1296 fdreserve_locked(fdp, fd, 1);
1297 spin_unlock(&fdp->fd_spin);
1298 return (0);
1299 }
1300
1301 /*
1302 * Check to see whether n user file descriptors
1303 * are available to the process p.
1304 *
1305 * MPSAFE
1306 */
1307 int
1308 fdavail(struct proc *p, int n)
1309 {
1310 struct filedesc *fdp = p->p_fd;
1311 struct fdnode *fdnode;
1312 int i, lim, last;
1313
1314 spin_lock(&p->p_limit->p_spin);
1315 if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX)
1316 lim = INT_MAX;
1317 else
1318 lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur;
1319 spin_unlock(&p->p_limit->p_spin);
1320
1321 if (lim > maxfilesperproc)
1322 lim = maxfilesperproc;
1323 if (lim < minfilesperproc)
1324 lim = minfilesperproc;
1325
1326 spin_lock(&fdp->fd_spin);
1327 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
1328 spin_unlock(&fdp->fd_spin);
1329 return (1);
1330 }
1331 last = min(fdp->fd_nfiles, lim);
1332 fdnode = &fdp->fd_files[fdp->fd_freefile];
1333 for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) {
1334 if (fdnode->fp == NULL && --n <= 0) {
1335 spin_unlock(&fdp->fd_spin);
1336 return (1);
1337 }
1338 }
1339 spin_unlock(&fdp->fd_spin);
1340 return (0);
1341 }
1342
1343 /*
1344 * Revoke open descriptors referencing (f_data, f_type)
1345 *
1346 * Any revoke executed within a prison is only able to
1347 * revoke descriptors for processes within that prison.
1348 *
1349 * Returns 0 on success or an error code.
1350 */
1351 struct fdrevoke_info {
1352 void *data;
1353 short type;
1354 short unused;
1355 int count;
1356 int intransit;
1357 struct ucred *cred;
1358 struct file *nfp;
1359 };
1360
1361 static int fdrevoke_check_callback(struct file *fp, void *vinfo);
1362 static int fdrevoke_proc_callback(struct proc *p, void *vinfo);
1363
1364 int
1365 fdrevoke(void *f_data, short f_type, struct ucred *cred)
1366 {
1367 struct fdrevoke_info info;
1368 int error;
1369
1370 bzero(&info, sizeof(info));
1371 info.data = f_data;
1372 info.type = f_type;
1373 info.cred = cred;
1374 error = falloc(NULL, &info.nfp, NULL);
1375 if (error)
1376 return (error);
1377
1378 /*
1379 * Scan the file pointer table once. dups do not dup file pointers,
1380 * only descriptors, so there is no leak. Set FREVOKED on the fps
1381 * being revoked.
1382 */
1383 allfiles_scan_exclusive(fdrevoke_check_callback, &info);
1384
1385 /*
1386 * If any fps were marked track down the related descriptors
1387 * and close them. Any dup()s at this point will notice
1388 * the FREVOKED already set in the fp and do the right thing.
1389 *
1390 * Any fps with non-zero msgcounts (aka sent over a unix-domain
1391 * socket) bumped the intransit counter and will require a
1392 * scan. Races against fps leaving the socket are closed by
1393 * the socket code checking for FREVOKED.
1394 */
1395 if (info.count)
1396 allproc_scan(fdrevoke_proc_callback, &info);
1397 if (info.intransit)
1398 unp_revoke_gc(info.nfp);
1399 fdrop(info.nfp);
1400 return(0);
1401 }
1402
1403 /*
1404 * Locate matching file pointers directly.
1405 *
1406 * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls!
1407 */
1408 static int
1409 fdrevoke_check_callback(struct file *fp, void *vinfo)
1410 {
1411 struct fdrevoke_info *info = vinfo;
1412
1413 /*
1414 * File pointers already flagged for revokation are skipped.
1415 */
1416 if (fp->f_flag & FREVOKED)
1417 return(0);
1418
1419 /*
1420 * If revoking from a prison file pointers created outside of
1421 * that prison, or file pointers without creds, cannot be revoked.
1422 */
1423 if (info->cred->cr_prison &&
1424 (fp->f_cred == NULL ||
1425 info->cred->cr_prison != fp->f_cred->cr_prison)) {
1426 return(0);
1427 }
1428
1429 /*
1430 * If the file pointer matches then mark it for revocation. The
1431 * flag is currently only used by unp_revoke_gc().
1432 *
1433 * info->count is a heuristic and can race in a SMP environment.
1434 */
1435 if (info->data == fp->f_data && info->type == fp->f_type) {
1436 atomic_set_int(&fp->f_flag, FREVOKED);
1437 info->count += fp->f_count;
1438 if (fp->f_msgcount)
1439 ++info->intransit;
1440 }
1441 return(0);
1442 }
1443
1444 /*
1445 * Locate matching file pointers via process descriptor tables.
1446 */
1447 static int
1448 fdrevoke_proc_callback(struct proc *p, void *vinfo)
1449 {
1450 struct fdrevoke_info *info = vinfo;
1451 struct filedesc *fdp;
1452 struct file *fp;
1453 int n;
1454
1455 if (p->p_stat == SIDL || p->p_stat == SZOMB)
1456 return(0);
1457 if (info->cred->cr_prison &&
1458 info->cred->cr_prison != p->p_ucred->cr_prison) {
1459 return(0);
1460 }
1461
1462 /*
1463 * If the controlling terminal of the process matches the
1464 * vnode being revoked we clear the controlling terminal.
1465 *
1466 * The normal spec_close() may not catch this because it
1467 * uses curproc instead of p.
1468 */
1469 if (p->p_session && info->type == DTYPE_VNODE &&
1470 info->data == p->p_session->s_ttyvp) {
1471 p->p_session->s_ttyvp = NULL;
1472 vrele(info->data);
1473 }
1474
1475 /*
1476 * Softref the fdp to prevent it from being destroyed
1477 */
1478 spin_lock(&p->p_spin);
1479 if ((fdp = p->p_fd) == NULL) {
1480 spin_unlock(&p->p_spin);
1481 return(0);
1482 }
1483 atomic_add_int(&fdp->fd_softrefs, 1);
1484 spin_unlock(&p->p_spin);
1485
1486 /*
1487 * Locate and close any matching file descriptors.
1488 */
1489 spin_lock(&fdp->fd_spin);
1490 for (n = 0; n < fdp->fd_nfiles; ++n) {
1491 if ((fp = fdp->fd_files[n].fp) == NULL)
1492 continue;
1493 if (fp->f_flag & FREVOKED) {
1494 fhold(info->nfp);
1495 fdp->fd_files[n].fp = info->nfp;
1496 spin_unlock(&fdp->fd_spin);
1497 knote_fdclose(fp, fdp, n); /* XXX */
1498 closef(fp, p);
1499 spin_lock(&fdp->fd_spin);
1500 --info->count;
1501 }
1502 }
1503 spin_unlock(&fdp->fd_spin);
1504 atomic_subtract_int(&fdp->fd_softrefs, 1);
1505 return(0);
1506 }
1507
1508 /*
1509 * falloc:
1510 * Create a new open file structure and reserve a file decriptor
1511 * for the process that refers to it.
1512 *
1513 * Root creds are checked using lp, or assumed if lp is NULL. If
1514 * resultfd is non-NULL then lp must also be non-NULL. No file
1515 * descriptor is reserved (and no process context is needed) if
1516 * resultfd is NULL.
1517 *
1518 * A file pointer with a refcount of 1 is returned. Note that the
1519 * file pointer is NOT associated with the descriptor. If falloc
1520 * returns success, fsetfd() MUST be called to either associate the
1521 * file pointer or clear the reservation.
1522 *
1523 * MPSAFE
1524 */
1525 int
1526 falloc(struct lwp *lp, struct file **resultfp, int *resultfd)
1527 {
1528 static struct timeval lastfail;
1529 static int curfail;
1530 struct file *fp;
1531 struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred;
1532 int error;
1533
1534 fp = NULL;
1535
1536 /*
1537 * Handle filetable full issues and root overfill.
1538 */
1539 if (nfiles >= maxfiles - maxfilesrootres &&
1540 (cred->cr_ruid != 0 || nfiles >= maxfiles)) {
1541 if (ppsratecheck(&lastfail, &curfail, 1)) {
1542 kprintf("kern.maxfiles limit exceeded by uid %d, "
1543 "please see tuning(7).\n",
1544 cred->cr_ruid);
1545 }
1546 error = ENFILE;
1547 goto done;
1548 }
1549
1550 /*
1551 * Allocate a new file descriptor.
1552 */
1553 fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO);
1554 spin_init(&fp->f_spin);
1555 SLIST_INIT(&fp->f_klist);
1556 fp->f_count = 1;
1557 fp->f_ops = &badfileops;
1558 fp->f_seqcount = 1;
1559 fsetcred(fp, cred);
1560 spin_lock(&filehead_spin);
1561 nfiles++;
1562 LIST_INSERT_HEAD(&filehead, fp, f_list);
1563 spin_unlock(&filehead_spin);
1564 if (resultfd) {
1565 if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) {
1566 fdrop(fp);
1567 fp = NULL;
1568 }
1569 } else {
1570 error = 0;
1571 }
1572 done:
1573 *resultfp = fp;
1574 return (error);
1575 }
1576
1577 /*
1578 * Check for races against a file descriptor by determining that the
1579 * file pointer is still associated with the specified file descriptor,
1580 * and a close is not currently in progress.
1581 *
1582 * MPSAFE
1583 */
1584 int
1585 checkfdclosed(struct filedesc *fdp, int fd, struct file *fp)
1586 {
1587 int error;
1588
1589 spin_lock_shared(&fdp->fd_spin);
1590 if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp)
1591 error = EBADF;
1592 else
1593 error = 0;
1594 spin_unlock_shared(&fdp->fd_spin);
1595 return (error);
1596 }
1597
1598 /*
1599 * Associate a file pointer with a previously reserved file descriptor.
1600 * This function always succeeds.
1601 *
1602 * If fp is NULL, the file descriptor is returned to the pool.
1603 */
1604
1605 /*
1606 * MPSAFE (exclusive spinlock must be held on call)
1607 */
1608 static void
1609 fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd)
1610 {
1611 KKASSERT((unsigned)fd < fdp->fd_nfiles);
1612 KKASSERT(fdp->fd_files[fd].reserved != 0);
1613 if (fp) {
1614 fhold(fp);
1615 fdp->fd_files[fd].fp = fp;
1616 fdp->fd_files[fd].reserved = 0;
1617 } else {
1618 fdp->fd_files[fd].reserved = 0;
1619 fdreserve_locked(fdp, fd, -1);
1620 fdfixup_locked(fdp, fd);
1621 }
1622 }
1623
1624 /*
1625 * MPSAFE
1626 */
1627 void
1628 fsetfd(struct filedesc *fdp, struct file *fp, int fd)
1629 {
1630 spin_lock(&fdp->fd_spin);
1631 fsetfd_locked(fdp, fp, fd);
1632 spin_unlock(&fdp->fd_spin);
1633 }
1634
1635 /*
1636 * MPSAFE (exclusive spinlock must be held on call)
1637 */
1638 static
1639 struct file *
1640 funsetfd_locked(struct filedesc *fdp, int fd)
1641 {
1642 struct file *fp;
1643
1644 if ((unsigned)fd >= fdp->fd_nfiles)
1645 return (NULL);
1646 if ((fp = fdp->fd_files[fd].fp) == NULL)
1647 return (NULL);
1648 fdp->fd_files[fd].fp = NULL;
1649 fdp->fd_files[fd].fileflags = 0;
1650
1651 fdreserve_locked(fdp, fd, -1);
1652 fdfixup_locked(fdp, fd);
1653 return(fp);
1654 }
1655
1656 /*
1657 * MPSAFE
1658 */
1659 int
1660 fgetfdflags(struct filedesc *fdp, int fd, int *flagsp)
1661 {
1662 int error;
1663
1664 spin_lock(&fdp->fd_spin);
1665 if (((u_int)fd) >= fdp->fd_nfiles) {
1666 error = EBADF;
1667 } else if (fdp->fd_files[fd].fp == NULL) {
1668 error = EBADF;
1669 } else {
1670 *flagsp = fdp->fd_files[fd].fileflags;
1671 error = 0;
1672 }
1673 spin_unlock(&fdp->fd_spin);
1674 return (error);
1675 }
1676
1677 /*
1678 * MPSAFE
1679 */
1680 int
1681 fsetfdflags(struct filedesc *fdp, int fd, int add_flags)
1682 {
1683 int error;
1684
1685 spin_lock(&fdp->fd_spin);
1686 if (((u_int)fd) >= fdp->fd_nfiles) {
1687 error = EBADF;
1688 } else if (fdp->fd_files[fd].fp == NULL) {
1689 error = EBADF;
1690 } else {
1691 fdp->fd_files[fd].fileflags |= add_flags;
1692 error = 0;
1693 }
1694 spin_unlock(&fdp->fd_spin);
1695 return (error);
1696 }
1697
1698 /*
1699 * MPSAFE
1700 */
1701 int
1702 fclrfdflags(struct filedesc *fdp, int fd, int rem_flags)
1703 {
1704 int error;
1705
1706 spin_lock(&fdp->fd_spin);
1707 if (((u_int)fd) >= fdp->fd_nfiles) {
1708 error = EBADF;
1709 } else if (fdp->fd_files[fd].fp == NULL) {
1710 error = EBADF;
1711 } else {
1712 fdp->fd_files[fd].fileflags &= ~rem_flags;
1713 error = 0;
1714 }
1715 spin_unlock(&fdp->fd_spin);
1716 return (error);
1717 }
1718
1719 /*
1720 * Set/Change/Clear the creds for a fp and synchronize the uidinfo.
1721 */
1722 void
1723 fsetcred(struct file *fp, struct ucred *ncr)
1724 {
1725 struct ucred *ocr;
1726 struct uidinfo *uip;
1727
1728 ocr = fp->f_cred;
1729 if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) {
1730 if (ocr) {
1731 uip = ocr->cr_uidinfo;
1732 atomic_add_int(&uip->ui_openfiles, -1);
1733 }
1734 if (ncr) {
1735 uip = ncr->cr_uidinfo;
1736 atomic_add_int(&uip->ui_openfiles, 1);
1737 }
1738 }
1739 if (ncr)
1740 crhold(ncr);
1741 fp->f_cred = ncr;
1742 if (ocr)
1743 crfree(ocr);
1744 }
1745
1746 /*
1747 * Free a file descriptor.
1748 */
1749 static
1750 void
1751 ffree(struct file *fp)
1752 {
1753 KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!"));
1754 spin_lock(&filehead_spin);
1755 LIST_REMOVE(fp, f_list);
1756 nfiles--;
1757 spin_unlock(&filehead_spin);
1758 fsetcred(fp, NULL);
1759 if (fp->f_nchandle.ncp)
1760 cache_drop(&fp->f_nchandle);
1761 kfree(fp, M_FILE);
1762 }
1763
1764 /*
1765 * called from init_main, initialize filedesc0 for proc0.
1766 */
1767 void
1768 fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask)
1769 {
1770 p0->p_fd = fdp0;
1771 p0->p_fdtol = NULL;
1772 fdp0->fd_refcnt = 1;
1773 fdp0->fd_cmask = cmask;
1774 fdp0->fd_files = fdp0->fd_builtin_files;
1775 fdp0->fd_nfiles = NDFILE;
1776 fdp0->fd_lastfile = -1;
1777 spin_init(&fdp0->fd_spin);
1778 }
1779
1780 /*
1781 * Build a new filedesc structure.
1782 *
1783 * NOT MPSAFE (vref)
1784 */
1785 struct filedesc *
1786 fdinit(struct proc *p)
1787 {
1788 struct filedesc *newfdp;
1789 struct filedesc *fdp = p->p_fd;
1790
1791 newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO);
1792 spin_lock(&fdp->fd_spin);
1793 if (fdp->fd_cdir) {
1794 newfdp->fd_cdir = fdp->fd_cdir;
1795 vref(newfdp->fd_cdir);
1796 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1797 }
1798
1799 /*
1800 * rdir may not be set in e.g. proc0 or anything vm_fork'd off of
1801 * proc0, but should unconditionally exist in other processes.
1802 */
1803 if (fdp->fd_rdir) {
1804 newfdp->fd_rdir = fdp->fd_rdir;
1805 vref(newfdp->fd_rdir);
1806 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1807 }
1808 if (fdp->fd_jdir) {
1809 newfdp->fd_jdir = fdp->fd_jdir;
1810 vref(newfdp->fd_jdir);
1811 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1812 }
1813 spin_unlock(&fdp->fd_spin);
1814
1815 /* Create the file descriptor table. */
1816 newfdp->fd_refcnt = 1;
1817 newfdp->fd_cmask = cmask;
1818 newfdp->fd_files = newfdp->fd_builtin_files;
1819 newfdp->fd_nfiles = NDFILE;
1820 newfdp->fd_lastfile = -1;
1821 spin_init(&newfdp->fd_spin);
1822
1823 return (newfdp);
1824 }
1825
1826 /*
1827 * Share a filedesc structure.
1828 *
1829 * MPSAFE
1830 */
1831 struct filedesc *
1832 fdshare(struct proc *p)
1833 {
1834 struct filedesc *fdp;
1835
1836 fdp = p->p_fd;
1837 spin_lock(&fdp->fd_spin);
1838 fdp->fd_refcnt++;
1839 spin_unlock(&fdp->fd_spin);
1840 return (fdp);
1841 }
1842
1843 /*
1844 * Copy a filedesc structure.
1845 *
1846 * MPSAFE
1847 */
1848 int
1849 fdcopy(struct proc *p, struct filedesc **fpp)
1850 {
1851 struct filedesc *fdp = p->p_fd;
1852 struct filedesc *newfdp;
1853 struct fdnode *fdnode;
1854 int i;
1855 int ni;
1856
1857 /*
1858 * Certain daemons might not have file descriptors.
1859 */
1860 if (fdp == NULL)
1861 return (0);
1862
1863 /*
1864 * Allocate the new filedesc and fd_files[] array. This can race
1865 * with operations by other threads on the fdp so we have to be
1866 * careful.
1867 */
1868 newfdp = kmalloc(sizeof(struct filedesc),
1869 M_FILEDESC, M_WAITOK | M_ZERO | M_NULLOK);
1870 if (newfdp == NULL) {
1871 *fpp = NULL;
1872 return (-1);
1873 }
1874 again:
1875 spin_lock(&fdp->fd_spin);
1876 if (fdp->fd_lastfile < NDFILE) {
1877 newfdp->fd_files = newfdp->fd_builtin_files;
1878 i = NDFILE;
1879 } else {
1880 /*
1881 * We have to allocate (N^2-1) entries for our in-place
1882 * binary tree. Allow the table to shrink.
1883 */
1884 i = fdp->fd_nfiles;
1885 ni = (i - 1) / 2;
1886 while (ni > fdp->fd_lastfile && ni > NDFILE) {
1887 i = ni;
1888 ni = (i - 1) / 2;
1889 }
1890 spin_unlock(&fdp->fd_spin);
1891 newfdp->fd_files = kmalloc(i * sizeof(struct fdnode),
1892 M_FILEDESC, M_WAITOK | M_ZERO);
1893
1894 /*
1895 * Check for race, retry
1896 */
1897 spin_lock(&fdp->fd_spin);
1898 if (i <= fdp->fd_lastfile) {
1899 spin_unlock(&fdp->fd_spin);
1900 kfree(newfdp->fd_files, M_FILEDESC);
1901 goto again;
1902 }
1903 }
1904
1905 /*
1906 * Dup the remaining fields. vref() and cache_hold() can be
1907 * safely called while holding the read spinlock on fdp.
1908 *
1909 * The read spinlock on fdp is still being held.
1910 *
1911 * NOTE: vref and cache_hold calls for the case where the vnode
1912 * or cache entry already has at least one ref may be called
1913 * while holding spin locks.
1914 */
1915 if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) {
1916 vref(newfdp->fd_cdir);
1917 cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir);
1918 }
1919 /*
1920 * We must check for fd_rdir here, at least for now because
1921 * the init process is created before we have access to the
1922 * rootvode to take a reference to it.
1923 */
1924 if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) {
1925 vref(newfdp->fd_rdir);
1926 cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir);
1927 }
1928 if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) {
1929 vref(newfdp->fd_jdir);
1930 cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir);
1931 }
1932 newfdp->fd_refcnt = 1;
1933 newfdp->fd_nfiles = i;
1934 newfdp->fd_lastfile = fdp->fd_lastfile;
1935 newfdp->fd_freefile = fdp->fd_freefile;
1936 newfdp->fd_cmask = fdp->fd_cmask;
1937 spin_init(&newfdp->fd_spin);
1938
1939 /*
1940 * Copy the descriptor table through (i). This also copies the
1941 * allocation state. Then go through and ref the file pointers
1942 * and clean up any KQ descriptors.
1943 *
1944 * kq descriptors cannot be copied. Since we haven't ref'd the
1945 * copied files yet we can ignore the return value from funsetfd().
1946 *
1947 * The read spinlock on fdp is still being held.
1948 */
1949 bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode));
1950 for (i = 0 ; i < newfdp->fd_nfiles; ++i) {
1951 fdnode = &newfdp->fd_files[i];
1952 if (fdnode->reserved) {
1953 fdreserve_locked(newfdp, i, -1);
1954 fdnode->reserved = 0;
1955 fdfixup_locked(newfdp, i);
1956 } else if (fdnode->fp) {
1957 if (fdnode->fp->f_type == DTYPE_KQUEUE) {
1958 (void)funsetfd_locked(newfdp, i);
1959 } else {
1960 fhold(fdnode->fp);
1961 }
1962 }
1963 }
1964 spin_unlock(&fdp->fd_spin);
1965 *fpp = newfdp;
1966 return (0);
1967 }
1968
1969 /*
1970 * Release a filedesc structure.
1971 *
1972 * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE)
1973 */
1974 void
1975 fdfree(struct proc *p, struct filedesc *repl)
1976 {
1977 struct filedesc *fdp;
1978 struct fdnode *fdnode;
1979 int i;
1980 struct filedesc_to_leader *fdtol;
1981 struct file *fp;
1982 struct vnode *vp;
1983 struct flock lf;
1984
1985 /*
1986 * Certain daemons might not have file descriptors.
1987 */
1988 fdp = p->p_fd;
1989 if (fdp == NULL) {
1990 p->p_fd = repl;
1991 return;
1992 }
1993
1994 /*
1995 * Severe messing around to follow.
1996 */
1997 spin_lock(&fdp->fd_spin);
1998
1999 /* Check for special need to clear POSIX style locks */
2000 fdtol = p->p_fdtol;
2001 if (fdtol != NULL) {
2002 KASSERT(fdtol->fdl_refcount > 0,
2003 ("filedesc_to_refcount botch: fdl_refcount=%d",
2004 fdtol->fdl_refcount));
2005 if (fdtol->fdl_refcount == 1 &&
2006 (p->p_leader->p_flags & P_ADVLOCK) != 0) {
2007 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2008 fdnode = &fdp->fd_files[i];
2009 if (fdnode->fp == NULL ||
2010 fdnode->fp->f_type != DTYPE_VNODE) {
2011 continue;
2012 }
2013 fp = fdnode->fp;
2014 fhold(fp);
2015 spin_unlock(&fdp->fd_spin);
2016
2017 lf.l_whence = SEEK_SET;
2018 lf.l_start = 0;
2019 lf.l_len = 0;
2020 lf.l_type = F_UNLCK;
2021 vp = (struct vnode *)fp->f_data;
2022 (void) VOP_ADVLOCK(vp,
2023 (caddr_t)p->p_leader,
2024 F_UNLCK,
2025 &lf,
2026 F_POSIX);
2027 fdrop(fp);
2028 spin_lock(&fdp->fd_spin);
2029 }
2030 }
2031 retry:
2032 if (fdtol->fdl_refcount == 1) {
2033 if (fdp->fd_holdleaderscount > 0 &&
2034 (p->p_leader->p_flags & P_ADVLOCK) != 0) {
2035 /*
2036 * close() or do_dup() has cleared a reference
2037 * in a shared file descriptor table.
2038 */
2039 fdp->fd_holdleaderswakeup = 1;
2040 ssleep(&fdp->fd_holdleaderscount,
2041 &fdp->fd_spin, 0, "fdlhold", 0);
2042 goto retry;
2043 }
2044 if (fdtol->fdl_holdcount > 0) {
2045 /*
2046 * Ensure that fdtol->fdl_leader
2047 * remains valid in closef().
2048 */
2049 fdtol->fdl_wakeup = 1;
2050 ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0);
2051 goto retry;
2052 }
2053 }
2054 fdtol->fdl_refcount--;
2055 if (fdtol->fdl_refcount == 0 &&
2056 fdtol->fdl_holdcount == 0) {
2057 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2058 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2059 } else {
2060 fdtol = NULL;
2061 }
2062 p->p_fdtol = NULL;
2063 if (fdtol != NULL) {
2064 spin_unlock(&fdp->fd_spin);
2065 kfree(fdtol, M_FILEDESC_TO_LEADER);
2066 spin_lock(&fdp->fd_spin);
2067 }
2068 }
2069 if (--fdp->fd_refcnt > 0) {
2070 spin_unlock(&fdp->fd_spin);
2071 spin_lock(&p->p_spin);
2072 p->p_fd = repl;
2073 spin_unlock(&p->p_spin);
2074 return;
2075 }
2076
2077 /*
2078 * Even though we are the last reference to the structure allproc
2079 * scans may still reference the structure. Maintain proper
2080 * locks until we can replace p->p_fd.
2081 *
2082 * Also note that kqueue's closef still needs to reference the
2083 * fdp via p->p_fd, so we have to close the descriptors before
2084 * we replace p->p_fd.
2085 */
2086 for (i = 0; i <= fdp->fd_lastfile; ++i) {
2087 if (fdp->fd_files[i].fp) {
2088 fp = funsetfd_locked(fdp, i);
2089 if (fp) {
2090 spin_unlock(&fdp->fd_spin);
2091 if (SLIST_FIRST(&fp->f_klist))
2092 knote_fdclose(fp, fdp, i);
2093 closef(fp, p);
2094 spin_lock(&fdp->fd_spin);
2095 }
2096 }
2097 }
2098 spin_unlock(&fdp->fd_spin);
2099
2100 /*
2101 * Interlock against an allproc scan operations (typically frevoke).
2102 */
2103 spin_lock(&p->p_spin);
2104 p->p_fd = repl;
2105 spin_unlock(&p->p_spin);
2106
2107 /*
2108 * Wait for any softrefs to go away. This race rarely occurs so
2109 * we can use a non-critical-path style poll/sleep loop. The
2110 * race only occurs against allproc scans.
2111 *
2112 * No new softrefs can occur with the fdp disconnected from the
2113 * process.
2114 */
2115 if (fdp->fd_softrefs) {
2116 kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid);
2117 while (fdp->fd_softrefs)
2118 tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1);
2119 }
2120
2121 if (fdp->fd_files != fdp->fd_builtin_files)
2122 kfree(fdp->fd_files, M_FILEDESC);
2123 if (fdp->fd_cdir) {
2124 cache_drop(&fdp->fd_ncdir);
2125 vrele(fdp->fd_cdir);
2126 }
2127 if (fdp->fd_rdir) {
2128 cache_drop(&fdp->fd_nrdir);
2129 vrele(fdp->fd_rdir);
2130 }
2131 if (fdp->fd_jdir) {
2132 cache_drop(&fdp->fd_njdir);
2133 vrele(fdp->fd_jdir);
2134 }
2135 kfree(fdp, M_FILEDESC);
2136 }
2137
2138 /*
2139 * Retrieve and reference the file pointer associated with a descriptor.
2140 *
2141 * MPSAFE
2142 */
2143 struct file *
2144 holdfp(struct filedesc *fdp, int fd, int flag)
2145 {
2146 struct file* fp;
2147
2148 spin_lock_shared(&fdp->fd_spin);
2149 if (((u_int)fd) >= fdp->fd_nfiles) {
2150 fp = NULL;
2151 goto done;
2152 }
2153 if ((fp = fdp->fd_files[fd].fp) == NULL)
2154 goto done;
2155 if ((fp->f_flag & flag) == 0 && flag != -1) {
2156 fp = NULL;
2157 goto done;
2158 }
2159 fhold(fp);
2160 done:
2161 spin_unlock_shared(&fdp->fd_spin);
2162 return (fp);
2163 }
2164
2165 /*
2166 * holdsock() - load the struct file pointer associated
2167 * with a socket into *fpp. If an error occurs, non-zero
2168 * will be returned and *fpp will be set to NULL.
2169 *
2170 * MPSAFE
2171 */
2172 int
2173 holdsock(struct filedesc *fdp, int fd, struct file **fpp)
2174 {
2175 struct file *fp;
2176 int error;
2177
2178 spin_lock_shared(&fdp->fd_spin);
2179 if ((unsigned)fd >= fdp->fd_nfiles) {
2180 error = EBADF;
2181 fp = NULL;
2182 goto done;
2183 }
2184 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2185 error = EBADF;
2186 goto done;
2187 }
2188 if (fp->f_type != DTYPE_SOCKET) {
2189 error = ENOTSOCK;
2190 goto done;
2191 }
2192 fhold(fp);
2193 error = 0;
2194 done:
2195 spin_unlock_shared(&fdp->fd_spin);
2196 *fpp = fp;
2197 return (error);
2198 }
2199
2200 /*
2201 * Convert a user file descriptor to a held file pointer.
2202 *
2203 * MPSAFE
2204 */
2205 int
2206 holdvnode(struct filedesc *fdp, int fd, struct file **fpp)
2207 {
2208 struct file *fp;
2209 int error;
2210
2211 spin_lock_shared(&fdp->fd_spin);
2212 if ((unsigned)fd >= fdp->fd_nfiles) {
2213 error = EBADF;
2214 fp = NULL;
2215 goto done;
2216 }
2217 if ((fp = fdp->fd_files[fd].fp) == NULL) {
2218 error = EBADF;
2219 goto done;
2220 }
2221 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
2222 fp = NULL;
2223 error = EINVAL;
2224 goto done;
2225 }
2226 fhold(fp);
2227 error = 0;
2228 done:
2229 spin_unlock_shared(&fdp->fd_spin);
2230 *fpp = fp;
2231 return (error);
2232 }
2233
2234 /*
2235 * For setugid programs, we don't want to people to use that setugidness
2236 * to generate error messages which write to a file which otherwise would
2237 * otherwise be off-limits to the process.
2238 *
2239 * This is a gross hack to plug the hole. A better solution would involve
2240 * a special vop or other form of generalized access control mechanism. We
2241 * go ahead and just reject all procfs file systems accesses as dangerous.
2242 *
2243 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2244 * sufficient. We also don't for check setugidness since we know we are.
2245 */
2246 static int
2247 is_unsafe(struct file *fp)
2248 {
2249 if (fp->f_type == DTYPE_VNODE &&
2250 ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS)
2251 return (1);
2252 return (0);
2253 }
2254
2255 /*
2256 * Make this setguid thing safe, if at all possible.
2257 *
2258 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2259 */
2260 void
2261 setugidsafety(struct proc *p)
2262 {
2263 struct filedesc *fdp = p->p_fd;
2264 int i;
2265
2266 /* Certain daemons might not have file descriptors. */
2267 if (fdp == NULL)
2268 return;
2269
2270 /*
2271 * note: fdp->fd_files may be reallocated out from under us while
2272 * we are blocked in a close. Be careful!
2273 */
2274 for (i = 0; i <= fdp->fd_lastfile; i++) {
2275 if (i > 2)
2276 break;
2277 if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) {
2278 struct file *fp;
2279
2280 /*
2281 * NULL-out descriptor prior to close to avoid
2282 * a race while close blocks.
2283 */
2284 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2285 knote_fdclose(fp, fdp, i);
2286 closef(fp, p);
2287 }
2288 }
2289 }
2290 }
2291
2292 /*
2293 * Close any files on exec?
2294 *
2295 * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose()
2296 */
2297 void
2298 fdcloseexec(struct proc *p)
2299 {
2300 struct filedesc *fdp = p->p_fd;
2301 int i;
2302
2303 /* Certain daemons might not have file descriptors. */
2304 if (fdp == NULL)
2305 return;
2306
2307 /*
2308 * We cannot cache fd_files since operations may block and rip
2309 * them out from under us.
2310 */
2311 for (i = 0; i <= fdp->fd_lastfile; i++) {
2312 if (fdp->fd_files[i].fp != NULL &&
2313 (fdp->fd_files[i].fileflags & UF_EXCLOSE)) {
2314 struct file *fp;
2315
2316 /*
2317 * NULL-out descriptor prior to close to avoid
2318 * a race while close blocks.
2319 */
2320 if ((fp = funsetfd_locked(fdp, i)) != NULL) {
2321 knote_fdclose(fp, fdp, i);
2322 closef(fp, p);
2323 }
2324 }
2325 }
2326 }
2327
2328 /*
2329 * It is unsafe for set[ug]id processes to be started with file
2330 * descriptors 0..2 closed, as these descriptors are given implicit
2331 * significance in the Standard C library. fdcheckstd() will create a
2332 * descriptor referencing /dev/null for each of stdin, stdout, and
2333 * stderr that is not already open.
2334 *
2335 * NOT MPSAFE - calls falloc, vn_open, etc
2336 */
2337 int
2338 fdcheckstd(struct lwp *lp)
2339 {
2340 struct nlookupdata nd;
2341 struct filedesc *fdp;
2342 struct file *fp;
2343 int retval;
2344 int i, error, flags, devnull;
2345
2346 fdp = lp->lwp_proc->p_fd;
2347 if (fdp == NULL)
2348 return (0);
2349 devnull = -1;
2350 error = 0;
2351 for (i = 0; i < 3; i++) {
2352 if (fdp->fd_files[i].fp != NULL)
2353 continue;
2354 if (devnull < 0) {
2355 if ((error = falloc(lp, &fp, &devnull)) != 0)
2356 break;
2357
2358 error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE,
2359 NLC_FOLLOW|NLC_LOCKVP);
2360 flags = FREAD | FWRITE;
2361 if (error == 0)
2362 error = vn_open(&nd, fp, flags, 0);
2363 if (error == 0)
2364 fsetfd(fdp, fp, devnull);
2365 else
2366 fsetfd(fdp, NULL, devnull);
2367 fdrop(fp);
2368 nlookup_done(&nd);
2369 if (error)
2370 break;
2371 KKASSERT(i == devnull);
2372 } else {
2373 error = kern_dup(DUP_FIXED, devnull, i, &retval);
2374 if (error != 0)
2375 break;
2376 }
2377 }
2378 return (error);
2379 }
2380
2381 /*
2382 * Internal form of close.
2383 * Decrement reference count on file structure.
2384 * Note: td and/or p may be NULL when closing a file
2385 * that was being passed in a message.
2386 *
2387 * MPALMOSTSAFE - acquires mplock for VOP operations
2388 */
2389 int
2390 closef(struct file *fp, struct proc *p)
2391 {
2392 struct vnode *vp;
2393 struct flock lf;
2394 struct filedesc_to_leader *fdtol;
2395
2396 if (fp == NULL)
2397 return (0);
2398
2399 /*
2400 * POSIX record locking dictates that any close releases ALL
2401 * locks owned by this process. This is handled by setting
2402 * a flag in the unlock to free ONLY locks obeying POSIX
2403 * semantics, and not to free BSD-style file locks.
2404 * If the descriptor was in a message, POSIX-style locks
2405 * aren't passed with the descriptor.
2406 */
2407 if (p != NULL && fp->f_type == DTYPE_VNODE &&
2408 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2409 ) {
2410 if ((p->p_leader->p_flags & P_ADVLOCK) != 0) {
2411 lf.l_whence = SEEK_SET;
2412 lf.l_start = 0;
2413 lf.l_len = 0;
2414 lf.l_type = F_UNLCK;
2415 vp = (struct vnode *)fp->f_data;
2416 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
2417 &lf, F_POSIX);
2418 }
2419 fdtol = p->p_fdtol;
2420 if (fdtol != NULL) {
2421 lwkt_gettoken(&p->p_token);
2422 /*
2423 * Handle special case where file descriptor table
2424 * is shared between multiple process leaders.
2425 */
2426 for (fdtol = fdtol->fdl_next;
2427 fdtol != p->p_fdtol;
2428 fdtol = fdtol->fdl_next) {
2429 if ((fdtol->fdl_leader->p_flags &
2430 P_ADVLOCK) == 0)
2431 continue;
2432 fdtol->fdl_holdcount++;
2433 lf.l_whence = SEEK_SET;
2434 lf.l_start = 0;
2435 lf.l_len = 0;
2436 lf.l_type = F_UNLCK;
2437 vp = (struct vnode *)fp->f_data;
2438 (void) VOP_ADVLOCK(vp,
2439 (caddr_t)fdtol->fdl_leader,
2440 F_UNLCK, &lf, F_POSIX);
2441 fdtol->fdl_holdcount--;
2442 if (fdtol->fdl_holdcount == 0 &&
2443 fdtol->fdl_wakeup != 0) {
2444 fdtol->fdl_wakeup = 0;
2445 wakeup(fdtol);
2446 }
2447 }
2448 lwkt_reltoken(&p->p_token);
2449 }
2450 }
2451 return (fdrop(fp));
2452 }
2453
2454 /*
2455 * MPSAFE
2456 *
2457 * fhold() can only be called if f_count is already at least 1 (i.e. the
2458 * caller of fhold() already has a reference to the file pointer in some
2459 * manner or other).
2460 *
2461 * f_count is not spin-locked. Instead, atomic ops are used for
2462 * incrementing, decrementing, and handling the 1->0 transition.
2463 */
2464 void
2465 fhold(struct file *fp)
2466 {
2467 atomic_add_int(&fp->f_count, 1);
2468 }
2469
2470 /*
2471 * fdrop() - drop a reference to a descriptor
2472 *
2473 * MPALMOSTSAFE - acquires mplock for final close sequence
2474 */
2475 int
2476 fdrop(struct file *fp)
2477 {
2478 struct flock lf;
2479 struct vnode *vp;
2480 int error;
2481
2482 /*
2483 * A combined fetch and subtract is needed to properly detect
2484 * 1->0 transitions, otherwise two cpus dropping from a ref
2485 * count of 2 might both try to run the 1->0 code.
2486 */
2487 if (atomic_fetchadd_int(&fp->f_count, -1) > 1)
2488 return (0);
2489
2490 KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL);
2491
2492 /*
2493 * The last reference has gone away, we own the fp structure free
2494 * and clear.
2495 */
2496 if (fp->f_count < 0)
2497 panic("fdrop: count < 0");
2498 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE &&
2499 (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS)
2500 ) {
2501 lf.l_whence = SEEK_SET;
2502 lf.l_start = 0;
2503 lf.l_len = 0;
2504 lf.l_type = F_UNLCK;
2505 vp = (struct vnode *)fp->f_data;
2506 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2507 }
2508 if (fp->f_ops != &badfileops)
2509 error = fo_close(fp);
2510 else
2511 error = 0;
2512 ffree(fp);
2513 return (error);
2514 }
2515
2516 /*
2517 * Apply an advisory lock on a file descriptor.
2518 *
2519 * Just attempt to get a record lock of the requested type on
2520 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2521 *
2522 * MPALMOSTSAFE
2523 */
2524 int
2525 sys_flock(struct flock_args *uap)
2526 {
2527 struct proc *p = curproc;
2528 struct file *fp;
2529 struct vnode *vp;
2530 struct flock lf;
2531 int error;
2532
2533 if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL)
2534 return (EBADF);
2535 if (fp->f_type != DTYPE_VNODE) {
2536 error = EOPNOTSUPP;
2537 goto done;
2538 }
2539 vp = (struct vnode *)fp->f_data;
2540 lf.l_whence = SEEK_SET;
2541 lf.l_start = 0;
2542 lf.l_len = 0;
2543 if (uap->how & LOCK_UN) {
2544 lf.l_type = F_UNLCK;
2545 fp->f_flag &= ~FHASLOCK;
2546 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0);
2547 goto done;
2548 }
2549 if (uap->how & LOCK_EX)
2550 lf.l_type = F_WRLCK;
2551 else if (uap->how & LOCK_SH)
2552 lf.l_type = F_RDLCK;
2553 else {
2554 error = EBADF;
2555 goto done;
2556 }
2557 fp->f_flag |= FHASLOCK;
2558 if (uap->how & LOCK_NB)
2559 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0);
2560 else
2561 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT);
2562 done:
2563 fdrop(fp);
2564 return (error);
2565 }
2566
2567 /*
2568 * File Descriptor pseudo-device driver (/dev/fd/).
2569 *
2570 * Opening minor device N dup()s the file (if any) connected to file
2571 * descriptor N belonging to the calling process. Note that this driver
2572 * consists of only the ``open()'' routine, because all subsequent
2573 * references to this file will be direct to the other driver.
2574 */
2575 static int
2576 fdopen(struct dev_open_args *ap)
2577 {
2578 thread_t td = curthread;
2579
2580 KKASSERT(td->td_lwp != NULL);
2581
2582 /*
2583 * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the
2584 * the file descriptor being sought for duplication. The error
2585 * return ensures that the vnode for this device will be released
2586 * by vn_open. Open will detect this special error and take the
2587 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2588 * will simply report the error.
2589 */
2590 td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev);
2591 return (ENODEV);
2592 }
2593
2594 /*
2595 * The caller has reserved the file descriptor dfd for us. On success we
2596 * must fsetfd() it. On failure the caller will clean it up.
2597 *
2598 * MPSAFE
2599 */
2600 int
2601 dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error)
2602 {
2603 struct file *wfp;
2604 struct file *xfp;
2605 int werror;
2606
2607 if ((wfp = holdfp(fdp, sfd, -1)) == NULL)
2608 return (EBADF);
2609
2610 /*
2611 * Close a revoke/dup race. Duping a descriptor marked as revoked
2612 * will dup a dummy descriptor instead of the real one.
2613 */
2614 if (wfp->f_flag & FREVOKED) {
2615 kprintf("Warning: attempt to dup() a revoked descriptor\n");
2616 fdrop(wfp);
2617 wfp = NULL;
2618 werror = falloc(NULL, &wfp, NULL);
2619 if (werror)
2620 return (werror);
2621 }
2622
2623 /*
2624 * There are two cases of interest here.
2625 *
2626 * For ENODEV simply dup sfd to file descriptor dfd and return.
2627 *
2628 * For ENXIO steal away the file structure from sfd and store it
2629 * dfd. sfd is effectively closed by this operation.
2630 *
2631 * Any other error code is just returned.
2632 */
2633 switch (error) {
2634 case ENODEV:
2635 /*
2636 * Check that the mode the file is being opened for is a
2637 * subset of the mode of the existing descriptor.
2638 */
2639 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2640 error = EACCES;
2641 break;
2642 }
2643 spin_lock(&fdp->fd_spin);
2644 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2645 fsetfd_locked(fdp, wfp, dfd);
2646 spin_unlock(&fdp->fd_spin);
2647 error = 0;
2648 break;
2649 case ENXIO:
2650 /*
2651 * Steal away the file pointer from dfd, and stuff it into indx.
2652 */
2653 spin_lock(&fdp->fd_spin);
2654 fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags;
2655 fsetfd(fdp, wfp, dfd);
2656 if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) {
2657 spin_unlock(&fdp->fd_spin);
2658 fdrop(xfp);
2659 } else {
2660 spin_unlock(&fdp->fd_spin);
2661 }
2662 error = 0;
2663 break;
2664 default:
2665 break;
2666 }
2667 fdrop(wfp);
2668 return (error);
2669 }
2670
2671 /*
2672 * NOT MPSAFE - I think these refer to a common file descriptor table
2673 * and we need to spinlock that to link fdtol in.
2674 */
2675 struct filedesc_to_leader *
2676 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
2677 struct proc *leader)
2678 {
2679 struct filedesc_to_leader *fdtol;
2680
2681 fdtol = kmalloc(sizeof(struct filedesc_to_leader),
2682 M_FILEDESC_TO_LEADER, M_WAITOK | M_ZERO);
2683 fdtol->fdl_refcount = 1;
2684 fdtol->fdl_holdcount = 0;
2685 fdtol->fdl_wakeup = 0;
2686 fdtol->fdl_leader = leader;
2687 if (old != NULL) {
2688 fdtol->fdl_next = old->fdl_next;
2689 fdtol->fdl_prev = old;
2690 old->fdl_next = fdtol;
2691 fdtol->fdl_next->fdl_prev = fdtol;
2692 } else {
2693 fdtol->fdl_next = fdtol;
2694 fdtol->fdl_prev = fdtol;
2695 }
2696 return fdtol;
2697 }
2698
2699 /*
2700 * Scan all file pointers in the system. The callback is made with
2701 * the master list spinlock held exclusively.
2702 *
2703 * MPSAFE
2704 */
2705 void
2706 allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data)
2707 {
2708 struct file *fp;
2709 int res;
2710
2711 spin_lock(&filehead_spin);
2712 LIST_FOREACH(fp, &filehead, f_list) {
2713 res = callback(fp, data);
2714 if (res < 0)
2715 break;
2716 }
2717 spin_unlock(&filehead_spin);
2718 }
2719
2720 /*
2721 * Get file structures.
2722 *
2723 * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe)
2724 */
2725
2726 struct sysctl_kern_file_info {
2727 int count;
2728 int error;
2729 struct sysctl_req *req;
2730 };
2731
2732 static int sysctl_kern_file_callback(struct proc *p, void *data);
2733
2734 static int
2735 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2736 {
2737 struct sysctl_kern_file_info info;
2738
2739 /*
2740 * Note: because the number of file descriptors is calculated
2741 * in different ways for sizing vs returning the data,
2742 * there is information leakage from the first loop. However,
2743 * it is of a similar order of magnitude to the leakage from
2744 * global system statistics such as kern.openfiles.
2745 *
2746 * When just doing a count, note that we cannot just count
2747 * the elements and add f_count via the filehead list because
2748 * threaded processes share their descriptor table and f_count might
2749 * still be '1' in that case.
2750 *
2751 * Since the SYSCTL op can block, we must hold the process to
2752 * prevent it being ripped out from under us either in the
2753 * file descriptor loop or in the greater LIST_FOREACH. The
2754 * process may be in varying states of disrepair. If the process
2755 * is in SZOMB we may have caught it just as it is being removed
2756 * from the allproc list, we must skip it in that case to maintain
2757 * an unbroken chain through the allproc list.
2758 */
2759 info.count = 0;
2760 info.error = 0;
2761 info.req = req;
2762 allproc_scan(sysctl_kern_file_callback, &info);
2763
2764 /*
2765 * When just calculating the size, overestimate a bit to try to
2766 * prevent system activity from causing the buffer-fill call
2767 * to fail later on.
2768 */
2769 if (req->oldptr == NULL) {
2770 info.count = (info.count + 16) + (info.count / 10);
2771 info.error = SYSCTL_OUT(req, NULL,
2772 info.count * sizeof(struct kinfo_file));
2773 }
2774 return (info.error);
2775 }
2776
2777 static int
2778 sysctl_kern_file_callback(struct proc *p, void *data)
2779 {
2780 struct sysctl_kern_file_info *info = data;
2781 struct kinfo_file kf;
2782 struct filedesc *fdp;
2783 struct file *fp;
2784 uid_t uid;
2785 int n;
2786
2787 if (p->p_stat == SIDL || p->p_stat == SZOMB)
2788 return(0);
2789 if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0)
2790 return(0);
2791
2792 /*
2793 * Softref the fdp to prevent it from being destroyed
2794 */
2795 spin_lock(&p->p_spin);
2796 if ((fdp = p->p_fd) == NULL) {
2797 spin_unlock(&p->p_spin);
2798 return(0);
2799 }
2800 atomic_add_int(&fdp->fd_softrefs, 1);
2801 spin_unlock(&p->p_spin);
2802
2803 /*
2804 * The fdp's own spinlock prevents the contents from being
2805 * modified.
2806 */
2807 spin_lock_shared(&fdp->fd_spin);
2808 for (n = 0; n < fdp->fd_nfiles; ++n) {
2809 if ((fp = fdp->fd_files[n].fp) == NULL)
2810 continue;
2811 if (info->req->oldptr == NULL) {
2812 ++info->count;
2813 } else {
2814 uid = p->p_ucred ? p->p_ucred->cr_uid : -1;
2815 kcore_make_file(&kf, fp, p->p_pid, uid, n);
2816 spin_unlock_shared(&fdp->fd_spin);
2817 info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf));
2818 spin_lock_shared(&fdp->fd_spin);
2819 if (info->error)
2820 break;
2821 }
2822 }
2823 spin_unlock_shared(&fdp->fd_spin);
2824 atomic_subtract_int(&fdp->fd_softrefs, 1);
2825 if (info->error)
2826 return(-1);
2827 return(0);
2828 }
2829
2830 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2831 0, 0, sysctl_kern_file, "S,file", "Entire file table");
2832
2833 SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW,
2834 &minfilesperproc, 0, "Minimum files allowed open per process");
2835 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2836 &maxfilesperproc, 0, "Maximum files allowed open per process");
2837 SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW,
2838 &maxfilesperuser, 0, "Maximum files allowed open per user");
2839
2840 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2841 &maxfiles, 0, "Maximum number of files");
2842
2843 SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW,
2844 &maxfilesrootres, 0, "Descriptors reserved for root use");
2845
2846 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2847 &nfiles, 0, "System-wide number of open files");
2848
2849 static void
2850 fildesc_drvinit(void *unused)
2851 {
2852 int fd;
2853
2854 for (fd = 0; fd < NUMFDESC; fd++) {
2855 make_dev(&fildesc_ops, fd,
2856 UID_BIN, GID_BIN, 0666, "fd/%d", fd);
2857 }
2858
2859 make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin");
2860 make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout");
2861 make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr");
2862 }
2863
2864 /*
2865 * MPSAFE
2866 */
2867 struct fileops badfileops = {
2868 .fo_read = badfo_readwrite,
2869 .fo_write = badfo_readwrite,
2870 .fo_ioctl = badfo_ioctl,
2871 .fo_kqfilter = badfo_kqfilter,
2872 .fo_stat = badfo_stat,
2873 .fo_close = badfo_close,
2874 .fo_shutdown = badfo_shutdown
2875 };
2876
2877 int
2878 badfo_readwrite(
2879 struct file *fp,
2880 struct uio *uio,
2881 struct ucred *cred,
2882 int flags
2883 ) {
2884 return (EBADF);
2885 }
2886
2887 int
2888 badfo_ioctl(struct file *fp, u_long com, caddr_t data,
2889 struct ucred *cred, struct sysmsg *msgv)
2890 {
2891 return (EBADF);
2892 }
2893
2894 /*
2895 * Must return an error to prevent registration, typically
2896 * due to a revoked descriptor (file_filtops assigned).
2897 */
2898 int
2899 badfo_kqfilter(struct file *fp, struct knote *kn)
2900 {
2901 return (EOPNOTSUPP);
2902 }
2903
2904 /*
2905 * MPSAFE
2906 */
2907 int
2908 badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
2909 {
2910 return (EBADF);
2911 }
2912
2913 /*
2914 * MPSAFE
2915 */
2916 int
2917 badfo_close(struct file *fp)
2918 {
2919 return (EBADF);
2920 }
2921
2922 /*
2923 * MPSAFE
2924 */
2925 int
2926 badfo_shutdown(struct file *fp, int how)
2927 {
2928 return (EBADF);
2929 }
2930
2931 /*
2932 * MPSAFE
2933 */
2934 int
2935 nofo_shutdown(struct file *fp, int how)
2936 {
2937 return (EOPNOTSUPP);
2938 }
2939
2940 SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,
2941 fildesc_drvinit,NULL)
Cache object: 1b24b56a8c27dfe7325543408f149784
|