1/* $NetBSD: kern_descrip.c,v 1.229 2015/08/03 04:55:15 christos Exp $ */
2
3/*-
4 * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Andrew Doran.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32/*
33 * Copyright (c) 1982, 1986, 1989, 1991, 1993
34 * The Regents of the University of California. All rights reserved.
35 * (c) UNIX System Laboratories, Inc.
36 * All or some portions of this file are derived from material licensed
37 * to the University of California by American Telephone and Telegraph
38 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
39 * the permission of UNIX System Laboratories, Inc.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 * notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 * notice, this list of conditions and the following disclaimer in the
48 * documentation and/or other materials provided with the distribution.
49 * 3. Neither the name of the University nor the names of its contributors
50 * may be used to endorse or promote products derived from this software
51 * without specific prior written permission.
52 *
53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63 * SUCH DAMAGE.
64 *
65 * @(#)kern_descrip.c 8.8 (Berkeley) 2/14/95
66 */
67
68/*
69 * File descriptor management.
70 */
71
72#include <sys/cdefs.h>
73__KERNEL_RCSID(0, "$NetBSD: kern_descrip.c,v 1.229 2015/08/03 04:55:15 christos Exp $");
74
75#include <sys/param.h>
76#include <sys/systm.h>
77#include <sys/filedesc.h>
78#include <sys/kernel.h>
79#include <sys/proc.h>
80#include <sys/file.h>
81#include <sys/socket.h>
82#include <sys/socketvar.h>
83#include <sys/stat.h>
84#include <sys/ioctl.h>
85#include <sys/fcntl.h>
86#include <sys/pool.h>
87#include <sys/unistd.h>
88#include <sys/resourcevar.h>
89#include <sys/conf.h>
90#include <sys/event.h>
91#include <sys/kauth.h>
92#include <sys/atomic.h>
93#include <sys/syscallargs.h>
94#include <sys/cpu.h>
95#include <sys/kmem.h>
96#include <sys/vnode.h>
97#include <sys/sysctl.h>
98#include <sys/ktrace.h>
99
100/*
101 * A list (head) of open files, counter, and lock protecting them.
102 */
103struct filelist filehead __cacheline_aligned;
104static u_int nfiles __cacheline_aligned;
105kmutex_t filelist_lock __cacheline_aligned;
106
107static pool_cache_t filedesc_cache __read_mostly;
108static pool_cache_t file_cache __read_mostly;
109static pool_cache_t fdfile_cache __read_mostly;
110
111static int file_ctor(void *, void *, int);
112static void file_dtor(void *, void *);
113static int fdfile_ctor(void *, void *, int);
114static void fdfile_dtor(void *, void *);
115static int filedesc_ctor(void *, void *, int);
116static void filedesc_dtor(void *, void *);
117static int filedescopen(dev_t, int, int, lwp_t *);
118
119static int sysctl_kern_file(SYSCTLFN_PROTO);
120static int sysctl_kern_file2(SYSCTLFN_PROTO);
121static void fill_file(struct kinfo_file *, const file_t *, const fdfile_t *,
122 int, pid_t);
123
124const struct cdevsw filedesc_cdevsw = {
125 .d_open = filedescopen,
126 .d_close = noclose,
127 .d_read = noread,
128 .d_write = nowrite,
129 .d_ioctl = noioctl,
130 .d_stop = nostop,
131 .d_tty = notty,
132 .d_poll = nopoll,
133 .d_mmap = nommap,
134 .d_kqfilter = nokqfilter,
135 .d_discard = nodiscard,
136 .d_flag = D_OTHER | D_MPSAFE
137};
138
139/* For ease of reading. */
140__strong_alias(fd_putvnode,fd_putfile)
141__strong_alias(fd_putsock,fd_putfile)
142
143/*
144 * Initialize the descriptor system.
145 */
146void
147fd_sys_init(void)
148{
149 static struct sysctllog *clog;
150
151 mutex_init(&filelist_lock, MUTEX_DEFAULT, IPL_NONE);
152
153 file_cache = pool_cache_init(sizeof(file_t), coherency_unit, 0,
154 0, "file", NULL, IPL_NONE, file_ctor, file_dtor, NULL);
155 KASSERT(file_cache != NULL);
156
157 fdfile_cache = pool_cache_init(sizeof(fdfile_t), coherency_unit, 0,
158 PR_LARGECACHE, "fdfile", NULL, IPL_NONE, fdfile_ctor, fdfile_dtor,
159 NULL);
160 KASSERT(fdfile_cache != NULL);
161
162 filedesc_cache = pool_cache_init(sizeof(filedesc_t), coherency_unit,
163 0, 0, "filedesc", NULL, IPL_NONE, filedesc_ctor, filedesc_dtor,
164 NULL);
165 KASSERT(filedesc_cache != NULL);
166
167 sysctl_createv(&clog, 0, NULL, NULL,
168 CTLFLAG_PERMANENT,
169 CTLTYPE_STRUCT, "file",
170 SYSCTL_DESCR("System open file table"),
171 sysctl_kern_file, 0, NULL, 0,
172 CTL_KERN, KERN_FILE, CTL_EOL);
173 sysctl_createv(&clog, 0, NULL, NULL,
174 CTLFLAG_PERMANENT,
175 CTLTYPE_STRUCT, "file2",
176 SYSCTL_DESCR("System open file table"),
177 sysctl_kern_file2, 0, NULL, 0,
178 CTL_KERN, KERN_FILE2, CTL_EOL);
179}
180
181static bool
182fd_isused(filedesc_t *fdp, unsigned fd)
183{
184 u_int off = fd >> NDENTRYSHIFT;
185
186 KASSERT(fd < fdp->fd_dt->dt_nfiles);
187
188 return (fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0;
189}
190
191/*
192 * Verify that the bitmaps match the descriptor table.
193 */
194static inline void
195fd_checkmaps(filedesc_t *fdp)
196{
197#ifdef DEBUG
198 fdtab_t *dt;
199 u_int fd;
200
201 dt = fdp->fd_dt;
202 if (fdp->fd_refcnt == -1) {
203 /*
204 * fd_free tears down the table without maintaining its bitmap.
205 */
206 return;
207 }
208 for (fd = 0; fd < dt->dt_nfiles; fd++) {
209 if (fd < NDFDFILE) {
210 KASSERT(dt->dt_ff[fd] ==
211 (fdfile_t *)fdp->fd_dfdfile[fd]);
212 }
213 if (dt->dt_ff[fd] == NULL) {
214 KASSERT(!fd_isused(fdp, fd));
215 } else if (dt->dt_ff[fd]->ff_file != NULL) {
216 KASSERT(fd_isused(fdp, fd));
217 }
218 }
219#endif
220}
221
222static int
223fd_next_zero(filedesc_t *fdp, uint32_t *bitmap, int want, u_int bits)
224{
225 int i, off, maxoff;
226 uint32_t sub;
227
228 KASSERT(mutex_owned(&fdp->fd_lock));
229
230 fd_checkmaps(fdp);
231
232 if (want > bits)
233 return -1;
234
235 off = want >> NDENTRYSHIFT;
236 i = want & NDENTRYMASK;
237 if (i) {
238 sub = bitmap[off] | ((u_int)~0 >> (NDENTRIES - i));
239 if (sub != ~0)
240 goto found;
241 off++;
242 }
243
244 maxoff = NDLOSLOTS(bits);
245 while (off < maxoff) {
246 if ((sub = bitmap[off]) != ~0)
247 goto found;
248 off++;
249 }
250
251 return -1;
252
253 found:
254 return (off << NDENTRYSHIFT) + ffs(~sub) - 1;
255}
256
257static int
258fd_last_set(filedesc_t *fd, int last)
259{
260 int off, i;
261 fdfile_t **ff = fd->fd_dt->dt_ff;
262 uint32_t *bitmap = fd->fd_lomap;
263
264 KASSERT(mutex_owned(&fd->fd_lock));
265
266 fd_checkmaps(fd);
267
268 off = (last - 1) >> NDENTRYSHIFT;
269
270 while (off >= 0 && !bitmap[off])
271 off--;
272
273 if (off < 0)
274 return -1;
275
276 i = ((off + 1) << NDENTRYSHIFT) - 1;
277 if (i >= last)
278 i = last - 1;
279
280 /* XXX should use bitmap */
281 while (i > 0 && (ff[i] == NULL || !ff[i]->ff_allocated))
282 i--;
283
284 return i;
285}
286
287static inline void
288fd_used(filedesc_t *fdp, unsigned fd)
289{
290 u_int off = fd >> NDENTRYSHIFT;
291 fdfile_t *ff;
292
293 ff = fdp->fd_dt->dt_ff[fd];
294
295 KASSERT(mutex_owned(&fdp->fd_lock));
296 KASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) == 0);
297 KASSERT(ff != NULL);
298 KASSERT(ff->ff_file == NULL);
299 KASSERT(!ff->ff_allocated);
300
301 ff->ff_allocated = true;
302 fdp->fd_lomap[off] |= 1 << (fd & NDENTRYMASK);
303 if (__predict_false(fdp->fd_lomap[off] == ~0)) {
304 KASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
305 (1 << (off & NDENTRYMASK))) == 0);
306 fdp->fd_himap[off >> NDENTRYSHIFT] |= 1 << (off & NDENTRYMASK);
307 }
308
309 if ((int)fd > fdp->fd_lastfile) {
310 fdp->fd_lastfile = fd;
311 }
312
313 fd_checkmaps(fdp);
314}
315
316static inline void
317fd_unused(filedesc_t *fdp, unsigned fd)
318{
319 u_int off = fd >> NDENTRYSHIFT;
320 fdfile_t *ff;
321
322 ff = fdp->fd_dt->dt_ff[fd];
323
324 /*
325 * Don't assert the lock is held here, as we may be copying
326 * the table during exec() and it is not needed there.
327 * procfs and sysctl are locked out by proc::p_reflock.
328 *
329 * KASSERT(mutex_owned(&fdp->fd_lock));
330 */
331 KASSERT(ff != NULL);
332 KASSERT(ff->ff_file == NULL);
333 KASSERT(ff->ff_allocated);
334
335 if (fd < fdp->fd_freefile) {
336 fdp->fd_freefile = fd;
337 }
338
339 if (fdp->fd_lomap[off] == ~0) {
340 KASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
341 (1 << (off & NDENTRYMASK))) != 0);
342 fdp->fd_himap[off >> NDENTRYSHIFT] &=
343 ~(1 << (off & NDENTRYMASK));
344 }
345 KASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0);
346 fdp->fd_lomap[off] &= ~(1 << (fd & NDENTRYMASK));
347 ff->ff_allocated = false;
348
349 KASSERT(fd <= fdp->fd_lastfile);
350 if (fd == fdp->fd_lastfile) {
351 fdp->fd_lastfile = fd_last_set(fdp, fd);
352 }
353 fd_checkmaps(fdp);
354}
355
356/*
357 * Look up the file structure corresponding to a file descriptor
358 * and return the file, holding a reference on the descriptor.
359 */
360file_t *
361fd_getfile(unsigned fd)
362{
363 filedesc_t *fdp;
364 fdfile_t *ff;
365 file_t *fp;
366 fdtab_t *dt;
367
368 /*
369 * Look up the fdfile structure representing this descriptor.
370 * We are doing this unlocked. See fd_tryexpand().
371 */
372 fdp = curlwp->l_fd;
373 dt = fdp->fd_dt;
374 if (__predict_false(fd >= dt->dt_nfiles)) {
375 return NULL;
376 }
377 ff = dt->dt_ff[fd];
378 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
379 if (__predict_false(ff == NULL)) {
380 return NULL;
381 }
382
383 /* Now get a reference to the descriptor. */
384 if (fdp->fd_refcnt == 1) {
385 /*
386 * Single threaded: don't need to worry about concurrent
387 * access (other than earlier calls to kqueue, which may
388 * hold a reference to the descriptor).
389 */
390 ff->ff_refcnt++;
391 } else {
392 /*
393 * Multi threaded: issue a memory barrier to ensure that we
394 * acquire the file pointer _after_ adding a reference. If
395 * no memory barrier, we could fetch a stale pointer.
396 */
397 atomic_inc_uint(&ff->ff_refcnt);
398#ifndef __HAVE_ATOMIC_AS_MEMBAR
399 membar_enter();
400#endif
401 }
402
403 /*
404 * If the file is not open or is being closed then put the
405 * reference back.
406 */
407 fp = ff->ff_file;
408 if (__predict_true(fp != NULL)) {
409 return fp;
410 }
411 fd_putfile(fd);
412 return NULL;
413}
414
415/*
416 * Release a reference to a file descriptor acquired with fd_getfile().
417 */
418void
419fd_putfile(unsigned fd)
420{
421 filedesc_t *fdp;
422 fdfile_t *ff;
423 u_int u, v;
424
425 fdp = curlwp->l_fd;
426 ff = fdp->fd_dt->dt_ff[fd];
427
428 KASSERT(fd < fdp->fd_dt->dt_nfiles);
429 KASSERT(ff != NULL);
430 KASSERT((ff->ff_refcnt & FR_MASK) > 0);
431 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
432
433 if (fdp->fd_refcnt == 1) {
434 /*
435 * Single threaded: don't need to worry about concurrent
436 * access (other than earlier calls to kqueue, which may
437 * hold a reference to the descriptor).
438 */
439 if (__predict_false((ff->ff_refcnt & FR_CLOSING) != 0)) {
440 fd_close(fd);
441 return;
442 }
443 ff->ff_refcnt--;
444 return;
445 }
446
447 /*
448 * Ensure that any use of the file is complete and globally
449 * visible before dropping the final reference. If no membar,
450 * the current CPU could still access memory associated with
451 * the file after it has been freed or recycled by another
452 * CPU.
453 */
454#ifndef __HAVE_ATOMIC_AS_MEMBAR
455 membar_exit();
456#endif
457
458 /*
459 * Be optimistic and start out with the assumption that no other
460 * threads are trying to close the descriptor. If the CAS fails,
461 * we lost a race and/or it's being closed.
462 */
463 for (u = ff->ff_refcnt & FR_MASK;; u = v) {
464 v = atomic_cas_uint(&ff->ff_refcnt, u, u - 1);
465 if (__predict_true(u == v)) {
466 return;
467 }
468 if (__predict_false((v & FR_CLOSING) != 0)) {
469 break;
470 }
471 }
472
473 /* Another thread is waiting to close the file: join it. */
474 (void)fd_close(fd);
475}
476
477/*
478 * Convenience wrapper around fd_getfile() that returns reference
479 * to a vnode.
480 */
481int
482fd_getvnode(unsigned fd, file_t **fpp)
483{
484 vnode_t *vp;
485 file_t *fp;
486
487 fp = fd_getfile(fd);
488 if (__predict_false(fp == NULL)) {
489 return EBADF;
490 }
491 if (__predict_false(fp->f_type != DTYPE_VNODE)) {
492 fd_putfile(fd);
493 return EINVAL;
494 }
495 vp = fp->f_vnode;
496 if (__predict_false(vp->v_type == VBAD)) {
497 /* XXX Is this case really necessary? */
498 fd_putfile(fd);
499 return EBADF;
500 }
501 *fpp = fp;
502 return 0;
503}
504
505/*
506 * Convenience wrapper around fd_getfile() that returns reference
507 * to a socket.
508 */
509int
510fd_getsock1(unsigned fd, struct socket **sop, file_t **fp)
511{
512 *fp = fd_getfile(fd);
513 if (__predict_false(*fp == NULL)) {
514 return EBADF;
515 }
516 if (__predict_false((*fp)->f_type != DTYPE_SOCKET)) {
517 fd_putfile(fd);
518 return ENOTSOCK;
519 }
520 *sop = (*fp)->f_socket;
521 return 0;
522}
523
524int
525fd_getsock(unsigned fd, struct socket **sop)
526{
527 file_t *fp;
528 return fd_getsock1(fd, sop, &fp);
529}
530
531/*
532 * Look up the file structure corresponding to a file descriptor
533 * and return it with a reference held on the file, not the
534 * descriptor.
535 *
536 * This is heavyweight and only used when accessing descriptors
537 * from a foreign process. The caller must ensure that `p' does
538 * not exit or fork across this call.
539 *
540 * To release the file (not descriptor) reference, use closef().
541 */
542file_t *
543fd_getfile2(proc_t *p, unsigned fd)
544{
545 filedesc_t *fdp;
546 fdfile_t *ff;
547 file_t *fp;
548 fdtab_t *dt;
549
550 fdp = p->p_fd;
551 mutex_enter(&fdp->fd_lock);
552 dt = fdp->fd_dt;
553 if (fd >= dt->dt_nfiles) {
554 mutex_exit(&fdp->fd_lock);
555 return NULL;
556 }
557 if ((ff = dt->dt_ff[fd]) == NULL) {
558 mutex_exit(&fdp->fd_lock);
559 return NULL;
560 }
561 if ((fp = ff->ff_file) == NULL) {
562 mutex_exit(&fdp->fd_lock);
563 return NULL;
564 }
565 mutex_enter(&fp->f_lock);
566 fp->f_count++;
567 mutex_exit(&fp->f_lock);
568 mutex_exit(&fdp->fd_lock);
569
570 return fp;
571}
572
573/*
574 * Internal form of close. Must be called with a reference to the
575 * descriptor, and will drop the reference. When all descriptor
576 * references are dropped, releases the descriptor slot and a single
577 * reference to the file structure.
578 */
579int
580fd_close(unsigned fd)
581{
582 struct flock lf;
583 filedesc_t *fdp;
584 fdfile_t *ff;
585 file_t *fp;
586 proc_t *p;
587 lwp_t *l;
588 u_int refcnt;
589
590 l = curlwp;
591 p = l->l_proc;
592 fdp = l->l_fd;
593 ff = fdp->fd_dt->dt_ff[fd];
594
595 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
596
597 mutex_enter(&fdp->fd_lock);
598 KASSERT((ff->ff_refcnt & FR_MASK) > 0);
599 if (__predict_false(ff->ff_file == NULL)) {
600 /*
601 * Another user of the file is already closing, and is
602 * waiting for other users of the file to drain. Release
603 * our reference, and wake up the closer.
604 */
605 atomic_dec_uint(&ff->ff_refcnt);
606 cv_broadcast(&ff->ff_closing);
607 mutex_exit(&fdp->fd_lock);
608
609 /*
610 * An application error, so pretend that the descriptor
611 * was already closed. We can't safely wait for it to
612 * be closed without potentially deadlocking.
613 */
614 return (EBADF);
615 }
616 KASSERT((ff->ff_refcnt & FR_CLOSING) == 0);
617
618 /*
619 * There may be multiple users of this file within the process.
620 * Notify existing and new users that the file is closing. This
621 * will prevent them from adding additional uses to this file
622 * while we are closing it.
623 */
624 fp = ff->ff_file;
625 ff->ff_file = NULL;
626 ff->ff_exclose = false;
627
628 /*
629 * We expect the caller to hold a descriptor reference - drop it.
630 * The reference count may increase beyond zero at this point due
631 * to an erroneous descriptor reference by an application, but
632 * fd_getfile() will notice that the file is being closed and drop
633 * the reference again.
634 */
635 if (fdp->fd_refcnt == 1) {
636 /* Single threaded. */
637 refcnt = --(ff->ff_refcnt);
638 } else {
639 /* Multi threaded. */
640#ifndef __HAVE_ATOMIC_AS_MEMBAR
641 membar_producer();
642#endif
643 refcnt = atomic_dec_uint_nv(&ff->ff_refcnt);
644 }
645 if (__predict_false(refcnt != 0)) {
646 /*
647 * Wait for other references to drain. This is typically
648 * an application error - the descriptor is being closed
649 * while still in use.
650 * (Or just a threaded application trying to unblock its
651 * thread that sleeps in (say) accept()).
652 */
653 atomic_or_uint(&ff->ff_refcnt, FR_CLOSING);
654
655 /*
656 * Remove any knotes attached to the file. A knote
657 * attached to the descriptor can hold references on it.
658 */
659 mutex_exit(&fdp->fd_lock);
660 if (!SLIST_EMPTY(&ff->ff_knlist)) {
661 knote_fdclose(fd);
662 }
663
664 /*
665 * Since the file system code doesn't know which fd
666 * each request came from (think dup()), we have to
667 * ask it to return ERESTART for any long-term blocks.
668 * The re-entry through read/write/etc will detect the
669 * closed fd and return EBAFD.
670 * Blocked partial writes may return a short length.
671 */
672 (*fp->f_ops->fo_restart)(fp);
673 mutex_enter(&fdp->fd_lock);
674
675 /*
676 * We need to see the count drop to zero at least once,
677 * in order to ensure that all pre-existing references
678 * have been drained. New references past this point are
679 * of no interest.
680 * XXX (dsl) this may need to call fo_restart() after a
681 * timeout to guarantee that all the system calls exit.
682 */
683 while ((ff->ff_refcnt & FR_MASK) != 0) {
684 cv_wait(&ff->ff_closing, &fdp->fd_lock);
685 }
686 atomic_and_uint(&ff->ff_refcnt, ~FR_CLOSING);
687 } else {
688 /* If no references, there must be no knotes. */
689 KASSERT(SLIST_EMPTY(&ff->ff_knlist));
690 }
691
692 /*
693 * POSIX record locking dictates that any close releases ALL
694 * locks owned by this process. This is handled by setting
695 * a flag in the unlock to free ONLY locks obeying POSIX
696 * semantics, and not to free BSD-style file locks.
697 * If the descriptor was in a message, POSIX-style locks
698 * aren't passed with the descriptor.
699 */
700 if (__predict_false((p->p_flag & PK_ADVLOCK) != 0 &&
701 fp->f_type == DTYPE_VNODE)) {
702 lf.l_whence = SEEK_SET;
703 lf.l_start = 0;
704 lf.l_len = 0;
705 lf.l_type = F_UNLCK;
706 mutex_exit(&fdp->fd_lock);
707 (void)VOP_ADVLOCK(fp->f_vnode, p, F_UNLCK, &lf, F_POSIX);
708 mutex_enter(&fdp->fd_lock);
709 }
710
711 /* Free descriptor slot. */
712 fd_unused(fdp, fd);
713 mutex_exit(&fdp->fd_lock);
714
715 /* Now drop reference to the file itself. */
716 return closef(fp);
717}
718
719/*
720 * Duplicate a file descriptor.
721 */
722int
723fd_dup(file_t *fp, int minfd, int *newp, bool exclose)
724{
725 proc_t *p = curproc;
726 int error;
727
728 while ((error = fd_alloc(p, minfd, newp)) != 0) {
729 if (error != ENOSPC) {
730 return error;
731 }
732 fd_tryexpand(p);
733 }
734
735 curlwp->l_fd->fd_dt->dt_ff[*newp]->ff_exclose = exclose;
736 fd_affix(p, fp, *newp);
737 return 0;
738}
739
740/*
741 * dup2 operation.
742 */
743int
744fd_dup2(file_t *fp, unsigned newfd, int flags)
745{
746 filedesc_t *fdp = curlwp->l_fd;
747 fdfile_t *ff;
748 fdtab_t *dt;
749
750 if (flags & ~(O_CLOEXEC|O_NONBLOCK))
751 return EINVAL;
752 /*
753 * Ensure there are enough slots in the descriptor table,
754 * and allocate an fdfile_t up front in case we need it.
755 */
756 while (newfd >= fdp->fd_dt->dt_nfiles) {
757 fd_tryexpand(curproc);
758 }
759 ff = pool_cache_get(fdfile_cache, PR_WAITOK);
760
761 /*
762 * If there is already a file open, close it. If the file is
763 * half open, wait for it to be constructed before closing it.
764 * XXX Potential for deadlock here?
765 */
766 mutex_enter(&fdp->fd_lock);
767 while (fd_isused(fdp, newfd)) {
768 mutex_exit(&fdp->fd_lock);
769 if (fd_getfile(newfd) != NULL) {
770 (void)fd_close(newfd);
771 } else {
772 /*
773 * Crummy, but unlikely to happen.
774 * Can occur if we interrupt another
775 * thread while it is opening a file.
776 */
777 kpause("dup2", false, 1, NULL);
778 }
779 mutex_enter(&fdp->fd_lock);
780 }
781 dt = fdp->fd_dt;
782 if (dt->dt_ff[newfd] == NULL) {
783 KASSERT(newfd >= NDFDFILE);
784 dt->dt_ff[newfd] = ff;
785 ff = NULL;
786 }
787 fd_used(fdp, newfd);
788 mutex_exit(&fdp->fd_lock);
789
790 dt->dt_ff[newfd]->ff_exclose = (flags & O_CLOEXEC) != 0;
791 fp->f_flag |= flags & FNONBLOCK;
792 /* Slot is now allocated. Insert copy of the file. */
793 fd_affix(curproc, fp, newfd);
794 if (ff != NULL) {
795 pool_cache_put(fdfile_cache, ff);
796 }
797 return 0;
798}
799
800/*
801 * Drop reference to a file structure.
802 */
803int
804closef(file_t *fp)
805{
806 struct flock lf;
807 int error;
808
809 /*
810 * Drop reference. If referenced elsewhere it's still open
811 * and we have nothing more to do.
812 */
813 mutex_enter(&fp->f_lock);
814 KASSERT(fp->f_count > 0);
815 if (--fp->f_count > 0) {
816 mutex_exit(&fp->f_lock);
817 return 0;
818 }
819 KASSERT(fp->f_count == 0);
820 mutex_exit(&fp->f_lock);
821
822 /* We held the last reference - release locks, close and free. */
823 if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
824 lf.l_whence = SEEK_SET;
825 lf.l_start = 0;
826 lf.l_len = 0;
827 lf.l_type = F_UNLCK;
828 (void)VOP_ADVLOCK(fp->f_vnode, fp, F_UNLCK, &lf, F_FLOCK);
829 }
830 if (fp->f_ops != NULL) {
831 error = (*fp->f_ops->fo_close)(fp);
832 } else {
833 error = 0;
834 }
835 KASSERT(fp->f_count == 0);
836 KASSERT(fp->f_cred != NULL);
837 pool_cache_put(file_cache, fp);
838
839 return error;
840}
841
842/*
843 * Allocate a file descriptor for the process.
844 */
845int
846fd_alloc(proc_t *p, int want, int *result)
847{
848 filedesc_t *fdp = p->p_fd;
849 int i, lim, last, error, hi;
850 u_int off;
851 fdtab_t *dt;
852
853 KASSERT(p == curproc || p == &proc0);
854
855 /*
856 * Search for a free descriptor starting at the higher
857 * of want or fd_freefile.
858 */
859 mutex_enter(&fdp->fd_lock);
860 fd_checkmaps(fdp);
861 dt = fdp->fd_dt;
862 KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
863 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
864 last = min(dt->dt_nfiles, lim);
865 for (;;) {
866 if ((i = want) < fdp->fd_freefile)
867 i = fdp->fd_freefile;
868 off = i >> NDENTRYSHIFT;
869 hi = fd_next_zero(fdp, fdp->fd_himap, off,
870 (last + NDENTRIES - 1) >> NDENTRYSHIFT);
871 if (hi == -1)
872 break;
873 i = fd_next_zero(fdp, &fdp->fd_lomap[hi],
874 hi > off ? 0 : i & NDENTRYMASK, NDENTRIES);
875 if (i == -1) {
876 /*
877 * Free file descriptor in this block was
878 * below want, try again with higher want.
879 */
880 want = (hi + 1) << NDENTRYSHIFT;
881 continue;
882 }
883 i += (hi << NDENTRYSHIFT);
884 if (i >= last) {
885 break;
886 }
887 if (dt->dt_ff[i] == NULL) {
888 KASSERT(i >= NDFDFILE);
889 dt->dt_ff[i] = pool_cache_get(fdfile_cache, PR_WAITOK);
890 }
891 KASSERT(dt->dt_ff[i]->ff_file == NULL);
892 fd_used(fdp, i);
893 if (want <= fdp->fd_freefile) {
894 fdp->fd_freefile = i;
895 }
896 *result = i;
897 KASSERT(i >= NDFDFILE ||
898 dt->dt_ff[i] == (fdfile_t *)fdp->fd_dfdfile[i]);
899 fd_checkmaps(fdp);
900 mutex_exit(&fdp->fd_lock);
901 return 0;
902 }
903
904 /* No space in current array. Let the caller expand and retry. */
905 error = (dt->dt_nfiles >= lim) ? EMFILE : ENOSPC;
906 mutex_exit(&fdp->fd_lock);
907 return error;
908}
909
910/*
911 * Allocate memory for a descriptor table.
912 */
913static fdtab_t *
914fd_dtab_alloc(int n)
915{
916 fdtab_t *dt;
917 size_t sz;
918
919 KASSERT(n > NDFILE);
920
921 sz = sizeof(*dt) + (n - NDFILE) * sizeof(dt->dt_ff[0]);
922 dt = kmem_alloc(sz, KM_SLEEP);
923#ifdef DIAGNOSTIC
924 memset(dt, 0xff, sz);
925#endif
926 dt->dt_nfiles = n;
927 dt->dt_link = NULL;
928 return dt;
929}
930
931/*
932 * Free a descriptor table, and all tables linked for deferred free.
933 */
934static void
935fd_dtab_free(fdtab_t *dt)
936{
937 fdtab_t *next;
938 size_t sz;
939
940 do {
941 next = dt->dt_link;
942 KASSERT(dt->dt_nfiles > NDFILE);
943 sz = sizeof(*dt) +
944 (dt->dt_nfiles - NDFILE) * sizeof(dt->dt_ff[0]);
945#ifdef DIAGNOSTIC
946 memset(dt, 0xff, sz);
947#endif
948 kmem_free(dt, sz);
949 dt = next;
950 } while (dt != NULL);
951}
952
953/*
954 * Allocate descriptor bitmap.
955 */
956static void
957fd_map_alloc(int n, uint32_t **lo, uint32_t **hi)
958{
959 uint8_t *ptr;
960 size_t szlo, szhi;
961
962 KASSERT(n > NDENTRIES);
963
964 szlo = NDLOSLOTS(n) * sizeof(uint32_t);
965 szhi = NDHISLOTS(n) * sizeof(uint32_t);
966 ptr = kmem_alloc(szlo + szhi, KM_SLEEP);
967 *lo = (uint32_t *)ptr;
968 *hi = (uint32_t *)(ptr + szlo);
969}
970
971/*
972 * Free descriptor bitmap.
973 */
974static void
975fd_map_free(int n, uint32_t *lo, uint32_t *hi)
976{
977 size_t szlo, szhi;
978
979 KASSERT(n > NDENTRIES);
980
981 szlo = NDLOSLOTS(n) * sizeof(uint32_t);
982 szhi = NDHISLOTS(n) * sizeof(uint32_t);
983 KASSERT(hi == (uint32_t *)((uint8_t *)lo + szlo));
984 kmem_free(lo, szlo + szhi);
985}
986
987/*
988 * Expand a process' descriptor table.
989 */
990void
991fd_tryexpand(proc_t *p)
992{
993 filedesc_t *fdp;
994 int i, numfiles, oldnfiles;
995 fdtab_t *newdt, *dt;
996 uint32_t *newhimap, *newlomap;
997
998 KASSERT(p == curproc || p == &proc0);
999
1000 fdp = p->p_fd;
1001 newhimap = NULL;
1002 newlomap = NULL;
1003 oldnfiles = fdp->fd_dt->dt_nfiles;
1004
1005 if (oldnfiles < NDEXTENT)
1006 numfiles = NDEXTENT;
1007 else
1008 numfiles = 2 * oldnfiles;
1009
1010 newdt = fd_dtab_alloc(numfiles);
1011 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
1012 fd_map_alloc(numfiles, &newlomap, &newhimap);
1013 }
1014
1015 mutex_enter(&fdp->fd_lock);
1016 dt = fdp->fd_dt;
1017 KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
1018 if (dt->dt_nfiles != oldnfiles) {
1019 /* fdp changed; caller must retry */
1020 mutex_exit(&fdp->fd_lock);
1021 fd_dtab_free(newdt);
1022 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
1023 fd_map_free(numfiles, newlomap, newhimap);
1024 }
1025 return;
1026 }
1027
1028 /* Copy the existing descriptor table and zero the new portion. */
1029 i = sizeof(fdfile_t *) * oldnfiles;
1030 memcpy(newdt->dt_ff, dt->dt_ff, i);
1031 memset((uint8_t *)newdt->dt_ff + i, 0,
1032 numfiles * sizeof(fdfile_t *) - i);
1033
1034 /*
1035 * Link old descriptor array into list to be discarded. We defer
1036 * freeing until the last reference to the descriptor table goes
1037 * away (usually process exit). This allows us to do lockless
1038 * lookups in fd_getfile().
1039 */
1040 if (oldnfiles > NDFILE) {
1041 if (fdp->fd_refcnt > 1) {
1042 newdt->dt_link = dt;
1043 } else {
1044 fd_dtab_free(dt);
1045 }
1046 }
1047
1048 if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
1049 i = NDHISLOTS(oldnfiles) * sizeof(uint32_t);
1050 memcpy(newhimap, fdp->fd_himap, i);
1051 memset((uint8_t *)newhimap + i, 0,
1052 NDHISLOTS(numfiles) * sizeof(uint32_t) - i);
1053
1054 i = NDLOSLOTS(oldnfiles) * sizeof(uint32_t);
1055 memcpy(newlomap, fdp->fd_lomap, i);
1056 memset((uint8_t *)newlomap + i, 0,
1057 NDLOSLOTS(numfiles) * sizeof(uint32_t) - i);
1058
1059 if (NDHISLOTS(oldnfiles) > NDHISLOTS(NDFILE)) {
1060 fd_map_free(oldnfiles, fdp->fd_lomap, fdp->fd_himap);
1061 }
1062 fdp->fd_himap = newhimap;
1063 fdp->fd_lomap = newlomap;
1064 }
1065
1066 /*
1067 * All other modifications must become globally visible before
1068 * the change to fd_dt. See fd_getfile().
1069 */
1070 membar_producer();
1071 fdp->fd_dt = newdt;
1072 KASSERT(newdt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
1073 fd_checkmaps(fdp);
1074 mutex_exit(&fdp->fd_lock);
1075}
1076
1077/*
1078 * Create a new open file structure and allocate a file descriptor
1079 * for the current process.
1080 */
1081int
1082fd_allocfile(file_t **resultfp, int *resultfd)
1083{
1084 proc_t *p = curproc;
1085 kauth_cred_t cred;
1086 file_t *fp;
1087 int error;
1088
1089 while ((error = fd_alloc(p, 0, resultfd)) != 0) {
1090 if (error != ENOSPC) {
1091 return error;
1092 }
1093 fd_tryexpand(p);
1094 }
1095
1096 fp = pool_cache_get(file_cache, PR_WAITOK);
1097 if (fp == NULL) {
1098 fd_abort(p, NULL, *resultfd);
1099 return ENFILE;
1100 }
1101 KASSERT(fp->f_count == 0);
1102 KASSERT(fp->f_msgcount == 0);
1103 KASSERT(fp->f_unpcount == 0);
1104
1105 /* Replace cached credentials if not what we need. */
1106 cred = curlwp->l_cred;
1107 if (__predict_false(cred != fp->f_cred)) {
1108 kauth_cred_free(fp->f_cred);
1109 kauth_cred_hold(cred);
1110 fp->f_cred = cred;
1111 }
1112
1113 /*
1114 * Don't allow recycled files to be scanned.
1115 * See uipc_usrreq.c.
1116 */
1117 if (__predict_false((fp->f_flag & FSCAN) != 0)) {
1118 mutex_enter(&fp->f_lock);
1119 atomic_and_uint(&fp->f_flag, ~FSCAN);
1120 mutex_exit(&fp->f_lock);
1121 }
1122
1123 fp->f_advice = 0;
1124 fp->f_offset = 0;
1125 *resultfp = fp;
1126
1127 return 0;
1128}
1129
1130/*
1131 * Successful creation of a new descriptor: make visible to the process.
1132 */
1133void
1134fd_affix(proc_t *p, file_t *fp, unsigned fd)
1135{
1136 fdfile_t *ff;
1137 filedesc_t *fdp;
1138
1139 KASSERT(p == curproc || p == &proc0);
1140
1141 /* Add a reference to the file structure. */
1142 mutex_enter(&fp->f_lock);
1143 fp->f_count++;
1144 mutex_exit(&fp->f_lock);
1145
1146 /*
1147 * Insert the new file into the descriptor slot.
1148 *
1149 * The memory barriers provided by lock activity in this routine
1150 * ensure that any updates to the file structure become globally
1151 * visible before the file becomes visible to other LWPs in the
1152 * current process.
1153 */
1154 fdp = p->p_fd;
1155 ff = fdp->fd_dt->dt_ff[fd];
1156
1157 KASSERT(ff != NULL);
1158 KASSERT(ff->ff_file == NULL);
1159 KASSERT(ff->ff_allocated);
1160 KASSERT(fd_isused(fdp, fd));
1161 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
1162
1163 /* No need to lock in order to make file initially visible. */
1164 ff->ff_file = fp;
1165}
1166
1167/*
1168 * Abort creation of a new descriptor: free descriptor slot and file.
1169 */
1170void
1171fd_abort(proc_t *p, file_t *fp, unsigned fd)
1172{
1173 filedesc_t *fdp;
1174 fdfile_t *ff;
1175
1176 KASSERT(p == curproc || p == &proc0);
1177
1178 fdp = p->p_fd;
1179 ff = fdp->fd_dt->dt_ff[fd];
1180 ff->ff_exclose = false;
1181
1182 KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
1183
1184 mutex_enter(&fdp->fd_lock);
1185 KASSERT(fd_isused(fdp, fd));
1186 fd_unused(fdp, fd);
1187 mutex_exit(&fdp->fd_lock);
1188
1189 if (fp != NULL) {
1190 KASSERT(fp->f_count == 0);
1191 KASSERT(fp->f_cred != NULL);
1192 pool_cache_put(file_cache, fp);
1193 }
1194}
1195
1196static int
1197file_ctor(void *arg, void *obj, int flags)
1198{
1199 file_t *fp = obj;
1200
1201 memset(fp, 0, sizeof(*fp));
1202
1203 mutex_enter(&filelist_lock);
1204 if (__predict_false(nfiles >= maxfiles)) {
1205 mutex_exit(&filelist_lock);
1206 tablefull("file", "increase kern.maxfiles or MAXFILES");
1207 return ENFILE;
1208 }
1209 nfiles++;
1210 LIST_INSERT_HEAD(&filehead, fp, f_list);
1211 mutex_init(&fp->f_lock, MUTEX_DEFAULT, IPL_NONE);
1212 fp->f_cred = curlwp->l_cred;
1213 kauth_cred_hold(fp->f_cred);
1214 mutex_exit(&filelist_lock);
1215
1216 return 0;
1217}
1218
1219static void
1220file_dtor(void *arg, void *obj)
1221{
1222 file_t *fp = obj;
1223
1224 mutex_enter(&filelist_lock);
1225 nfiles--;
1226 LIST_REMOVE(fp, f_list);
1227 mutex_exit(&filelist_lock);
1228
1229 kauth_cred_free(fp->f_cred);
1230 mutex_destroy(&fp->f_lock);
1231}
1232
1233static int
1234fdfile_ctor(void *arg, void *obj, int flags)
1235{
1236 fdfile_t *ff = obj;
1237
1238 memset(ff, 0, sizeof(*ff));
1239 cv_init(&ff->ff_closing, "fdclose");
1240
1241 return 0;
1242}
1243
1244static void
1245fdfile_dtor(void *arg, void *obj)
1246{
1247 fdfile_t *ff = obj;
1248
1249 cv_destroy(&ff->ff_closing);
1250}
1251
1252file_t *
1253fgetdummy(void)
1254{
1255 file_t *fp;
1256
1257 fp = kmem_zalloc(sizeof(*fp), KM_SLEEP);
1258 if (fp != NULL) {
1259 mutex_init(&fp->f_lock, MUTEX_DEFAULT, IPL_NONE);
1260 }
1261 return fp;
1262}
1263
1264void
1265fputdummy(file_t *fp)
1266{
1267
1268 mutex_destroy(&fp->f_lock);
1269 kmem_free(fp, sizeof(*fp));
1270}
1271
1272/*
1273 * Create an initial filedesc structure.
1274 */
1275filedesc_t *
1276fd_init(filedesc_t *fdp)
1277{
1278#ifdef DIAGNOSTIC
1279 unsigned fd;
1280#endif
1281
1282 if (__predict_true(fdp == NULL)) {
1283 fdp = pool_cache_get(filedesc_cache, PR_WAITOK);
1284 } else {
1285 KASSERT(fdp == &filedesc0);
1286 filedesc_ctor(NULL, fdp, PR_WAITOK);
1287 }
1288
1289#ifdef DIAGNOSTIC
1290 KASSERT(fdp->fd_lastfile == -1);
1291 KASSERT(fdp->fd_lastkqfile == -1);
1292 KASSERT(fdp->fd_knhash == NULL);
1293 KASSERT(fdp->fd_freefile == 0);
1294 KASSERT(fdp->fd_exclose == false);
1295 KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin);
1296 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE);
1297 for (fd = 0; fd < NDFDFILE; fd++) {
1298 KASSERT(fdp->fd_dtbuiltin.dt_ff[fd] ==
1299 (fdfile_t *)fdp->fd_dfdfile[fd]);
1300 }
1301 for (fd = NDFDFILE; fd < NDFILE; fd++) {
1302 KASSERT(fdp->fd_dtbuiltin.dt_ff[fd] == NULL);
1303 }
1304 KASSERT(fdp->fd_himap == fdp->fd_dhimap);
1305 KASSERT(fdp->fd_lomap == fdp->fd_dlomap);
1306#endif /* DIAGNOSTIC */
1307
1308 fdp->fd_refcnt = 1;
1309 fd_checkmaps(fdp);
1310
1311 return fdp;
1312}
1313
1314/*
1315 * Initialize a file descriptor table.
1316 */
1317static int
1318filedesc_ctor(void *arg, void *obj, int flag)
1319{
1320 filedesc_t *fdp = obj;
1321 fdfile_t **ffp;
1322 int i;
1323
1324 memset(fdp, 0, sizeof(*fdp));
1325 mutex_init(&fdp->fd_lock, MUTEX_DEFAULT, IPL_NONE);
1326 fdp->fd_lastfile = -1;
1327 fdp->fd_lastkqfile = -1;
1328 fdp->fd_dt = &fdp->fd_dtbuiltin;
1329 fdp->fd_dtbuiltin.dt_nfiles = NDFILE;
1330 fdp->fd_himap = fdp->fd_dhimap;
1331 fdp->fd_lomap = fdp->fd_dlomap;
1332
1333 CTASSERT(sizeof(fdp->fd_dfdfile[0]) >= sizeof(fdfile_t));
1334 for (i = 0, ffp = fdp->fd_dt->dt_ff; i < NDFDFILE; i++, ffp++) {
1335 *ffp = (fdfile_t *)fdp->fd_dfdfile[i];
1336 (void)fdfile_ctor(NULL, fdp->fd_dfdfile[i], PR_WAITOK);
1337 }
1338
1339 return 0;
1340}
1341
1342static void
1343filedesc_dtor(void *arg, void *obj)
1344{
1345 filedesc_t *fdp = obj;
1346 int i;
1347
1348 for (i = 0; i < NDFDFILE; i++) {
1349 fdfile_dtor(NULL, fdp->fd_dfdfile[i]);
1350 }
1351
1352 mutex_destroy(&fdp->fd_lock);
1353}
1354
1355/*
1356 * Make p share curproc's filedesc structure.
1357 */
1358void
1359fd_share(struct proc *p)
1360{
1361 filedesc_t *fdp;
1362
1363 fdp = curlwp->l_fd;
1364 p->p_fd = fdp;
1365 atomic_inc_uint(&fdp->fd_refcnt);
1366}
1367
1368/*
1369 * Acquire a hold on a filedesc structure.
1370 */
1371void
1372fd_hold(lwp_t *l)
1373{
1374 filedesc_t *fdp = l->l_fd;
1375
1376 atomic_inc_uint(&fdp->fd_refcnt);
1377}
1378
1379/*
1380 * Copy a filedesc structure.
1381 */
1382filedesc_t *
1383fd_copy(void)
1384{
1385 filedesc_t *newfdp, *fdp;
1386 fdfile_t *ff, **ffp, **nffp, *ff2;
1387 int i, j, numfiles, lastfile, newlast;
1388 file_t *fp;
1389 fdtab_t *newdt;
1390
1391 fdp = curproc->p_fd;
1392 newfdp = pool_cache_get(filedesc_cache, PR_WAITOK);
1393 newfdp->fd_refcnt = 1;
1394
1395#ifdef DIAGNOSTIC
1396 KASSERT(newfdp->fd_lastfile == -1);
1397 KASSERT(newfdp->fd_lastkqfile == -1);
1398 KASSERT(newfdp->fd_knhash == NULL);
1399 KASSERT(newfdp->fd_freefile == 0);
1400 KASSERT(newfdp->fd_exclose == false);
1401 KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin);
1402 KASSERT(newfdp->fd_dtbuiltin.dt_nfiles == NDFILE);
1403 for (i = 0; i < NDFDFILE; i++) {
1404 KASSERT(newfdp->fd_dtbuiltin.dt_ff[i] ==
1405 (fdfile_t *)&newfdp->fd_dfdfile[i]);
1406 }
1407 for (i = NDFDFILE; i < NDFILE; i++) {
1408 KASSERT(newfdp->fd_dtbuiltin.dt_ff[i] == NULL);
1409 }
1410#endif /* DIAGNOSTIC */
1411
1412 mutex_enter(&fdp->fd_lock);
1413 fd_checkmaps(fdp);
1414 numfiles = fdp->fd_dt->dt_nfiles;
1415 lastfile = fdp->fd_lastfile;
1416
1417 /*
1418 * If the number of open files fits in the internal arrays
1419 * of the open file structure, use them, otherwise allocate
1420 * additional memory for the number of descriptors currently
1421 * in use.
1422 */
1423 if (lastfile < NDFILE) {
1424 i = NDFILE;
1425 newdt = newfdp->fd_dt;
1426 KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin);
1427 } else {
1428 /*
1429 * Compute the smallest multiple of NDEXTENT needed
1430 * for the file descriptors currently in use,
1431 * allowing the table to shrink.
1432 */
1433 i = numfiles;
1434 while (i >= 2 * NDEXTENT && i > lastfile * 2) {
1435 i /= 2;
1436 }
1437 KASSERT(i > NDFILE);
1438 newdt = fd_dtab_alloc(i);
1439 newfdp->fd_dt = newdt;
1440 memcpy(newdt->dt_ff, newfdp->fd_dtbuiltin.dt_ff,
1441 NDFDFILE * sizeof(fdfile_t **));
1442 memset(newdt->dt_ff + NDFDFILE, 0,
1443 (i - NDFDFILE) * sizeof(fdfile_t **));
1444 }
1445 if (NDHISLOTS(i) <= NDHISLOTS(NDFILE)) {
1446 newfdp->fd_himap = newfdp->fd_dhimap;
1447 newfdp->fd_lomap = newfdp->fd_dlomap;
1448 } else {
1449 fd_map_alloc(i, &newfdp->fd_lomap, &newfdp->fd_himap);
1450 KASSERT(i >= NDENTRIES * NDENTRIES);
1451 memset(newfdp->fd_himap, 0, NDHISLOTS(i)*sizeof(uint32_t));
1452 memset(newfdp->fd_lomap, 0, NDLOSLOTS(i)*sizeof(uint32_t));
1453 }
1454 newfdp->fd_freefile = fdp->fd_freefile;
1455 newfdp->fd_exclose = fdp->fd_exclose;
1456
1457 ffp = fdp->fd_dt->dt_ff;
1458 nffp = newdt->dt_ff;
1459 newlast = -1;
1460 for (i = 0; i <= lastfile; i++, ffp++, nffp++) {
1461 KASSERT(i >= NDFDFILE ||
1462 *nffp == (fdfile_t *)newfdp->fd_dfdfile[i]);
1463 ff = *ffp;
1464 if (ff == NULL || (fp = ff->ff_file) == NULL) {
1465 /* Descriptor unused, or descriptor half open. */
1466 KASSERT(!fd_isused(newfdp, i));
1467 continue;
1468 }
1469 if (__predict_false(fp->f_type == DTYPE_KQUEUE)) {
1470 /* kqueue descriptors cannot be copied. */
1471 if (i < newfdp->fd_freefile) {
1472 newfdp->fd_freefile = i;
1473 }
1474 continue;
1475 }
1476 /* It's active: add a reference to the file. */
1477 mutex_enter(&fp->f_lock);
1478 fp->f_count++;
1479 mutex_exit(&fp->f_lock);
1480
1481 /* Allocate an fdfile_t to represent it. */
1482 if (i >= NDFDFILE) {
1483 ff2 = pool_cache_get(fdfile_cache, PR_WAITOK);
1484 *nffp = ff2;
1485 } else {
1486 ff2 = newdt->dt_ff[i];
1487 }
1488 ff2->ff_file = fp;
1489 ff2->ff_exclose = ff->ff_exclose;
1490 ff2->ff_allocated = true;
1491
1492 /* Fix up bitmaps. */
1493 j = i >> NDENTRYSHIFT;
1494 KASSERT((newfdp->fd_lomap[j] & (1 << (i & NDENTRYMASK))) == 0);
1495 newfdp->fd_lomap[j] |= 1 << (i & NDENTRYMASK);
1496 if (__predict_false(newfdp->fd_lomap[j] == ~0)) {
1497 KASSERT((newfdp->fd_himap[j >> NDENTRYSHIFT] &
1498 (1 << (j & NDENTRYMASK))) == 0);
1499 newfdp->fd_himap[j >> NDENTRYSHIFT] |=
1500 1 << (j & NDENTRYMASK);
1501 }
1502 newlast = i;
1503 }
1504 KASSERT(newdt->dt_ff[0] == (fdfile_t *)newfdp->fd_dfdfile[0]);
1505 newfdp->fd_lastfile = newlast;
1506 fd_checkmaps(newfdp);
1507 mutex_exit(&fdp->fd_lock);
1508
1509 return newfdp;
1510}
1511
1512/*
1513 * Release a filedesc structure.
1514 */
1515void
1516fd_free(void)
1517{
1518 fdfile_t *ff;
1519 file_t *fp;
1520 int fd, nf;
1521 fdtab_t *dt;
1522 lwp_t * const l = curlwp;
1523 filedesc_t * const fdp = l->l_fd;
1524 const bool noadvlock = (l->l_proc->p_flag & PK_ADVLOCK) == 0;
1525
1526 KASSERT(fdp->fd_dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
1527 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE);
1528 KASSERT(fdp->fd_dtbuiltin.dt_link == NULL);
1529
1530#ifndef __HAVE_ATOMIC_AS_MEMBAR
1531 membar_exit();
1532#endif
1533 if (atomic_dec_uint_nv(&fdp->fd_refcnt) > 0)
1534 return;
1535
1536 /*
1537 * Close any files that the process holds open.
1538 */
1539 dt = fdp->fd_dt;
1540 fd_checkmaps(fdp);
1541#ifdef DEBUG
1542 fdp->fd_refcnt = -1; /* see fd_checkmaps */
1543#endif
1544 for (fd = 0, nf = dt->dt_nfiles; fd < nf; fd++) {
1545 ff = dt->dt_ff[fd];
1546 KASSERT(fd >= NDFDFILE ||
1547 ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
1548 if (ff == NULL)
1549 continue;
1550 if ((fp = ff->ff_file) != NULL) {
1551 /*
1552 * Must use fd_close() here if there is
1553 * a reference from kqueue or we might have posix
1554 * advisory locks.
1555 */
1556 if (__predict_true(ff->ff_refcnt == 0) &&
1557 (noadvlock || fp->f_type != DTYPE_VNODE)) {
1558 ff->ff_file = NULL;
1559 ff->ff_exclose = false;
1560 ff->ff_allocated = false;
1561 closef(fp);
1562 } else {
1563 ff->ff_refcnt++;
1564 fd_close(fd);
1565 }
1566 }
1567 KASSERT(ff->ff_refcnt == 0);
1568 KASSERT(ff->ff_file == NULL);
1569 KASSERT(!ff->ff_exclose);
1570 KASSERT(!ff->ff_allocated);
1571 if (fd >= NDFDFILE) {
1572 pool_cache_put(fdfile_cache, ff);
1573 dt->dt_ff[fd] = NULL;
1574 }
1575 }
1576
1577 /*
1578 * Clean out the descriptor table for the next user and return
1579 * to the cache.
1580 */
1581 if (__predict_false(dt != &fdp->fd_dtbuiltin)) {
1582 fd_dtab_free(fdp->fd_dt);
1583 /* Otherwise, done above. */
1584 memset(&fdp->fd_dtbuiltin.dt_ff[NDFDFILE], 0,
1585 (NDFILE - NDFDFILE) * sizeof(fdp->fd_dtbuiltin.dt_ff[0]));
1586 fdp->fd_dt = &fdp->fd_dtbuiltin;
1587 }
1588 if (__predict_false(NDHISLOTS(nf) > NDHISLOTS(NDFILE))) {
1589 KASSERT(fdp->fd_himap != fdp->fd_dhimap);
1590 KASSERT(fdp->fd_lomap != fdp->fd_dlomap);
1591 fd_map_free(nf, fdp->fd_lomap, fdp->fd_himap);
1592 }
1593 if (__predict_false(fdp->fd_knhash != NULL)) {
1594 hashdone(fdp->fd_knhash, HASH_LIST, fdp->fd_knhashmask);
1595 fdp->fd_knhash = NULL;
1596 fdp->fd_knhashmask = 0;
1597 } else {
1598 KASSERT(fdp->fd_knhashmask == 0);
1599 }
1600 fdp->fd_dt = &fdp->fd_dtbuiltin;
1601 fdp->fd_lastkqfile = -1;
1602 fdp->fd_lastfile = -1;
1603 fdp->fd_freefile = 0;
1604 fdp->fd_exclose = false;
1605 memset(&fdp->fd_startzero, 0, sizeof(*fdp) -
1606 offsetof(filedesc_t, fd_startzero));
1607 fdp->fd_himap = fdp->fd_dhimap;
1608 fdp->fd_lomap = fdp->fd_dlomap;
1609 KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE);
1610 KASSERT(fdp->fd_dtbuiltin.dt_link == NULL);
1611 KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin);
1612#ifdef DEBUG
1613 fdp->fd_refcnt = 0; /* see fd_checkmaps */
1614#endif
1615 fd_checkmaps(fdp);
1616 pool_cache_put(filedesc_cache, fdp);
1617}
1618
1619/*
1620 * File Descriptor pseudo-device driver (/dev/fd/).
1621 *
1622 * Opening minor device N dup()s the file (if any) connected to file
1623 * descriptor N belonging to the calling process. Note that this driver
1624 * consists of only the ``open()'' routine, because all subsequent
1625 * references to this file will be direct to the other driver.
1626 */
1627static int
1628filedescopen(dev_t dev, int mode, int type, lwp_t *l)
1629{
1630
1631 /*
1632 * XXX Kludge: set dupfd to contain the value of the
1633 * the file descriptor being sought for duplication. The error
1634 * return ensures that the vnode for this device will be released
1635 * by vn_open. Open will detect this special error and take the
1636 * actions in fd_dupopen below. Other callers of vn_open or VOP_OPEN
1637 * will simply report the error.
1638 */
1639 l->l_dupfd = minor(dev); /* XXX */
1640 return EDUPFD;
1641}
1642
1643/*
1644 * Duplicate the specified descriptor to a free descriptor.
1645 */
1646int
1647fd_dupopen(int old, int *newp, int mode, int error)
1648{
1649 filedesc_t *fdp;
1650 fdfile_t *ff;
1651 file_t *fp;
1652 fdtab_t *dt;
1653
1654 if ((fp = fd_getfile(old)) == NULL) {
1655 return EBADF;
1656 }
1657 fdp = curlwp->l_fd;
1658 dt = fdp->fd_dt;
1659 ff = dt->dt_ff[old];
1660
1661 /*
1662 * There are two cases of interest here.
1663 *
1664 * For EDUPFD simply dup (old) to file descriptor
1665 * (new) and return.
1666 *
1667 * For EMOVEFD steal away the file structure from (old) and
1668 * store it in (new). (old) is effectively closed by
1669 * this operation.
1670 *
1671 * Any other error code is just returned.
1672 */
1673 switch (error) {
1674 case EDUPFD:
1675 /*
1676 * Check that the mode the file is being opened for is a
1677 * subset of the mode of the existing descriptor.
1678 */
1679 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
1680 error = EACCES;
1681 break;
1682 }
1683
1684 /* Copy it. */
1685 error = fd_dup(fp, 0, newp, ff->ff_exclose);
1686 break;
1687
1688 case EMOVEFD:
1689 /* Copy it. */
1690 error = fd_dup(fp, 0, newp, ff->ff_exclose);
1691 if (error != 0) {
1692 break;
1693 }
1694
1695 /* Steal away the file pointer from 'old'. */
1696 (void)fd_close(old);
1697 return 0;
1698 }
1699
1700 fd_putfile(old);
1701 return error;
1702}
1703
1704/*
1705 * Close open files on exec.
1706 */
1707void
1708fd_closeexec(void)
1709{
1710 proc_t *p;
1711 filedesc_t *fdp;
1712 fdfile_t *ff;
1713 lwp_t *l;
1714 fdtab_t *dt;
1715 int fd;
1716
1717 l = curlwp;
1718 p = l->l_proc;
1719 fdp = p->p_fd;
1720
1721 if (fdp->fd_refcnt > 1) {
1722 fdp = fd_copy();
1723 fd_free();
1724 p->p_fd = fdp;
1725 l->l_fd = fdp;
1726 }
1727 if (!fdp->fd_exclose) {
1728 return;
1729 }
1730 fdp->fd_exclose = false;
1731 dt = fdp->fd_dt;
1732
1733 for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
1734 if ((ff = dt->dt_ff[fd]) == NULL) {
1735 KASSERT(fd >= NDFDFILE);
1736 continue;
1737 }
1738 KASSERT(fd >= NDFDFILE ||
1739 ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
1740 if (ff->ff_file == NULL)
1741 continue;
1742 if (ff->ff_exclose) {
1743 /*
1744 * We need a reference to close the file.
1745 * No other threads can see the fdfile_t at
1746 * this point, so don't bother locking.
1747 */
1748 KASSERT((ff->ff_refcnt & FR_CLOSING) == 0);
1749 ff->ff_refcnt++;
1750 fd_close(fd);
1751 }
1752 }
1753}
1754
1755/*
1756 * Sets descriptor owner. If the owner is a process, 'pgid'
1757 * is set to positive value, process ID. If the owner is process group,
1758 * 'pgid' is set to -pg_id.
1759 */
1760int
1761fsetown(pid_t *pgid, u_long cmd, const void *data)
1762{
1763 pid_t id = *(const pid_t *)data;
1764 int error;
1765
1766 switch (cmd) {
1767 case TIOCSPGRP:
1768 if (id < 0)
1769 return EINVAL;
1770 id = -id;
1771 break;
1772 default:
1773 break;
1774 }
1775 if (id > 0) {
1776 mutex_enter(proc_lock);
1777 error = proc_find(id) ? 0 : ESRCH;
1778 mutex_exit(proc_lock);
1779 } else if (id < 0) {
1780 error = pgid_in_session(curproc, -id);
1781 } else {
1782 error = 0;
1783 }
1784 if (!error) {
1785 *pgid = id;
1786 }
1787 return error;
1788}
1789
1790void
1791fd_set_exclose(struct lwp *l, int fd, bool exclose)
1792{
1793 filedesc_t *fdp = l->l_fd;
1794 fdfile_t *ff = fdp->fd_dt->dt_ff[fd];
1795
1796 ff->ff_exclose = exclose;
1797 if (exclose)
1798 fdp->fd_exclose = true;
1799}
1800
1801/*
1802 * Return descriptor owner information. If the value is positive,
1803 * it's process ID. If it's negative, it's process group ID and
1804 * needs the sign removed before use.
1805 */
1806int
1807fgetown(pid_t pgid, u_long cmd, void *data)
1808{
1809
1810 switch (cmd) {
1811 case TIOCGPGRP:
1812 *(int *)data = -pgid;
1813 break;
1814 default:
1815 *(int *)data = pgid;
1816 break;
1817 }
1818 return 0;
1819}
1820
1821/*
1822 * Send signal to descriptor owner, either process or process group.
1823 */
1824void
1825fownsignal(pid_t pgid, int signo, int code, int band, void *fdescdata)
1826{
1827 ksiginfo_t ksi;
1828
1829 KASSERT(!cpu_intr_p());
1830
1831 if (pgid == 0) {
1832 return;
1833 }
1834
1835 KSI_INIT(&ksi);
1836 ksi.ksi_signo = signo;
1837 ksi.ksi_code = code;
1838 ksi.ksi_band = band;
1839
1840 mutex_enter(proc_lock);
1841 if (pgid > 0) {
1842 struct proc *p1;
1843
1844 p1 = proc_find(pgid);
1845 if (p1 != NULL) {
1846 kpsignal(p1, &ksi, fdescdata);
1847 }
1848 } else {
1849 struct pgrp *pgrp;
1850
1851 KASSERT(pgid < 0);
1852 pgrp = pgrp_find(-pgid);
1853 if (pgrp != NULL) {
1854 kpgsignal(pgrp, &ksi, fdescdata, 0);
1855 }
1856 }
1857 mutex_exit(proc_lock);
1858}
1859
1860int
1861fd_clone(file_t *fp, unsigned fd, int flag, const struct fileops *fops,
1862 void *data)
1863{
1864 fdfile_t *ff;
1865 filedesc_t *fdp;
1866
1867 fp->f_flag |= flag & FMASK;
1868 fdp = curproc->p_fd;
1869 ff = fdp->fd_dt->dt_ff[fd];
1870 KASSERT(ff != NULL);
1871 ff->ff_exclose = (flag & O_CLOEXEC) != 0;
1872 fp->f_type = DTYPE_MISC;
1873 fp->f_ops = fops;
1874 fp->f_data = data;
1875 curlwp->l_dupfd = fd;
1876 fd_affix(curproc, fp, fd);
1877
1878 return EMOVEFD;
1879}
1880
1881int
1882fnullop_fcntl(file_t *fp, u_int cmd, void *data)
1883{
1884
1885 if (cmd == F_SETFL)
1886 return 0;
1887
1888 return EOPNOTSUPP;
1889}
1890
1891int
1892fnullop_poll(file_t *fp, int which)
1893{
1894
1895 return 0;
1896}
1897
1898int
1899fnullop_kqfilter(file_t *fp, struct knote *kn)
1900{
1901
1902 return EOPNOTSUPP;
1903}
1904
1905void
1906fnullop_restart(file_t *fp)
1907{
1908
1909}
1910
1911int
1912fbadop_read(file_t *fp, off_t *offset, struct uio *uio,
1913 kauth_cred_t cred, int flags)
1914{
1915
1916 return EOPNOTSUPP;
1917}
1918
1919int
1920fbadop_write(file_t *fp, off_t *offset, struct uio *uio,
1921 kauth_cred_t cred, int flags)
1922{
1923
1924 return EOPNOTSUPP;
1925}
1926
1927int
1928fbadop_ioctl(file_t *fp, u_long com, void *data)
1929{
1930
1931 return EOPNOTSUPP;
1932}
1933
1934int
1935fbadop_stat(file_t *fp, struct stat *sb)
1936{
1937
1938 return EOPNOTSUPP;
1939}
1940
1941int
1942fbadop_close(file_t *fp)
1943{
1944
1945 return EOPNOTSUPP;
1946}
1947
1948/*
1949 * sysctl routines pertaining to file descriptors
1950 */
1951
1952/* Initialized in sysctl_init() for now... */
1953extern kmutex_t sysctl_file_marker_lock;
1954static u_int sysctl_file_marker = 1;
1955
1956/*
1957 * Expects to be called with proc_lock and sysctl_file_marker_lock locked.
1958 */
1959static void
1960sysctl_file_marker_reset(void)
1961{
1962 struct proc *p;
1963
1964 PROCLIST_FOREACH(p, &allproc) {
1965 struct filedesc *fd = p->p_fd;
1966 fdtab_t *dt;
1967 u_int i;
1968
1969 mutex_enter(&fd->fd_lock);
1970 dt = fd->fd_dt;
1971 for (i = 0; i < dt->dt_nfiles; i++) {
1972 struct file *fp;
1973 fdfile_t *ff;
1974
1975 if ((ff = dt->dt_ff[i]) == NULL) {
1976 continue;
1977 }
1978 if ((fp = ff->ff_file) == NULL) {
1979 continue;
1980 }
1981 fp->f_marker = 0;
1982 }
1983 mutex_exit(&fd->fd_lock);
1984 }
1985}
1986
1987/*
1988 * sysctl helper routine for kern.file pseudo-subtree.
1989 */
1990static int
1991sysctl_kern_file(SYSCTLFN_ARGS)
1992{
1993 int error;
1994 size_t buflen;
1995 struct file *fp, fbuf;
1996 char *start, *where;
1997 struct proc *p;
1998
1999 start = where = oldp;
2000 buflen = *oldlenp;
2001
2002 if (where == NULL) {
2003 /*
2004 * overestimate by 10 files
2005 */
2006 *oldlenp = sizeof(filehead) + (nfiles + 10) *
2007 sizeof(struct file);
2008 return 0;
2009 }
2010
2011 /*
2012 * first sysctl_copyout filehead
2013 */
2014 if (buflen < sizeof(filehead)) {
2015 *oldlenp = 0;
2016 return 0;
2017 }
2018 sysctl_unlock();
2019 error = sysctl_copyout(l, &filehead, where, sizeof(filehead));
2020 if (error) {
2021 sysctl_relock();
2022 return error;
2023 }
2024 buflen -= sizeof(filehead);
2025 where += sizeof(filehead);
2026
2027 /*
2028 * followed by an array of file structures
2029 */
2030 mutex_enter(&sysctl_file_marker_lock);
2031 mutex_enter(proc_lock);
2032 PROCLIST_FOREACH(p, &allproc) {
2033 struct filedesc *fd;
2034 fdtab_t *dt;
2035 u_int i;
2036
2037 if (p->p_stat == SIDL) {
2038 /* skip embryonic processes */
2039 continue;
2040 }
2041 mutex_enter(p->p_lock);
2042 error = kauth_authorize_process(l->l_cred,
2043 KAUTH_PROCESS_CANSEE, p,
2044 KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES),
2045 NULL, NULL);
2046 mutex_exit(p->p_lock);
2047 if (error != 0) {
2048 /*
2049 * Don't leak kauth retval if we're silently
2050 * skipping this entry.
2051 */
2052 error = 0;
2053 continue;
2054 }
2055
2056 /*
2057 * Grab a hold on the process.
2058 */
2059 if (!rw_tryenter(&p->p_reflock, RW_READER)) {
2060 continue;
2061 }
2062 mutex_exit(proc_lock);
2063
2064 fd = p->p_fd;
2065 mutex_enter(&fd->fd_lock);
2066 dt = fd->fd_dt;
2067 for (i = 0; i < dt->dt_nfiles; i++) {
2068 fdfile_t *ff;
2069
2070 if ((ff = dt->dt_ff[i]) == NULL) {
2071 continue;
2072 }
2073 if ((fp = ff->ff_file) == NULL) {
2074 continue;
2075 }
2076
2077 mutex_enter(&fp->f_lock);
2078
2079 if ((fp->f_count == 0) ||
2080 (fp->f_marker == sysctl_file_marker)) {
2081 mutex_exit(&fp->f_lock);
2082 continue;
2083 }
2084
2085 /* Check that we have enough space. */
2086 if (buflen < sizeof(struct file)) {
2087 *oldlenp = where - start;
2088 mutex_exit(&fp->f_lock);
2089 error = ENOMEM;
2090 break;
2091 }
2092
2093 memcpy(&fbuf, fp, sizeof(fbuf));
2094 mutex_exit(&fp->f_lock);
2095 error = sysctl_copyout(l, &fbuf, where, sizeof(fbuf));
2096 if (error) {
2097 break;
2098 }
2099 buflen -= sizeof(struct file);
2100 where += sizeof(struct file);
2101
2102 fp->f_marker = sysctl_file_marker;
2103 }
2104 mutex_exit(&fd->fd_lock);
2105
2106 /*
2107 * Release reference to process.
2108 */
2109 mutex_enter(proc_lock);
2110 rw_exit(&p->p_reflock);
2111
2112 if (error)
2113 break;
2114 }
2115
2116 sysctl_file_marker++;
2117 /* Reset all markers if wrapped. */
2118 if (sysctl_file_marker == 0) {
2119 sysctl_file_marker_reset();
2120 sysctl_file_marker++;
2121 }
2122
2123 mutex_exit(proc_lock);
2124 mutex_exit(&sysctl_file_marker_lock);
2125
2126 *oldlenp = where - start;
2127 sysctl_relock();
2128 return error;
2129}
2130
2131/*
2132 * sysctl helper function for kern.file2
2133 */
2134static int
2135sysctl_kern_file2(SYSCTLFN_ARGS)
2136{
2137 struct proc *p;
2138 struct file *fp;
2139 struct filedesc *fd;
2140 struct kinfo_file kf;
2141 char *dp;
2142 u_int i, op;
2143 size_t len, needed, elem_size, out_size;
2144 int error, arg, elem_count;
2145 fdfile_t *ff;
2146 fdtab_t *dt;
2147
2148 if (namelen == 1 && name[0] == CTL_QUERY)
2149 return sysctl_query(SYSCTLFN_CALL(rnode));
2150
2151 if (namelen != 4)
2152 return EINVAL;
2153
2154 error = 0;
2155 dp = oldp;
2156 len = (oldp != NULL) ? *oldlenp : 0;
2157 op = name[0];
2158 arg = name[1];
2159 elem_size = name[2];
2160 elem_count = name[3];
2161 out_size = MIN(sizeof(kf), elem_size);
2162 needed = 0;
2163
2164 if (elem_size < 1 || elem_count < 0)
2165 return EINVAL;
2166
2167 switch (op) {
2168 case KERN_FILE_BYFILE:
2169 case KERN_FILE_BYPID:
2170 /*
2171 * We're traversing the process list in both cases; the BYFILE
2172 * case does additional work of keeping track of files already
2173 * looked at.
2174 */
2175
2176 /* doesn't use arg so it must be zero */
2177 if ((op == KERN_FILE_BYFILE) && (arg != 0))
2178 return EINVAL;
2179
2180 if ((op == KERN_FILE_BYPID) && (arg < -1))
2181 /* -1 means all processes */
2182 return EINVAL;
2183
2184 sysctl_unlock();
2185 if (op == KERN_FILE_BYFILE)
2186 mutex_enter(&sysctl_file_marker_lock);
2187 mutex_enter(proc_lock);
2188 PROCLIST_FOREACH(p, &allproc) {
2189 if (p->p_stat == SIDL) {
2190 /* skip embryonic processes */
2191 continue;
2192 }
2193 if (arg > 0 && p->p_pid != arg) {
2194 /* pick only the one we want */
2195 /* XXX want 0 to mean "kernel files" */
2196 continue;
2197 }
2198 mutex_enter(p->p_lock);
2199 error = kauth_authorize_process(l->l_cred,
2200 KAUTH_PROCESS_CANSEE, p,
2201 KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES),
2202 NULL, NULL);
2203 mutex_exit(p->p_lock);
2204 if (error != 0) {
2205 /*
2206 * Don't leak kauth retval if we're silently
2207 * skipping this entry.
2208 */
2209 error = 0;
2210 continue;
2211 }
2212
2213 /*
2214 * Grab a hold on the process.
2215 */
2216 if (!rw_tryenter(&p->p_reflock, RW_READER)) {
2217 continue;
2218 }
2219 mutex_exit(proc_lock);
2220
2221 fd = p->p_fd;
2222 mutex_enter(&fd->fd_lock);
2223 dt = fd->fd_dt;
2224 for (i = 0; i < dt->dt_nfiles; i++) {
2225 if ((ff = dt->dt_ff[i]) == NULL) {
2226 continue;
2227 }
2228 if ((fp = ff->ff_file) == NULL) {
2229 continue;
2230 }
2231
2232 if ((op == KERN_FILE_BYFILE) &&
2233 (fp->f_marker == sysctl_file_marker)) {
2234 continue;
2235 }
2236 if (len >= elem_size && elem_count > 0) {
2237 mutex_enter(&fp->f_lock);
2238 fill_file(&kf, fp, ff, i, p->p_pid);
2239 mutex_exit(&fp->f_lock);
2240 mutex_exit(&fd->fd_lock);
2241 error = sysctl_copyout(l,
2242 &kf, dp, out_size);
2243 mutex_enter(&fd->fd_lock);
2244 if (error)
2245 break;
2246 dp += elem_size;
2247 len -= elem_size;
2248 }
2249 if (op == KERN_FILE_BYFILE)
2250 fp->f_marker = sysctl_file_marker;
2251 needed += elem_size;
2252 if (elem_count > 0 && elem_count != INT_MAX)
2253 elem_count--;
2254 }
2255 mutex_exit(&fd->fd_lock);
2256
2257 /*
2258 * Release reference to process.
2259 */
2260 mutex_enter(proc_lock);
2261 rw_exit(&p->p_reflock);
2262 }
2263 if (op == KERN_FILE_BYFILE) {
2264 sysctl_file_marker++;
2265
2266 /* Reset all markers if wrapped. */
2267 if (sysctl_file_marker == 0) {
2268 sysctl_file_marker_reset();
2269 sysctl_file_marker++;
2270 }
2271 }
2272 mutex_exit(proc_lock);
2273 if (op == KERN_FILE_BYFILE)
2274 mutex_exit(&sysctl_file_marker_lock);
2275 sysctl_relock();
2276 break;
2277 default:
2278 return EINVAL;
2279 }
2280
2281 if (oldp == NULL)
2282 needed += KERN_FILESLOP * elem_size;
2283 *oldlenp = needed;
2284
2285 return error;
2286}
2287
2288static void
2289fill_file(struct kinfo_file *kp, const file_t *fp, const fdfile_t *ff,
2290 int i, pid_t pid)
2291{
2292
2293 memset(kp, 0, sizeof(*kp));
2294
2295 kp->ki_fileaddr = PTRTOUINT64(fp);
2296 kp->ki_flag = fp->f_flag;
2297 kp->ki_iflags = 0;
2298 kp->ki_ftype = fp->f_type;
2299 kp->ki_count = fp->f_count;
2300 kp->ki_msgcount = fp->f_msgcount;
2301 kp->ki_fucred = PTRTOUINT64(fp->f_cred);
2302 kp->ki_fuid = kauth_cred_geteuid(fp->f_cred);
2303 kp->ki_fgid = kauth_cred_getegid(fp->f_cred);
2304 kp->ki_fops = PTRTOUINT64(fp->f_ops);
2305 kp->ki_foffset = fp->f_offset;
2306 kp->ki_fdata = PTRTOUINT64(fp->f_data);
2307
2308 /* vnode information to glue this file to something */
2309 if (fp->f_type == DTYPE_VNODE) {
2310 struct vnode *vp = fp->f_vnode;
2311
2312 kp->ki_vun = PTRTOUINT64(vp->v_un.vu_socket);
2313 kp->ki_vsize = vp->v_size;
2314 kp->ki_vtype = vp->v_type;
2315 kp->ki_vtag = vp->v_tag;
2316 kp->ki_vdata = PTRTOUINT64(vp->v_data);
2317 }
2318
2319 /* process information when retrieved via KERN_FILE_BYPID */
2320 if (ff != NULL) {
2321 kp->ki_pid = pid;
2322 kp->ki_fd = i;
2323 kp->ki_ofileflags = ff->ff_exclose;
2324 kp->ki_usecount = ff->ff_refcnt;
2325 }
2326}
2327