]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/kern_descrip.c
projects / apple / xnu.git / blob
? search:


62a9d94f730065bccdf78dd1b77d895ad243f45d
[apple/xnu.git] / bsd / kern / kern_descrip.c
1 /*
2 * Copyright (c) 2000-2015 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /* Copyright (c) 1995, 1997 Apple Computer, Inc. All Rights Reserved */
29 /*
30 * Copyright (c) 1982, 1986, 1989, 1991, 1993
31 * The Regents of the University of California. All rights reserved.
32 * (c) UNIX System Laboratories, Inc.
33 * All or some portions of this file are derived from material licensed
34 * to the University of California by American Telephone and Telegraph
35 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
36 * the permission of UNIX System Laboratories, Inc.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * @(#)kern_descrip.c 8.8 (Berkeley) 2/14/95
67 */
68 /*
69 * NOTICE: This file was modified by SPARTA, Inc. in 2006 to introduce
70 * support for mandatory and extensible security protections. This notice
71 * is included in support of clause 2.2 (b) of the Apple Public License,
72 * Version 2.0.
73 */
74
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/filedesc.h>
78 #include <sys/kernel.h>
79 #include <sys/vnode_internal.h>
80 #include <sys/proc_internal.h>
81 #include <sys/kauth.h>
82 #include <sys/file_internal.h>
83 #include <sys/guarded.h>
84 #include <sys/priv.h>
85 #include <sys/socket.h>
86 #include <sys/socketvar.h>
87 #include <sys/stat.h>
88 #include <sys/ioctl.h>
89 #include <sys/fcntl.h>
90 #include <sys/malloc.h>
91 #include <sys/mman.h>
92 #include <sys/syslog.h>
93 #include <sys/unistd.h>
94 #include <sys/resourcevar.h>
95 #include <sys/aio_kern.h>
96 #include <sys/ev.h>
97 #include <kern/locks.h>
98 #include <sys/uio_internal.h>
99 #include <sys/codesign.h>
100 #include <sys/codedir_internal.h>
101
102 #include <security/audit/audit.h>
103
104 #include <sys/mount_internal.h>
105 #include <sys/kdebug.h>
106 #include <sys/sysproto.h>
107 #include <sys/pipe.h>
108 #include <sys/spawn.h>
109 #include <kern/kern_types.h>
110 #include <kern/kalloc.h>
111 #include <kern/waitq.h>
112 #include <libkern/OSAtomic.h>
113
114 #include <sys/ubc_internal.h>
115
116 #include <kern/ipc_misc.h>
117 #include <vm/vm_protos.h>
118
119 #include <mach/mach_port.h>
120 #include <stdbool.h>
121
122 #include <hfs/hfs.h>
123
124 kern_return_t ipc_object_copyin(ipc_space_t, mach_port_name_t,
125 mach_msg_type_name_t, ipc_port_t *);
126 void ipc_port_release_send(ipc_port_t);
127
128 struct psemnode;
129 struct pshmnode;
130
131 static int finishdup(proc_t p,
132 struct filedesc *fdp, int old, int new, int flags, int32_t *retval);
133
134 int falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx, int locked);
135 void fg_drop(struct fileproc * fp);
136 void fg_free(struct fileglob *fg);
137 void fg_ref(struct fileproc * fp);
138 void fileport_releasefg(struct fileglob *fg);
139
140 /* flags for close_internal_locked */
141 #define FD_DUP2RESV 1
142
143 /* We don't want these exported */
144
145 __private_extern__
146 int unlink1(vfs_context_t, vnode_t, user_addr_t, enum uio_seg, int);
147
148 static void _fdrelse(struct proc * p, int fd);
149
150
151 extern void file_lock_init(void);
152
153 extern kauth_scope_t kauth_scope_fileop;
154
155 /* Conflict wait queue for when selects collide (opaque type) */
156 extern struct waitq select_conflict_queue;
157
158 #define f_flag f_fglob->fg_flag
159 #define f_type f_fglob->fg_ops->fo_type
160 #define f_msgcount f_fglob->fg_msgcount
161 #define f_cred f_fglob->fg_cred
162 #define f_ops f_fglob->fg_ops
163 #define f_offset f_fglob->fg_offset
164 #define f_data f_fglob->fg_data
165 #define CHECK_ADD_OVERFLOW_INT64L(x, y) \
166 (((((x) > 0) && ((y) > 0) && ((x) > LLONG_MAX - (y))) || \
167 (((x) < 0) && ((y) < 0) && ((x) < LLONG_MIN - (y)))) \
168 ? 1 : 0)
169 /*
170 * Descriptor management.
171 */
172 struct fmsglist fmsghead; /* head of list of open files */
173 struct fmsglist fmsg_ithead; /* head of list of open files */
174 int nfiles; /* actual number of open files */
175
176
177 lck_grp_attr_t * file_lck_grp_attr;
178 lck_grp_t * file_lck_grp;
179 lck_attr_t * file_lck_attr;
180
181 lck_mtx_t * uipc_lock;
182
183
184 /*
185 * check_file_seek_range
186 *
187 * Description: Checks if seek offsets are in the range of 0 to LLONG_MAX.
188 *
189 * Parameters: fl Flock structure.
190 * cur_file_offset Current offset in the file.
191 *
192 * Returns: 0 on Success.
193 * EOVERFLOW on overflow.
194 * EINVAL on offset less than zero.
195 */
196
197 static int
198 check_file_seek_range(struct flock *fl, off_t cur_file_offset)
199 {
200 if (fl->l_whence == SEEK_CUR) {
201 /* Check if the start marker is beyond LLONG_MAX. */
202 if (CHECK_ADD_OVERFLOW_INT64L(fl->l_start, cur_file_offset)) {
203 /* Check if start marker is negative */
204 if (fl->l_start < 0) {
205 return EINVAL;
206 }
207 return EOVERFLOW;
208 }
209 /* Check if the start marker is negative. */
210 if (fl->l_start + cur_file_offset < 0) {
211 return EINVAL;
212 }
213 /* Check if end marker is beyond LLONG_MAX. */
214 if ((fl->l_len > 0) && (CHECK_ADD_OVERFLOW_INT64L(fl->l_start +
215 cur_file_offset, fl->l_len - 1))) {
216 return EOVERFLOW;
217 }
218 /* Check if the end marker is negative. */
219 if ((fl->l_len <= 0) && (fl->l_start + cur_file_offset +
220 fl->l_len < 0)) {
221 return EINVAL;
222 }
223 } else if (fl->l_whence == SEEK_SET) {
224 /* Check if the start marker is negative. */
225 if (fl->l_start < 0) {
226 return EINVAL;
227 }
228 /* Check if the end marker is beyond LLONG_MAX. */
229 if ((fl->l_len > 0) &&
230 CHECK_ADD_OVERFLOW_INT64L(fl->l_start, fl->l_len - 1)) {
231 return EOVERFLOW;
232 }
233 /* Check if the end marker is negative. */
234 if ((fl->l_len < 0) && fl->l_start + fl->l_len < 0) {
235 return EINVAL;
236 }
237 }
238 return 0;
239 }
240
241
242 /*
243 * file_lock_init
244 *
245 * Description: Initialize the file lock group and the uipc and flist locks
246 *
247 * Parameters: (void)
248 *
249 * Returns: void
250 *
251 * Notes: Called at system startup from bsd_init().
252 */
253 void
254 file_lock_init(void)
255 {
256 /* allocate file lock group attribute and group */
257 file_lck_grp_attr= lck_grp_attr_alloc_init();
258
259 file_lck_grp = lck_grp_alloc_init("file", file_lck_grp_attr);
260
261 /* Allocate file lock attribute */
262 file_lck_attr = lck_attr_alloc_init();
263
264 uipc_lock = lck_mtx_alloc_init(file_lck_grp, file_lck_attr);
265 }
266
267
268 /*
269 * proc_fdlock, proc_fdlock_spin
270 *
271 * Description: Lock to control access to the per process struct fileproc
272 * and struct filedesc
273 *
274 * Parameters: p Process to take the lock on
275 *
276 * Returns: void
277 *
278 * Notes: The lock is initialized in forkproc() and destroyed in
279 * reap_child_process().
280 */
281 void
282 proc_fdlock(proc_t p)
283 {
284 lck_mtx_lock(&p->p_fdmlock);
285 }
286
287 void
288 proc_fdlock_spin(proc_t p)
289 {
290 lck_mtx_lock_spin(&p->p_fdmlock);
291 }
292
293 void
294 proc_fdlock_assert(proc_t p, int assertflags)
295 {
296 lck_mtx_assert(&p->p_fdmlock, assertflags);
297 }
298
299
300 /*
301 * proc_fdunlock
302 *
303 * Description: Unlock the lock previously locked by a call to proc_fdlock()
304 *
305 * Parameters: p Process to drop the lock on
306 *
307 * Returns: void
308 */
309 void
310 proc_fdunlock(proc_t p)
311 {
312 lck_mtx_unlock(&p->p_fdmlock);
313 }
314
315
316 /*
317 * System calls on descriptors.
318 */
319
320
321 /*
322 * getdtablesize
323 *
324 * Description: Returns the per process maximum size of the descriptor table
325 *
326 * Parameters: p Process being queried
327 * retval Pointer to the call return area
328 *
329 * Returns: 0 Success
330 *
331 * Implicit returns:
332 * *retval (modified) Size of dtable
333 */
334 int
335 getdtablesize(proc_t p, __unused struct getdtablesize_args *uap, int32_t *retval)
336 {
337 proc_fdlock_spin(p);
338 *retval = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
339 proc_fdunlock(p);
340
341 return (0);
342 }
343
344
345 void
346 procfdtbl_reservefd(struct proc * p, int fd)
347 {
348 p->p_fd->fd_ofiles[fd] = NULL;
349 p->p_fd->fd_ofileflags[fd] |= UF_RESERVED;
350 }
351
352 void
353 procfdtbl_markclosefd(struct proc * p, int fd)
354 {
355 p->p_fd->fd_ofileflags[fd] |= (UF_RESERVED | UF_CLOSING);
356 }
357
358 void
359 procfdtbl_releasefd(struct proc * p, int fd, struct fileproc * fp)
360 {
361 if (fp != NULL)
362 p->p_fd->fd_ofiles[fd] = fp;
363 p->p_fd->fd_ofileflags[fd] &= ~UF_RESERVED;
364 if ((p->p_fd->fd_ofileflags[fd] & UF_RESVWAIT) == UF_RESVWAIT) {
365 p->p_fd->fd_ofileflags[fd] &= ~UF_RESVWAIT;
366 wakeup(&p->p_fd);
367 }
368 }
369
370 void
371 procfdtbl_waitfd(struct proc * p, int fd)
372 {
373 p->p_fd->fd_ofileflags[fd] |= UF_RESVWAIT;
374 msleep(&p->p_fd, &p->p_fdmlock, PRIBIO, "ftbl_waitfd", NULL);
375 }
376
377
378 void
379 procfdtbl_clearfd(struct proc * p, int fd)
380 {
381 int waiting;
382
383 waiting = (p->p_fd->fd_ofileflags[fd] & UF_RESVWAIT);
384 p->p_fd->fd_ofiles[fd] = NULL;
385 p->p_fd->fd_ofileflags[fd] = 0;
386 if ( waiting == UF_RESVWAIT) {
387 wakeup(&p->p_fd);
388 }
389 }
390
391 /*
392 * _fdrelse
393 *
394 * Description: Inline utility function to free an fd in a filedesc
395 *
396 * Parameters: fdp Pointer to filedesc fd lies in
397 * fd fd to free
398 * reserv fd should be reserved
399 *
400 * Returns: void
401 *
402 * Locks: Assumes proc_fdlock for process pointing to fdp is held by
403 * the caller
404 */
405 static void
406 _fdrelse(struct proc * p, int fd)
407 {
408 struct filedesc *fdp = p->p_fd;
409 int nfd = 0;
410
411 if (fd < fdp->fd_freefile)
412 fdp->fd_freefile = fd;
413 #if DIAGNOSTIC
414 if (fd > fdp->fd_lastfile)
415 panic("fdrelse: fd_lastfile inconsistent");
416 #endif
417 procfdtbl_clearfd(p, fd);
418
419 while ((nfd = fdp->fd_lastfile) > 0 &&
420 fdp->fd_ofiles[nfd] == NULL &&
421 !(fdp->fd_ofileflags[nfd] & UF_RESERVED))
422 fdp->fd_lastfile--;
423 }
424
425
426 int
427 fd_rdwr(
428 int fd,
429 enum uio_rw rw,
430 uint64_t base,
431 int64_t len,
432 enum uio_seg segflg,
433 off_t offset,
434 int io_flg,
435 int64_t *aresid)
436 {
437 struct fileproc *fp;
438 proc_t p;
439 int error = 0;
440 int flags = 0;
441 int spacetype;
442 uio_t auio = NULL;
443 char uio_buf[ UIO_SIZEOF(1) ];
444 struct vfs_context context = *(vfs_context_current());
445 bool wrote_some = false;
446
447 p = current_proc();
448
449 error = fp_lookup(p, fd, &fp, 0);
450 if (error)
451 return(error);
452
453 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_PIPE && fp->f_type != DTYPE_SOCKET) {
454 error = EINVAL;
455 goto out;
456 }
457 if (rw == UIO_WRITE && !(fp->f_flag & FWRITE)) {
458 error = EBADF;
459 goto out;
460 }
461
462 if (rw == UIO_READ && !(fp->f_flag & FREAD)) {
463 error = EBADF;
464 goto out;
465 }
466
467 context.vc_ucred = fp->f_fglob->fg_cred;
468
469 if (UIO_SEG_IS_USER_SPACE(segflg))
470 spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
471 else
472 spacetype = UIO_SYSSPACE;
473
474 auio = uio_createwithbuffer(1, offset, spacetype, rw, &uio_buf[0], sizeof(uio_buf));
475
476 uio_addiov(auio, base, len);
477
478 if ( !(io_flg & IO_APPEND))
479 flags = FOF_OFFSET;
480
481 if (rw == UIO_WRITE) {
482 user_ssize_t orig_resid = uio_resid(auio);
483 error = fo_write(fp, auio, flags, &context);
484 wrote_some = uio_resid(auio) < orig_resid;
485 } else
486 error = fo_read(fp, auio, flags, &context);
487
488 if (aresid)
489 *aresid = uio_resid(auio);
490 else {
491 if (uio_resid(auio) && error == 0)
492 error = EIO;
493 }
494 out:
495 if (wrote_some)
496 fp_drop_written(p, fd, fp);
497 else
498 fp_drop(p, fd, fp, 0);
499
500 return error;
501 }
502
503
504
505 /*
506 * dup
507 *
508 * Description: Duplicate a file descriptor.
509 *
510 * Parameters: p Process performing the dup
511 * uap->fd The fd to dup
512 * retval Pointer to the call return area
513 *
514 * Returns: 0 Success
515 * !0 Errno
516 *
517 * Implicit returns:
518 * *retval (modified) The new descriptor
519 */
520 int
521 dup(proc_t p, struct dup_args *uap, int32_t *retval)
522 {
523 struct filedesc *fdp = p->p_fd;
524 int old = uap->fd;
525 int new, error;
526 struct fileproc *fp;
527
528 proc_fdlock(p);
529 if ( (error = fp_lookup(p, old, &fp, 1)) ) {
530 proc_fdunlock(p);
531 return(error);
532 }
533 if (FP_ISGUARDED(fp, GUARD_DUP)) {
534 error = fp_guard_exception(p, old, fp, kGUARD_EXC_DUP);
535 (void) fp_drop(p, old, fp, 1);
536 proc_fdunlock(p);
537 return (error);
538 }
539 if ( (error = fdalloc(p, 0, &new)) ) {
540 fp_drop(p, old, fp, 1);
541 proc_fdunlock(p);
542 return (error);
543 }
544 error = finishdup(p, fdp, old, new, 0, retval);
545 fp_drop(p, old, fp, 1);
546 proc_fdunlock(p);
547
548 if (ENTR_SHOULDTRACE && fp->f_type == DTYPE_SOCKET) {
549 KERNEL_ENERGYTRACE(kEnTrActKernSocket, DBG_FUNC_START,
550 new, 0, (int64_t)VM_KERNEL_ADDRPERM(fp->f_data));
551 }
552
553 return (error);
554 }
555
556 /*
557 * dup2
558 *
559 * Description: Duplicate a file descriptor to a particular value.
560 *
561 * Parameters: p Process performing the dup
562 * uap->from The fd to dup
563 * uap->to The fd to dup it to
564 * retval Pointer to the call return area
565 *
566 * Returns: 0 Success
567 * !0 Errno
568 *
569 * Implicit returns:
570 * *retval (modified) The new descriptor
571 */
572 int
573 dup2(proc_t p, struct dup2_args *uap, int32_t *retval)
574 {
575 struct filedesc *fdp = p->p_fd;
576 int old = uap->from, new = uap->to;
577 int i, error;
578 struct fileproc *fp, *nfp;
579
580 proc_fdlock(p);
581
582 startover:
583 if ( (error = fp_lookup(p, old, &fp, 1)) ) {
584 proc_fdunlock(p);
585 return(error);
586 }
587 if (FP_ISGUARDED(fp, GUARD_DUP)) {
588 error = fp_guard_exception(p, old, fp, kGUARD_EXC_DUP);
589 (void) fp_drop(p, old, fp, 1);
590 proc_fdunlock(p);
591 return (error);
592 }
593 if (new < 0 ||
594 (rlim_t)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
595 new >= maxfiles) {
596 fp_drop(p, old, fp, 1);
597 proc_fdunlock(p);
598 return (EBADF);
599 }
600 if (old == new) {
601 fp_drop(p, old, fp, 1);
602 *retval = new;
603 proc_fdunlock(p);
604 return (0);
605 }
606 if (new < 0 || new >= fdp->fd_nfiles) {
607 if ( (error = fdalloc(p, new, &i)) ) {
608 fp_drop(p, old, fp, 1);
609 proc_fdunlock(p);
610 return (error);
611 }
612 if (new != i) {
613 fdrelse(p, i);
614 goto closeit;
615 }
616 } else {
617 closeit:
618 while ((fdp->fd_ofileflags[new] & UF_RESERVED) == UF_RESERVED) {
619 fp_drop(p, old, fp, 1);
620 procfdtbl_waitfd(p, new);
621 #if DIAGNOSTIC
622 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
623 #endif
624 goto startover;
625 }
626
627 if ((fdp->fd_ofiles[new] != NULL) &&
628 ((error = fp_lookup(p, new, &nfp, 1)) == 0)) {
629 fp_drop(p, old, fp, 1);
630 if (FP_ISGUARDED(nfp, GUARD_CLOSE)) {
631 error = fp_guard_exception(p,
632 new, nfp, kGUARD_EXC_CLOSE);
633 (void) fp_drop(p, new, nfp, 1);
634 proc_fdunlock(p);
635 return (error);
636 }
637 (void)close_internal_locked(p, new, nfp, FD_DUP2RESV);
638 #if DIAGNOSTIC
639 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
640 #endif
641 procfdtbl_clearfd(p, new);
642 goto startover;
643 } else {
644 #if DIAGNOSTIC
645 if (fdp->fd_ofiles[new] != NULL)
646 panic("dup2: no ref on fileproc %d", new);
647 #endif
648 procfdtbl_reservefd(p, new);
649 }
650
651 #if DIAGNOSTIC
652 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
653 #endif
654
655 }
656 #if DIAGNOSTIC
657 if (fdp->fd_ofiles[new] != 0)
658 panic("dup2: overwriting fd_ofiles with new %d", new);
659 if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0)
660 panic("dup2: unreserved fileflags with new %d", new);
661 #endif
662 error = finishdup(p, fdp, old, new, 0, retval);
663 fp_drop(p, old, fp, 1);
664 proc_fdunlock(p);
665
666 return(error);
667 }
668
669
670 /*
671 * fcntl
672 *
673 * Description: The file control system call.
674 *
675 * Parameters: p Process performing the fcntl
676 * uap->fd The fd to operate against
677 * uap->cmd The command to perform
678 * uap->arg Pointer to the command argument
679 * retval Pointer to the call return area
680 *
681 * Returns: 0 Success
682 * !0 Errno (see fcntl_nocancel)
683 *
684 * Implicit returns:
685 * *retval (modified) fcntl return value (if any)
686 *
687 * Notes: This system call differs from fcntl_nocancel() in that it
688 * tests for cancellation prior to performing a potentially
689 * blocking operation.
690 */
691 int
692 fcntl(proc_t p, struct fcntl_args *uap, int32_t *retval)
693 {
694 __pthread_testcancel(1);
695 return(fcntl_nocancel(p, (struct fcntl_nocancel_args *)uap, retval));
696 }
697
698
699 /*
700 * fcntl_nocancel
701 *
702 * Description: A non-cancel-testing file control system call.
703 *
704 * Parameters: p Process performing the fcntl
705 * uap->fd The fd to operate against
706 * uap->cmd The command to perform
707 * uap->arg Pointer to the command argument
708 * retval Pointer to the call return area
709 *
710 * Returns: 0 Success
711 * EINVAL
712 * fp_lookup:EBADF Bad file descriptor
713 * [F_DUPFD]
714 * fdalloc:EMFILE
715 * fdalloc:ENOMEM
716 * finishdup:EBADF
717 * finishdup:ENOMEM
718 * [F_SETOWN]
719 * ESRCH
720 * [F_SETLK]
721 * EBADF
722 * EOVERFLOW
723 * copyin:EFAULT
724 * vnode_getwithref:???
725 * VNOP_ADVLOCK:???
726 * msleep:ETIMEDOUT
727 * [F_GETLK]
728 * EBADF
729 * EOVERFLOW
730 * copyin:EFAULT
731 * copyout:EFAULT
732 * vnode_getwithref:???
733 * VNOP_ADVLOCK:???
734 * [F_PREALLOCATE]
735 * EBADF
736 * EINVAL
737 * copyin:EFAULT
738 * copyout:EFAULT
739 * vnode_getwithref:???
740 * VNOP_ALLOCATE:???
741 * [F_SETSIZE,F_RDADVISE]
742 * EBADF
743 * copyin:EFAULT
744 * vnode_getwithref:???
745 * [F_RDAHEAD,F_NOCACHE]
746 * EBADF
747 * vnode_getwithref:???
748 * [???]
749 *
750 * Implicit returns:
751 * *retval (modified) fcntl return value (if any)
752 */
753 int
754 fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval)
755 {
756 int fd = uap->fd;
757 struct filedesc *fdp = p->p_fd;
758 struct fileproc *fp;
759 char *pop;
760 struct vnode *vp = NULLVP; /* for AUDIT_ARG() at end */
761 int i, tmp, error, error2, flg = 0;
762 struct flock fl;
763 struct flocktimeout fltimeout;
764 struct timespec *timeout = NULL;
765 struct vfs_context context;
766 off_t offset;
767 int newmin;
768 daddr64_t lbn, bn;
769 unsigned int fflag;
770 user_addr_t argp;
771 boolean_t is64bit;
772
773 AUDIT_ARG(fd, uap->fd);
774 AUDIT_ARG(cmd, uap->cmd);
775
776 proc_fdlock(p);
777 if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
778 proc_fdunlock(p);
779 return(error);
780 }
781 context.vc_thread = current_thread();
782 context.vc_ucred = fp->f_cred;
783
784 is64bit = proc_is64bit(p);
785 if (is64bit) {
786 argp = uap->arg;
787 }
788 else {
789 /*
790 * Since the arg parameter is defined as a long but may be
791 * either a long or a pointer we must take care to handle
792 * sign extension issues. Our sys call munger will sign
793 * extend a long when we are called from a 32-bit process.
794 * Since we can never have an address greater than 32-bits
795 * from a 32-bit process we lop off the top 32-bits to avoid
796 * getting the wrong address
797 */
798 argp = CAST_USER_ADDR_T((uint32_t)uap->arg);
799 }
800
801 pop = &fdp->fd_ofileflags[fd];
802
803 #if CONFIG_MACF
804 error = mac_file_check_fcntl(proc_ucred(p), fp->f_fglob, uap->cmd,
805 uap->arg);
806 if (error)
807 goto out;
808 #endif
809
810 switch (uap->cmd) {
811
812 case F_DUPFD:
813 case F_DUPFD_CLOEXEC:
814 if (FP_ISGUARDED(fp, GUARD_DUP)) {
815 error = fp_guard_exception(p, fd, fp, kGUARD_EXC_DUP);
816 goto out;
817 }
818 newmin = CAST_DOWN_EXPLICIT(int, uap->arg); /* arg is an int, so we won't lose bits */
819 AUDIT_ARG(value32, newmin);
820 if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
821 newmin >= maxfiles) {
822 error = EINVAL;
823 goto out;
824 }
825 if ( (error = fdalloc(p, newmin, &i)) )
826 goto out;
827 error = finishdup(p, fdp, fd, i,
828 uap->cmd == F_DUPFD_CLOEXEC ? UF_EXCLOSE : 0, retval);
829 goto out;
830
831 case F_GETFD:
832 *retval = (*pop & UF_EXCLOSE)? FD_CLOEXEC : 0;
833 error = 0;
834 goto out;
835
836 case F_SETFD:
837 AUDIT_ARG(value32, uap->arg);
838 if (uap->arg & FD_CLOEXEC)
839 *pop |= UF_EXCLOSE;
840 else {
841 if (FILEPROC_TYPE(fp) == FTYPE_GUARDED) {
842 error = fp_guard_exception(p,
843 fd, fp, kGUARD_EXC_NOCLOEXEC);
844 goto out;
845 }
846 *pop &= ~UF_EXCLOSE;
847 }
848 error = 0;
849 goto out;
850
851 case F_GETFL:
852 *retval = OFLAGS(fp->f_flag);
853 error = 0;
854 goto out;
855
856 case F_SETFL:
857 fp->f_flag &= ~FCNTLFLAGS;
858 tmp = CAST_DOWN_EXPLICIT(int, uap->arg); /* arg is an int, so we won't lose bits */
859 AUDIT_ARG(value32, tmp);
860 fp->f_flag |= FFLAGS(tmp) & FCNTLFLAGS;
861 tmp = fp->f_flag & FNONBLOCK;
862 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
863 if (error)
864 goto out;
865 tmp = fp->f_flag & FASYNC;
866 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, &context);
867 if (!error)
868 goto out;
869 fp->f_flag &= ~FNONBLOCK;
870 tmp = 0;
871 (void)fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
872 goto out;
873
874 case F_GETOWN:
875 if (fp->f_type == DTYPE_SOCKET) {
876 *retval = ((struct socket *)fp->f_data)->so_pgid;
877 error = 0;
878 goto out;
879 }
880 error = fo_ioctl(fp, (int)TIOCGPGRP, (caddr_t)retval, &context);
881 *retval = -*retval;
882 goto out;
883
884 case F_SETOWN:
885 tmp = CAST_DOWN_EXPLICIT(pid_t, uap->arg); /* arg is an int, so we won't lose bits */
886 AUDIT_ARG(value32, tmp);
887 if (fp->f_type == DTYPE_SOCKET) {
888 ((struct socket *)fp->f_data)->so_pgid = tmp;
889 error =0;
890 goto out;
891 }
892 if (fp->f_type == DTYPE_PIPE) {
893 error = fo_ioctl(fp, TIOCSPGRP, (caddr_t)&tmp, &context);
894 goto out;
895 }
896
897 if (tmp <= 0) {
898 tmp = -tmp;
899 } else {
900 proc_t p1 = proc_find(tmp);
901 if (p1 == 0) {
902 error = ESRCH;
903 goto out;
904 }
905 tmp = (int)p1->p_pgrpid;
906 proc_rele(p1);
907 }
908 error = fo_ioctl(fp, (int)TIOCSPGRP, (caddr_t)&tmp, &context);
909 goto out;
910
911 case F_SETNOSIGPIPE:
912 tmp = CAST_DOWN_EXPLICIT(int, uap->arg);
913 if (fp->f_type == DTYPE_SOCKET) {
914 #if SOCKETS
915 error = sock_setsockopt((struct socket *)fp->f_data,
916 SOL_SOCKET, SO_NOSIGPIPE, &tmp, sizeof (tmp));
917 #else
918 error = EINVAL;
919 #endif
920 } else {
921 struct fileglob *fg = fp->f_fglob;
922
923 lck_mtx_lock_spin(&fg->fg_lock);
924 if (tmp)
925 fg->fg_lflags |= FG_NOSIGPIPE;
926 else
927 fg->fg_lflags &= FG_NOSIGPIPE;
928 lck_mtx_unlock(&fg->fg_lock);
929 error = 0;
930 }
931 goto out;
932
933 case F_GETNOSIGPIPE:
934 if (fp->f_type == DTYPE_SOCKET) {
935 #if SOCKETS
936 int retsize = sizeof (*retval);
937 error = sock_getsockopt((struct socket *)fp->f_data,
938 SOL_SOCKET, SO_NOSIGPIPE, retval, &retsize);
939 #else
940 error = EINVAL;
941 #endif
942 } else {
943 *retval = (fp->f_fglob->fg_lflags & FG_NOSIGPIPE) ?
944 1 : 0;
945 error = 0;
946 }
947 goto out;
948
949 case F_SETCONFINED:
950 /*
951 * If this is the only reference to this fglob in the process
952 * and it's already marked as close-on-fork then mark it as
953 * (immutably) "confined" i.e. any fd that points to it will
954 * forever be close-on-fork, and attempts to use an IPC
955 * mechanism to move the descriptor elsewhere will fail.
956 */
957 if (CAST_DOWN_EXPLICIT(int, uap->arg)) {
958 struct fileglob *fg = fp->f_fglob;
959
960 lck_mtx_lock_spin(&fg->fg_lock);
961 if (fg->fg_lflags & FG_CONFINED)
962 error = 0;
963 else if (1 != fg->fg_count)
964 error = EAGAIN; /* go close the dup .. */
965 else if (UF_FORKCLOSE == (*pop & UF_FORKCLOSE)) {
966 fg->fg_lflags |= FG_CONFINED;
967 error = 0;
968 } else
969 error = EBADF; /* open without O_CLOFORK? */
970 lck_mtx_unlock(&fg->fg_lock);
971 } else {
972 /*
973 * Other subsystems may have built on the immutability
974 * of FG_CONFINED; clearing it may be tricky.
975 */
976 error = EPERM; /* immutable */
977 }
978 goto out;
979
980 case F_GETCONFINED:
981 *retval = (fp->f_fglob->fg_lflags & FG_CONFINED) ? 1 : 0;
982 error = 0;
983 goto out;
984
985 case F_SETLKWTIMEOUT:
986 case F_SETLKW:
987 case F_OFD_SETLKWTIMEOUT:
988 case F_OFD_SETLKW:
989 flg |= F_WAIT;
990 /* Fall into F_SETLK */
991
992 case F_SETLK:
993 case F_OFD_SETLK:
994 if (fp->f_type != DTYPE_VNODE) {
995 error = EBADF;
996 goto out;
997 }
998 vp = (struct vnode *)fp->f_data;
999
1000 fflag = fp->f_flag;
1001 offset = fp->f_offset;
1002 proc_fdunlock(p);
1003
1004 /* Copy in the lock structure */
1005 if (F_SETLKWTIMEOUT == uap->cmd ||
1006 F_OFD_SETLKWTIMEOUT == uap->cmd) {
1007 error = copyin(argp, (caddr_t) &fltimeout, sizeof(fltimeout));
1008 if (error) {
1009 goto outdrop;
1010 }
1011 fl = fltimeout.fl;
1012 timeout = &fltimeout.timeout;
1013 } else {
1014 error = copyin(argp, (caddr_t)&fl, sizeof(fl));
1015 if (error) {
1016 goto outdrop;
1017 }
1018 }
1019
1020 /* Check starting byte and ending byte for EOVERFLOW in SEEK_CUR */
1021 /* and ending byte for EOVERFLOW in SEEK_SET */
1022 error = check_file_seek_range(&fl, offset);
1023 if (error) {
1024 goto outdrop;
1025 }
1026
1027 if ( (error = vnode_getwithref(vp)) ) {
1028 goto outdrop;
1029 }
1030 if (fl.l_whence == SEEK_CUR)
1031 fl.l_start += offset;
1032
1033 #if CONFIG_MACF
1034 error = mac_file_check_lock(proc_ucred(p), fp->f_fglob,
1035 F_SETLK, &fl);
1036 if (error) {
1037 (void)vnode_put(vp);
1038 goto outdrop;
1039 }
1040 #endif
1041 switch (uap->cmd) {
1042 case F_OFD_SETLK:
1043 case F_OFD_SETLKW:
1044 case F_OFD_SETLKWTIMEOUT:
1045 flg |= F_OFD_LOCK;
1046 switch (fl.l_type) {
1047 case F_RDLCK:
1048 if ((fflag & FREAD) == 0) {
1049 error = EBADF;
1050 break;
1051 }
1052 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1053 F_SETLK, &fl, flg, &context, timeout);
1054 break;
1055 case F_WRLCK:
1056 if ((fflag & FWRITE) == 0) {
1057 error = EBADF;
1058 break;
1059 }
1060 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1061 F_SETLK, &fl, flg, &context, timeout);
1062 break;
1063 case F_UNLCK:
1064 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1065 F_UNLCK, &fl, F_OFD_LOCK, &context,
1066 timeout);
1067 break;
1068 default:
1069 error = EINVAL;
1070 break;
1071 }
1072 if (0 == error &&
1073 (F_RDLCK == fl.l_type || F_WRLCK == fl.l_type)) {
1074 struct fileglob *fg = fp->f_fglob;
1075
1076 /*
1077 * arrange F_UNLCK on last close (once
1078 * set, FG_HAS_OFDLOCK is immutable)
1079 */
1080 if ((fg->fg_lflags & FG_HAS_OFDLOCK) == 0) {
1081 lck_mtx_lock_spin(&fg->fg_lock);
1082 fg->fg_lflags |= FG_HAS_OFDLOCK;
1083 lck_mtx_unlock(&fg->fg_lock);
1084 }
1085 }
1086 break;
1087 default:
1088 flg |= F_POSIX;
1089 switch (fl.l_type) {
1090 case F_RDLCK:
1091 if ((fflag & FREAD) == 0) {
1092 error = EBADF;
1093 break;
1094 }
1095 // XXX UInt32 unsafe for LP64 kernel
1096 OSBitOrAtomic(P_LADVLOCK, &p->p_ladvflag);
1097 error = VNOP_ADVLOCK(vp, (caddr_t)p,
1098 F_SETLK, &fl, flg, &context, timeout);
1099 break;
1100 case F_WRLCK:
1101 if ((fflag & FWRITE) == 0) {
1102 error = EBADF;
1103 break;
1104 }
1105 // XXX UInt32 unsafe for LP64 kernel
1106 OSBitOrAtomic(P_LADVLOCK, &p->p_ladvflag);
1107 error = VNOP_ADVLOCK(vp, (caddr_t)p,
1108 F_SETLK, &fl, flg, &context, timeout);
1109 break;
1110 case F_UNLCK:
1111 error = VNOP_ADVLOCK(vp, (caddr_t)p,
1112 F_UNLCK, &fl, F_POSIX, &context, timeout);
1113 break;
1114 default:
1115 error = EINVAL;
1116 break;
1117 }
1118 break;
1119 }
1120 (void) vnode_put(vp);
1121 goto outdrop;
1122
1123 case F_GETLK:
1124 case F_OFD_GETLK:
1125 if (fp->f_type != DTYPE_VNODE) {
1126 error = EBADF;
1127 goto out;
1128 }
1129 vp = (struct vnode *)fp->f_data;
1130
1131 offset = fp->f_offset;
1132 proc_fdunlock(p);
1133
1134 /* Copy in the lock structure */
1135 error = copyin(argp, (caddr_t)&fl, sizeof(fl));
1136 if (error)
1137 goto outdrop;
1138
1139 /* Check starting byte and ending byte for EOVERFLOW in SEEK_CUR */
1140 /* and ending byte for EOVERFLOW in SEEK_SET */
1141 error = check_file_seek_range(&fl, offset);
1142 if (error) {
1143 goto outdrop;
1144 }
1145
1146 if ((fl.l_whence == SEEK_SET) && (fl.l_start < 0)) {
1147 error = EINVAL;
1148 goto outdrop;
1149 }
1150
1151 switch (fl.l_type) {
1152 case F_RDLCK:
1153 case F_UNLCK:
1154 case F_WRLCK:
1155 break;
1156 default:
1157 error = EINVAL;
1158 goto outdrop;
1159 }
1160
1161 switch (fl.l_whence) {
1162 case SEEK_CUR:
1163 case SEEK_SET:
1164 case SEEK_END:
1165 break;
1166 default:
1167 error = EINVAL;
1168 goto outdrop;
1169 }
1170
1171 if ( (error = vnode_getwithref(vp)) == 0 ) {
1172 if (fl.l_whence == SEEK_CUR)
1173 fl.l_start += offset;
1174
1175 #if CONFIG_MACF
1176 error = mac_file_check_lock(proc_ucred(p), fp->f_fglob,
1177 uap->cmd, &fl);
1178 if (error == 0)
1179 #endif
1180 switch (uap->cmd) {
1181 case F_OFD_GETLK:
1182 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1183 F_GETLK, &fl, F_OFD_LOCK, &context, NULL);
1184 break;
1185 case F_OFD_GETLKPID:
1186 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1187 F_GETLKPID, &fl, F_OFD_LOCK, &context, NULL);
1188 break;
1189 default:
1190 error = VNOP_ADVLOCK(vp, (caddr_t)p,
1191 uap->cmd, &fl, F_POSIX, &context, NULL);
1192 break;
1193 }
1194
1195 (void)vnode_put(vp);
1196
1197 if (error == 0)
1198 error = copyout((caddr_t)&fl, argp, sizeof(fl));
1199 }
1200 goto outdrop;
1201
1202 case F_PREALLOCATE: {
1203 fstore_t alloc_struct; /* structure for allocate command */
1204 u_int32_t alloc_flags = 0;
1205
1206 if (fp->f_type != DTYPE_VNODE) {
1207 error = EBADF;
1208 goto out;
1209 }
1210
1211 vp = (struct vnode *)fp->f_data;
1212 proc_fdunlock(p);
1213
1214 /* make sure that we have write permission */
1215 if ((fp->f_flag & FWRITE) == 0) {
1216 error = EBADF;
1217 goto outdrop;
1218 }
1219
1220 error = copyin(argp, (caddr_t)&alloc_struct, sizeof(alloc_struct));
1221 if (error)
1222 goto outdrop;
1223
1224 /* now set the space allocated to 0 */
1225 alloc_struct.fst_bytesalloc = 0;
1226
1227 /*
1228 * Do some simple parameter checking
1229 */
1230
1231 /* set up the flags */
1232
1233 alloc_flags |= PREALLOCATE;
1234
1235 if (alloc_struct.fst_flags & F_ALLOCATECONTIG)
1236 alloc_flags |= ALLOCATECONTIG;
1237
1238 if (alloc_struct.fst_flags & F_ALLOCATEALL)
1239 alloc_flags |= ALLOCATEALL;
1240
1241 /*
1242 * Do any position mode specific stuff. The only
1243 * position mode supported now is PEOFPOSMODE
1244 */
1245
1246 switch (alloc_struct.fst_posmode) {
1247
1248 case F_PEOFPOSMODE:
1249 if (alloc_struct.fst_offset != 0) {
1250 error = EINVAL;
1251 goto outdrop;
1252 }
1253
1254 alloc_flags |= ALLOCATEFROMPEOF;
1255 break;
1256
1257 case F_VOLPOSMODE:
1258 if (alloc_struct.fst_offset <= 0) {
1259 error = EINVAL;
1260 goto outdrop;
1261 }
1262
1263 alloc_flags |= ALLOCATEFROMVOL;
1264 break;
1265
1266 default: {
1267 error = EINVAL;
1268 goto outdrop;
1269 }
1270 }
1271 if ( (error = vnode_getwithref(vp)) == 0 ) {
1272 /*
1273 * call allocate to get the space
1274 */
1275 error = VNOP_ALLOCATE(vp,alloc_struct.fst_length,alloc_flags,
1276 &alloc_struct.fst_bytesalloc, alloc_struct.fst_offset,
1277 &context);
1278 (void)vnode_put(vp);
1279
1280 error2 = copyout((caddr_t)&alloc_struct, argp, sizeof(alloc_struct));
1281
1282 if (error == 0)
1283 error = error2;
1284 }
1285 goto outdrop;
1286
1287 }
1288 case F_SETSIZE:
1289 if (fp->f_type != DTYPE_VNODE) {
1290 error = EBADF;
1291 goto out;
1292 }
1293 vp = (struct vnode *)fp->f_data;
1294 proc_fdunlock(p);
1295
1296 error = copyin(argp, (caddr_t)&offset, sizeof (off_t));
1297 if (error)
1298 goto outdrop;
1299 AUDIT_ARG(value64, offset);
1300
1301 error = vnode_getwithref(vp);
1302 if (error)
1303 goto outdrop;
1304
1305 #if CONFIG_MACF
1306 error = mac_vnode_check_truncate(&context,
1307 fp->f_fglob->fg_cred, vp);
1308 if (error) {
1309 (void)vnode_put(vp);
1310 goto outdrop;
1311 }
1312 #endif
1313 /*
1314 * Make sure that we are root. Growing a file
1315 * without zero filling the data is a security hole
1316 * root would have access anyway so we'll allow it
1317 */
1318 if (!kauth_cred_issuser(kauth_cred_get())) {
1319 error = EACCES;
1320 } else {
1321 /*
1322 * set the file size
1323 */
1324 error = vnode_setsize(vp, offset, IO_NOZEROFILL,
1325 &context);
1326 }
1327
1328 (void)vnode_put(vp);
1329 goto outdrop;
1330
1331 case F_RDAHEAD:
1332 if (fp->f_type != DTYPE_VNODE) {
1333 error = EBADF;
1334 goto out;
1335 }
1336 if (uap->arg)
1337 fp->f_fglob->fg_flag &= ~FNORDAHEAD;
1338 else
1339 fp->f_fglob->fg_flag |= FNORDAHEAD;
1340
1341 goto out;
1342
1343 case F_NOCACHE:
1344 if (fp->f_type != DTYPE_VNODE) {
1345 error = EBADF;
1346 goto out;
1347 }
1348 if (uap->arg)
1349 fp->f_fglob->fg_flag |= FNOCACHE;
1350 else
1351 fp->f_fglob->fg_flag &= ~FNOCACHE;
1352
1353 goto out;
1354
1355 case F_NODIRECT:
1356 if (fp->f_type != DTYPE_VNODE) {
1357 error = EBADF;
1358 goto out;
1359 }
1360 if (uap->arg)
1361 fp->f_fglob->fg_flag |= FNODIRECT;
1362 else
1363 fp->f_fglob->fg_flag &= ~FNODIRECT;
1364
1365 goto out;
1366
1367 case F_SINGLE_WRITER:
1368 if (fp->f_type != DTYPE_VNODE) {
1369 error = EBADF;
1370 goto out;
1371 }
1372 if (uap->arg)
1373 fp->f_fglob->fg_flag |= FSINGLE_WRITER;
1374 else
1375 fp->f_fglob->fg_flag &= ~FSINGLE_WRITER;
1376
1377 goto out;
1378
1379 case F_GLOBAL_NOCACHE:
1380 if (fp->f_type != DTYPE_VNODE) {
1381 error = EBADF;
1382 goto out;
1383 }
1384 vp = (struct vnode *)fp->f_data;
1385 proc_fdunlock(p);
1386
1387 if ( (error = vnode_getwithref(vp)) == 0 ) {
1388
1389 *retval = vnode_isnocache(vp);
1390
1391 if (uap->arg)
1392 vnode_setnocache(vp);
1393 else
1394 vnode_clearnocache(vp);
1395
1396 (void)vnode_put(vp);
1397 }
1398 goto outdrop;
1399
1400 case F_CHECK_OPENEVT:
1401 if (fp->f_type != DTYPE_VNODE) {
1402 error = EBADF;
1403 goto out;
1404 }
1405 vp = (struct vnode *)fp->f_data;
1406 proc_fdunlock(p);
1407
1408 if ( (error = vnode_getwithref(vp)) == 0 ) {
1409
1410 *retval = vnode_is_openevt(vp);
1411
1412 if (uap->arg)
1413 vnode_set_openevt(vp);
1414 else
1415 vnode_clear_openevt(vp);
1416
1417 (void)vnode_put(vp);
1418 }
1419 goto outdrop;
1420
1421 case F_RDADVISE: {
1422 struct radvisory ra_struct;
1423
1424 if (fp->f_type != DTYPE_VNODE) {
1425 error = EBADF;
1426 goto out;
1427 }
1428 vp = (struct vnode *)fp->f_data;
1429 proc_fdunlock(p);
1430
1431 if ( (error = copyin(argp, (caddr_t)&ra_struct, sizeof(ra_struct))) )
1432 goto outdrop;
1433 if ( (error = vnode_getwithref(vp)) == 0 ) {
1434 error = VNOP_IOCTL(vp, F_RDADVISE, (caddr_t)&ra_struct, 0, &context);
1435
1436 (void)vnode_put(vp);
1437 }
1438 goto outdrop;
1439 }
1440
1441 case F_FLUSH_DATA:
1442
1443 if (fp->f_type != DTYPE_VNODE) {
1444 error = EBADF;
1445 goto out;
1446 }
1447 vp = (struct vnode *)fp->f_data;
1448 proc_fdunlock(p);
1449
1450 if ( (error = vnode_getwithref(vp)) == 0 ) {
1451 error = cluster_push(vp, 0);
1452
1453 (void)vnode_put(vp);
1454 }
1455 goto outdrop;
1456
1457 case F_LOG2PHYS:
1458 case F_LOG2PHYS_EXT: {
1459 struct log2phys l2p_struct; /* structure for allocate command */
1460 int devBlockSize;
1461
1462 off_t file_offset = 0;
1463 size_t a_size = 0;
1464 size_t run = 0;
1465
1466 if (uap->cmd == F_LOG2PHYS_EXT) {
1467 error = copyin(argp, (caddr_t)&l2p_struct, sizeof(l2p_struct));
1468 if (error)
1469 goto out;
1470 file_offset = l2p_struct.l2p_devoffset;
1471 } else {
1472 file_offset = fp->f_offset;
1473 }
1474 if (fp->f_type != DTYPE_VNODE) {
1475 error = EBADF;
1476 goto out;
1477 }
1478 vp = (struct vnode *)fp->f_data;
1479 proc_fdunlock(p);
1480 if ( (error = vnode_getwithref(vp)) ) {
1481 goto outdrop;
1482 }
1483 error = VNOP_OFFTOBLK(vp, file_offset, &lbn);
1484 if (error) {
1485 (void)vnode_put(vp);
1486 goto outdrop;
1487 }
1488 error = VNOP_BLKTOOFF(vp, lbn, &offset);
1489 if (error) {
1490 (void)vnode_put(vp);
1491 goto outdrop;
1492 }
1493 devBlockSize = vfs_devblocksize(vnode_mount(vp));
1494 if (uap->cmd == F_LOG2PHYS_EXT) {
1495 if (l2p_struct.l2p_contigbytes < 0) {
1496 vnode_put(vp);
1497 error = EINVAL;
1498 goto outdrop;
1499 }
1500
1501 a_size = MIN((uint64_t)l2p_struct.l2p_contigbytes, SIZE_MAX);
1502 } else {
1503 a_size = devBlockSize;
1504 }
1505
1506 error = VNOP_BLOCKMAP(vp, offset, a_size, &bn, &run, NULL, 0, &context);
1507
1508 (void)vnode_put(vp);
1509
1510 if (!error) {
1511 l2p_struct.l2p_flags = 0; /* for now */
1512 if (uap->cmd == F_LOG2PHYS_EXT) {
1513 l2p_struct.l2p_contigbytes = run - (file_offset - offset);
1514 } else {
1515 l2p_struct.l2p_contigbytes = 0; /* for now */
1516 }
1517
1518 /*
1519 * The block number being -1 suggests that the file offset is not backed
1520 * by any real blocks on-disk. As a result, just let it be passed back up wholesale.
1521 */
1522 if (bn == -1) {
1523 /* Don't multiply it by the block size */
1524 l2p_struct.l2p_devoffset = bn;
1525 }
1526 else {
1527 l2p_struct.l2p_devoffset = bn * devBlockSize;
1528 l2p_struct.l2p_devoffset += file_offset - offset;
1529 }
1530 error = copyout((caddr_t)&l2p_struct, argp, sizeof(l2p_struct));
1531 }
1532 goto outdrop;
1533 }
1534 case F_GETPATH: {
1535 char *pathbufp;
1536 int pathlen;
1537
1538 if (fp->f_type != DTYPE_VNODE) {
1539 error = EBADF;
1540 goto out;
1541 }
1542 vp = (struct vnode *)fp->f_data;
1543 proc_fdunlock(p);
1544
1545 pathlen = MAXPATHLEN;
1546 MALLOC(pathbufp, char *, pathlen, M_TEMP, M_WAITOK);
1547 if (pathbufp == NULL) {
1548 error = ENOMEM;
1549 goto outdrop;
1550 }
1551 if ( (error = vnode_getwithref(vp)) == 0 ) {
1552 error = vn_getpath(vp, pathbufp, &pathlen);
1553 (void)vnode_put(vp);
1554
1555 if (error == 0)
1556 error = copyout((caddr_t)pathbufp, argp, pathlen);
1557 }
1558 FREE(pathbufp, M_TEMP);
1559 goto outdrop;
1560 }
1561
1562 case F_PATHPKG_CHECK: {
1563 char *pathbufp;
1564 size_t pathlen;
1565
1566 if (fp->f_type != DTYPE_VNODE) {
1567 error = EBADF;
1568 goto out;
1569 }
1570 vp = (struct vnode *)fp->f_data;
1571 proc_fdunlock(p);
1572
1573 pathlen = MAXPATHLEN;
1574 pathbufp = kalloc(MAXPATHLEN);
1575
1576 if ( (error = copyinstr(argp, pathbufp, MAXPATHLEN, &pathlen)) == 0 ) {
1577 if ( (error = vnode_getwithref(vp)) == 0 ) {
1578 AUDIT_ARG(text, pathbufp);
1579 error = vn_path_package_check(vp, pathbufp, pathlen, retval);
1580
1581 (void)vnode_put(vp);
1582 }
1583 }
1584 kfree(pathbufp, MAXPATHLEN);
1585 goto outdrop;
1586 }
1587
1588 case F_CHKCLEAN: // used by regression tests to see if all dirty pages got cleaned by fsync()
1589 case F_FULLFSYNC: // fsync + flush the journal + DKIOCSYNCHRONIZE
1590 case F_BARRIERFSYNC: // fsync + barrier
1591 case F_FREEZE_FS: // freeze all other fs operations for the fs of this fd
1592 case F_THAW_FS: { // thaw all frozen fs operations for the fs of this fd
1593 if (fp->f_type != DTYPE_VNODE) {
1594 error = EBADF;
1595 goto out;
1596 }
1597 vp = (struct vnode *)fp->f_data;
1598 proc_fdunlock(p);
1599
1600 if ( (error = vnode_getwithref(vp)) == 0 ) {
1601 error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)NULL, 0, &context);
1602
1603 (void)vnode_put(vp);
1604 }
1605 break;
1606 }
1607
1608 /*
1609 * SPI (private) for opening a file starting from a dir fd
1610 */
1611 case F_OPENFROM: {
1612 struct user_fopenfrom fopen;
1613 struct vnode_attr va;
1614 struct nameidata nd;
1615 int cmode;
1616
1617 /* Check if this isn't a valid file descriptor */
1618 if ((fp->f_type != DTYPE_VNODE) ||
1619 (fp->f_flag & FREAD) == 0) {
1620 error = EBADF;
1621 goto out;
1622 }
1623 vp = (struct vnode *)fp->f_data;
1624 proc_fdunlock(p);
1625
1626 if (vnode_getwithref(vp)) {
1627 error = ENOENT;
1628 goto outdrop;
1629 }
1630
1631 /* Only valid for directories */
1632 if (vp->v_type != VDIR) {
1633 vnode_put(vp);
1634 error = ENOTDIR;
1635 goto outdrop;
1636 }
1637
1638 /* Get flags, mode and pathname arguments. */
1639 if (IS_64BIT_PROCESS(p)) {
1640 error = copyin(argp, &fopen, sizeof(fopen));
1641 } else {
1642 struct user32_fopenfrom fopen32;
1643
1644 error = copyin(argp, &fopen32, sizeof(fopen32));
1645 fopen.o_flags = fopen32.o_flags;
1646 fopen.o_mode = fopen32.o_mode;
1647 fopen.o_pathname = CAST_USER_ADDR_T(fopen32.o_pathname);
1648 }
1649 if (error) {
1650 vnode_put(vp);
1651 goto outdrop;
1652 }
1653 AUDIT_ARG(fflags, fopen.o_flags);
1654 AUDIT_ARG(mode, fopen.o_mode);
1655 VATTR_INIT(&va);
1656 /* Mask off all but regular access permissions */
1657 cmode = ((fopen.o_mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1658 VATTR_SET(&va, va_mode, cmode & ACCESSPERMS);
1659
1660 /* Start the lookup relative to the file descriptor's vnode. */
1661 NDINIT(&nd, LOOKUP, OP_OPEN, USEDVP | FOLLOW | AUDITVNPATH1, UIO_USERSPACE,
1662 fopen.o_pathname, &context);
1663 nd.ni_dvp = vp;
1664
1665 error = open1(&context, &nd, fopen.o_flags, &va,
1666 fileproc_alloc_init, NULL, retval);
1667
1668 vnode_put(vp);
1669 break;
1670 }
1671 /*
1672 * SPI (private) for unlinking a file starting from a dir fd
1673 */
1674 case F_UNLINKFROM: {
1675 user_addr_t pathname;
1676
1677 /* Check if this isn't a valid file descriptor */
1678 if ((fp->f_type != DTYPE_VNODE) ||
1679 (fp->f_flag & FREAD) == 0) {
1680 error = EBADF;
1681 goto out;
1682 }
1683 vp = (struct vnode *)fp->f_data;
1684 proc_fdunlock(p);
1685
1686 if (vnode_getwithref(vp)) {
1687 error = ENOENT;
1688 goto outdrop;
1689 }
1690
1691 /* Only valid for directories */
1692 if (vp->v_type != VDIR) {
1693 vnode_put(vp);
1694 error = ENOTDIR;
1695 goto outdrop;
1696 }
1697
1698 /* Get flags, mode and pathname arguments. */
1699 if (IS_64BIT_PROCESS(p)) {
1700 pathname = (user_addr_t)argp;
1701 } else {
1702 pathname = CAST_USER_ADDR_T(argp);
1703 }
1704
1705 /* Start the lookup relative to the file descriptor's vnode. */
1706 error = unlink1(&context, vp, pathname, UIO_USERSPACE, 0);
1707
1708 vnode_put(vp);
1709 break;
1710
1711 }
1712
1713 case F_ADDSIGS:
1714 case F_ADDFILESIGS:
1715 case F_ADDFILESIGS_FOR_DYLD_SIM:
1716 case F_ADDFILESIGS_RETURN:
1717 {
1718 struct cs_blob *blob = NULL;
1719 struct user_fsignatures fs;
1720 kern_return_t kr;
1721 vm_offset_t kernel_blob_addr;
1722 vm_size_t kernel_blob_size;
1723 int blob_add_flags = 0;
1724
1725 if (fp->f_type != DTYPE_VNODE) {
1726 error = EBADF;
1727 goto out;
1728 }
1729 vp = (struct vnode *)fp->f_data;
1730 proc_fdunlock(p);
1731
1732 if (uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM) {
1733 blob_add_flags |= MAC_VNODE_CHECK_DYLD_SIM;
1734 if ((p->p_csflags & CS_KILL) == 0) {
1735 proc_lock(p);
1736 p->p_csflags |= CS_KILL;
1737 proc_unlock(p);
1738 }
1739 }
1740
1741 error = vnode_getwithref(vp);
1742 if (error)
1743 goto outdrop;
1744
1745 if (IS_64BIT_PROCESS(p)) {
1746 error = copyin(argp, &fs, sizeof (fs));
1747 } else {
1748 struct user32_fsignatures fs32;
1749
1750 error = copyin(argp, &fs32, sizeof (fs32));
1751 fs.fs_file_start = fs32.fs_file_start;
1752 fs.fs_blob_start = CAST_USER_ADDR_T(fs32.fs_blob_start);
1753 fs.fs_blob_size = fs32.fs_blob_size;
1754 }
1755
1756 if (error) {
1757 vnode_put(vp);
1758 goto outdrop;
1759 }
1760
1761 /*
1762 * First check if we have something loaded a this offset
1763 */
1764 blob = ubc_cs_blob_get(vp, CPU_TYPE_ANY, fs.fs_file_start);
1765 if (blob != NULL)
1766 {
1767 /* If this is for dyld_sim revalidate the blob */
1768 if (uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM) {
1769 error = ubc_cs_blob_revalidate(vp, blob, blob_add_flags);
1770 }
1771
1772 } else {
1773 /*
1774 * An arbitrary limit, to prevent someone from mapping in a 20GB blob. This should cover
1775 * our use cases for the immediate future, but note that at the time of this commit, some
1776 * platforms are nearing 2MB blob sizes (with a prior soft limit of 2.5MB).
1777 *
1778 * We should consider how we can manage this more effectively; the above means that some
1779 * platforms are using megabytes of memory for signing data; it merely hasn't crossed the
1780 * threshold considered ridiculous at the time of this change.
1781 */
1782 #define CS_MAX_BLOB_SIZE (40ULL * 1024ULL * 1024ULL)
1783 if (fs.fs_blob_size > CS_MAX_BLOB_SIZE) {
1784 error = E2BIG;
1785 vnode_put(vp);
1786 goto outdrop;
1787 }
1788
1789 kernel_blob_size = CAST_DOWN(vm_size_t, fs.fs_blob_size);
1790 kr = ubc_cs_blob_allocate(&kernel_blob_addr, &kernel_blob_size);
1791 if (kr != KERN_SUCCESS) {
1792 error = ENOMEM;
1793 vnode_put(vp);
1794 goto outdrop;
1795 }
1796
1797 if(uap->cmd == F_ADDSIGS) {
1798 error = copyin(fs.fs_blob_start,
1799 (void *) kernel_blob_addr,
1800 kernel_blob_size);
1801 } else /* F_ADDFILESIGS || F_ADDFILESIGS_RETURN || F_ADDFILESIGS_FOR_DYLD_SIM */ {
1802 int resid;
1803
1804 error = vn_rdwr(UIO_READ,
1805 vp,
1806 (caddr_t) kernel_blob_addr,
1807 kernel_blob_size,
1808 fs.fs_file_start + fs.fs_blob_start,
1809 UIO_SYSSPACE,
1810 0,
1811 kauth_cred_get(),
1812 &resid,
1813 p);
1814 if ((error == 0) && resid) {
1815 /* kernel_blob_size rounded to a page size, but signature may be at end of file */
1816 memset((void *)(kernel_blob_addr + (kernel_blob_size - resid)), 0x0, resid);
1817 }
1818 }
1819
1820 if (error) {
1821 ubc_cs_blob_deallocate(kernel_blob_addr,
1822 kernel_blob_size);
1823 vnode_put(vp);
1824 goto outdrop;
1825 }
1826
1827 blob = NULL;
1828 error = ubc_cs_blob_add(vp,
1829 CPU_TYPE_ANY, /* not for a specific architecture */
1830 fs.fs_file_start,
1831 kernel_blob_addr,
1832 kernel_blob_size,
1833 blob_add_flags,
1834 &blob);
1835 if (error) {
1836 ubc_cs_blob_deallocate(kernel_blob_addr,
1837 kernel_blob_size);
1838 } else {
1839 /* ubc_blob_add() has consumed "kernel_blob_addr" */
1840 #if CHECK_CS_VALIDATION_BITMAP
1841 ubc_cs_validation_bitmap_allocate( vp );
1842 #endif
1843 }
1844 }
1845
1846 if (uap->cmd == F_ADDFILESIGS_RETURN || uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM) {
1847 /*
1848 * The first element of the structure is a
1849 * off_t that happen to have the same size for
1850 * all archs. Lets overwrite that.
1851 */
1852 off_t end_offset = 0;
1853 if (blob)
1854 end_offset = blob->csb_end_offset;
1855 error = copyout(&end_offset, argp, sizeof (end_offset));
1856 }
1857
1858 (void) vnode_put(vp);
1859 break;
1860 }
1861 case F_FINDSIGS: {
1862 error = ENOTSUP;
1863 goto out;
1864 }
1865 #if CONFIG_PROTECT
1866 case F_GETPROTECTIONCLASS: {
1867 int class = 0;
1868
1869 if (fp->f_type != DTYPE_VNODE) {
1870 error = EBADF;
1871 goto out;
1872 }
1873 vp = (struct vnode *)fp->f_data;
1874
1875 proc_fdunlock(p);
1876
1877 if (vnode_getwithref(vp)) {
1878 error = ENOENT;
1879 goto outdrop;
1880 }
1881
1882 error = cp_vnode_getclass (vp, &class);
1883 if (error == 0) {
1884 *retval = class;
1885 }
1886
1887 vnode_put(vp);
1888 break;
1889 }
1890
1891 case F_SETPROTECTIONCLASS: {
1892 /* tmp must be a valid PROTECTION_CLASS_* */
1893 tmp = CAST_DOWN_EXPLICIT(uint32_t, uap->arg);
1894
1895 if (fp->f_type != DTYPE_VNODE) {
1896 error = EBADF;
1897 goto out;
1898 }
1899 vp = (struct vnode *)fp->f_data;
1900
1901 proc_fdunlock(p);
1902
1903 if (vnode_getwithref(vp)) {
1904 error = ENOENT;
1905 goto outdrop;
1906 }
1907
1908 /* Only go forward if you have write access */
1909 vfs_context_t ctx = vfs_context_current();
1910 if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
1911 vnode_put(vp);
1912 error = EBADF;
1913 goto outdrop;
1914 }
1915 error = cp_vnode_setclass (vp, tmp);
1916 vnode_put(vp);
1917 break;
1918 }
1919
1920 case F_TRANSCODEKEY: {
1921
1922 char *backup_keyp = NULL;
1923 unsigned backup_key_len = CP_MAX_WRAPPEDKEYSIZE;
1924
1925 if (fp->f_type != DTYPE_VNODE) {
1926 error = EBADF;
1927 goto out;
1928 }
1929
1930 vp = (struct vnode *)fp->f_data;
1931 proc_fdunlock(p);
1932
1933 if (vnode_getwithref(vp)) {
1934 error = ENOENT;
1935 goto outdrop;
1936 }
1937
1938 MALLOC(backup_keyp, char *, backup_key_len, M_TEMP, M_WAITOK);
1939 if (backup_keyp == NULL) {
1940 error = ENOMEM;
1941 goto outdrop;
1942 }
1943
1944 error = cp_vnode_transcode (vp, backup_keyp, &backup_key_len);
1945 vnode_put(vp);
1946
1947 if (error == 0) {
1948 error = copyout((caddr_t)backup_keyp, argp, backup_key_len);
1949 *retval = backup_key_len;
1950 }
1951
1952 FREE(backup_keyp, M_TEMP);
1953
1954 break;
1955 }
1956
1957 case F_GETPROTECTIONLEVEL: {
1958 uint32_t cp_version = 0;
1959
1960 if (fp->f_type != DTYPE_VNODE) {
1961 error = EBADF;
1962 goto out;
1963 }
1964
1965 vp = (struct vnode*) fp->f_data;
1966 proc_fdunlock (p);
1967
1968 if (vnode_getwithref(vp)) {
1969 error = ENOENT;
1970 goto outdrop;
1971 }
1972
1973 /*
1974 * if cp_get_major_vers fails, error will be set to proper errno
1975 * and cp_version will still be 0.
1976 */
1977
1978 error = cp_get_root_major_vers (vp, &cp_version);
1979 *retval = cp_version;
1980
1981 vnode_put (vp);
1982 break;
1983 }
1984
1985 case F_GETDEFAULTPROTLEVEL: {
1986 uint32_t cp_default = 0;
1987
1988 if (fp->f_type != DTYPE_VNODE) {
1989 error = EBADF;
1990 goto out;
1991 }
1992
1993 vp = (struct vnode*) fp->f_data;
1994 proc_fdunlock (p);
1995
1996 if (vnode_getwithref(vp)) {
1997 error = ENOENT;
1998 goto outdrop;
1999 }
2000
2001 /*
2002 * if cp_get_major_vers fails, error will be set to proper errno
2003 * and cp_version will still be 0.
2004 */
2005
2006 error = cp_get_default_level(vp, &cp_default);
2007 *retval = cp_default;
2008
2009 vnode_put (vp);
2010 break;
2011 }
2012
2013
2014 #endif /* CONFIG_PROTECT */
2015
2016 case F_MOVEDATAEXTENTS: {
2017 struct fileproc *fp2 = NULL;
2018 struct vnode *src_vp = NULLVP;
2019 struct vnode *dst_vp = NULLVP;
2020 /* We need to grab the 2nd FD out of the argments before moving on. */
2021 int fd2 = CAST_DOWN_EXPLICIT(int32_t, uap->arg);
2022
2023 error = priv_check_cred(kauth_cred_get(), PRIV_VFS_MOVE_DATA_EXTENTS, 0);
2024 if (error)
2025 goto out;
2026
2027 if (fp->f_type != DTYPE_VNODE) {
2028 error = EBADF;
2029 goto out;
2030 }
2031
2032 /* For now, special case HFS+ only, since this is SPI. */
2033 src_vp = (struct vnode *)fp->f_data;
2034 if (src_vp->v_tag != VT_HFS) {
2035 error = EINVAL;
2036 goto out;
2037 }
2038
2039 /*
2040 * Get the references before we start acquiring iocounts on the vnodes,
2041 * while we still hold the proc fd lock
2042 */
2043 if ( (error = fp_lookup(p, fd2, &fp2, 1)) ) {
2044 error = EBADF;
2045 goto out;
2046 }
2047 if (fp2->f_type != DTYPE_VNODE) {
2048 fp_drop(p, fd2, fp2, 1);
2049 error = EBADF;
2050 goto out;
2051 }
2052 dst_vp = (struct vnode *)fp2->f_data;
2053 if (dst_vp->v_tag != VT_HFS) {
2054 fp_drop(p, fd2, fp2, 1);
2055 error = EINVAL;
2056 goto out;
2057 }
2058
2059 #if CONFIG_MACF
2060 /* Re-do MAC checks against the new FD, pass in a fake argument */
2061 error = mac_file_check_fcntl(proc_ucred(p), fp2->f_fglob, uap->cmd, 0);
2062 if (error) {
2063 fp_drop(p, fd2, fp2, 1);
2064 goto out;
2065 }
2066 #endif
2067 /* Audit the 2nd FD */
2068 AUDIT_ARG(fd, fd2);
2069
2070 proc_fdunlock(p);
2071
2072 if (vnode_getwithref(src_vp)) {
2073 fp_drop(p, fd2, fp2, 0);
2074 error = ENOENT;
2075 goto outdrop;
2076 }
2077 if (vnode_getwithref(dst_vp)) {
2078 vnode_put (src_vp);
2079 fp_drop(p, fd2, fp2, 0);
2080 error = ENOENT;
2081 goto outdrop;
2082 }
2083
2084 /*
2085 * Basic asserts; validate they are not the same and that
2086 * both live on the same filesystem.
2087 */
2088 if (dst_vp == src_vp) {
2089 vnode_put (src_vp);
2090 vnode_put (dst_vp);
2091 fp_drop (p, fd2, fp2, 0);
2092 error = EINVAL;
2093 goto outdrop;
2094 }
2095
2096 if (dst_vp->v_mount != src_vp->v_mount) {
2097 vnode_put (src_vp);
2098 vnode_put (dst_vp);
2099 fp_drop (p, fd2, fp2, 0);
2100 error = EXDEV;
2101 goto outdrop;
2102 }
2103
2104 /* Now we have a legit pair of FDs. Go to work */
2105
2106 /* Now check for write access to the target files */
2107 if(vnode_authorize(src_vp, NULLVP,
2108 (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) {
2109 vnode_put(src_vp);
2110 vnode_put(dst_vp);
2111 fp_drop(p, fd2, fp2, 0);
2112 error = EBADF;
2113 goto outdrop;
2114 }
2115
2116 if(vnode_authorize(dst_vp, NULLVP,
2117 (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) {
2118 vnode_put(src_vp);
2119 vnode_put(dst_vp);
2120 fp_drop(p, fd2, fp2, 0);
2121 error = EBADF;
2122 goto outdrop;
2123 }
2124
2125 /* Verify that both vps point to files and not directories */
2126 if ( !vnode_isreg(src_vp) || !vnode_isreg(dst_vp)) {
2127 error = EINVAL;
2128 vnode_put (src_vp);
2129 vnode_put (dst_vp);
2130 fp_drop (p, fd2, fp2, 0);
2131 goto outdrop;
2132 }
2133
2134 /*
2135 * The exchangedata syscall handler passes in 0 for the flags to VNOP_EXCHANGE.
2136 * We'll pass in our special bit indicating that the new behavior is expected
2137 */
2138
2139 error = VNOP_EXCHANGE(src_vp, dst_vp, FSOPT_EXCHANGE_DATA_ONLY, &context);
2140
2141 vnode_put (src_vp);
2142 vnode_put (dst_vp);
2143 fp_drop(p, fd2, fp2, 0);
2144 break;
2145 }
2146
2147 /*
2148 * SPI for making a file compressed.
2149 */
2150 case F_MAKECOMPRESSED: {
2151 uint32_t gcounter = CAST_DOWN_EXPLICIT(uint32_t, uap->arg);
2152
2153 if (fp->f_type != DTYPE_VNODE) {
2154 error = EBADF;
2155 goto out;
2156 }
2157
2158 vp = (struct vnode*) fp->f_data;
2159 proc_fdunlock (p);
2160
2161 /* get the vnode */
2162 if (vnode_getwithref(vp)) {
2163 error = ENOENT;
2164 goto outdrop;
2165 }
2166
2167 /* Is it a file? */
2168 if ((vnode_isreg(vp) == 0) && (vnode_islnk(vp) == 0)) {
2169 vnode_put(vp);
2170 error = EBADF;
2171 goto outdrop;
2172 }
2173
2174 /* invoke ioctl to pass off to FS */
2175 /* Only go forward if you have write access */
2176 vfs_context_t ctx = vfs_context_current();
2177 if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2178 vnode_put(vp);
2179 error = EBADF;
2180 goto outdrop;
2181 }
2182
2183 error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)&gcounter, 0, &context);
2184
2185 vnode_put (vp);
2186 break;
2187 }
2188
2189 /*
2190 * SPI (private) for indicating to a filesystem that subsequent writes to
2191 * the open FD will written to the Fastflow.
2192 */
2193 case F_SET_GREEDY_MODE:
2194 /* intentionally drop through to the same handler as F_SETSTATIC.
2195 * both fcntls should pass the argument and their selector into VNOP_IOCTL.
2196 */
2197
2198 /*
2199 * SPI (private) for indicating to a filesystem that subsequent writes to
2200 * the open FD will represent static content.
2201 */
2202 case F_SETSTATICCONTENT: {
2203 caddr_t ioctl_arg = NULL;
2204
2205 if (uap->arg) {
2206 ioctl_arg = (caddr_t) 1;
2207 }
2208
2209 if (fp->f_type != DTYPE_VNODE) {
2210 error = EBADF;
2211 goto out;
2212 }
2213 vp = (struct vnode *)fp->f_data;
2214 proc_fdunlock(p);
2215
2216 error = vnode_getwithref(vp);
2217 if (error) {
2218 error = ENOENT;
2219 goto outdrop;
2220 }
2221
2222 /* Only go forward if you have write access */
2223 vfs_context_t ctx = vfs_context_current();
2224 if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2225 vnode_put(vp);
2226 error = EBADF;
2227 goto outdrop;
2228 }
2229
2230 error = VNOP_IOCTL(vp, uap->cmd, ioctl_arg, 0, &context);
2231 (void)vnode_put(vp);
2232
2233 break;
2234 }
2235
2236 /*
2237 * SPI (private) for indicating to the lower level storage driver that the
2238 * subsequent writes should be of a particular IO type (burst, greedy, static),
2239 * or other flavors that may be necessary.
2240 */
2241 case F_SETIOTYPE: {
2242 caddr_t param_ptr;
2243 uint32_t param;
2244
2245 if (uap->arg) {
2246 /* extract 32 bits of flags from userland */
2247 param_ptr = (caddr_t) uap->arg;
2248 param = (uint32_t) param_ptr;
2249 }
2250 else {
2251 /* If no argument is specified, error out */
2252 error = EINVAL;
2253 goto out;
2254 }
2255
2256 /*
2257 * Validate the different types of flags that can be specified:
2258 * all of them are mutually exclusive for now.
2259 */
2260 switch (param) {
2261 case F_IOTYPE_ISOCHRONOUS:
2262 break;
2263
2264 default:
2265 error = EINVAL;
2266 goto out;
2267 }
2268
2269
2270 if (fp->f_type != DTYPE_VNODE) {
2271 error = EBADF;
2272 goto out;
2273 }
2274 vp = (struct vnode *)fp->f_data;
2275 proc_fdunlock(p);
2276
2277 error = vnode_getwithref(vp);
2278 if (error) {
2279 error = ENOENT;
2280 goto outdrop;
2281 }
2282
2283 /* Only go forward if you have write access */
2284 vfs_context_t ctx = vfs_context_current();
2285 if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2286 vnode_put(vp);
2287 error = EBADF;
2288 goto outdrop;
2289 }
2290
2291 error = VNOP_IOCTL(vp, uap->cmd, param_ptr, 0, &context);
2292 (void)vnode_put(vp);
2293
2294 break;
2295 }
2296
2297
2298 /*
2299 * Extract the CodeDirectory of the vnode associated with
2300 * the file descriptor and copy it back to user space
2301 */
2302 case F_GETCODEDIR: {
2303 struct user_fcodeblobs args;
2304
2305 if (fp->f_type != DTYPE_VNODE) {
2306 error = EBADF;
2307 goto out;
2308 }
2309
2310 vp = (struct vnode *)fp->f_data;
2311 proc_fdunlock(p);
2312
2313 if ((fp->f_flag & FREAD) == 0) {
2314 error = EBADF;
2315 goto outdrop;
2316 }
2317
2318 if (IS_64BIT_PROCESS(p)) {
2319 struct user64_fcodeblobs args64;
2320
2321 error = copyin(argp, &args64, sizeof(args64));
2322 if (error)
2323 goto outdrop;
2324
2325 args.f_cd_hash = args64.f_cd_hash;
2326 args.f_hash_size = args64.f_hash_size;
2327 args.f_cd_buffer = args64.f_cd_buffer;
2328 args.f_cd_size = args64.f_cd_size;
2329 args.f_out_size = args64.f_out_size;
2330 args.f_arch = args64.f_arch;
2331 } else {
2332 struct user32_fcodeblobs args32;
2333
2334 error = copyin(argp, &args32, sizeof(args32));
2335 if (error)
2336 goto outdrop;
2337
2338 args.f_cd_hash = CAST_USER_ADDR_T(args32.f_cd_hash);
2339 args.f_hash_size = args32.f_hash_size;
2340 args.f_cd_buffer = CAST_USER_ADDR_T(args32.f_cd_buffer);
2341 args.f_cd_size = args32.f_cd_size;
2342 args.f_out_size = CAST_USER_ADDR_T(args32.f_out_size);
2343 args.f_arch = args32.f_arch;
2344 }
2345
2346 if (vp->v_ubcinfo == NULL) {
2347 error = EINVAL;
2348 goto outdrop;
2349 }
2350
2351 struct cs_blob *t_blob = vp->v_ubcinfo->cs_blobs;
2352
2353 /*
2354 * This call fails if there is no cs_blob corresponding to the
2355 * vnode, or if there are multiple cs_blobs present, and the caller
2356 * did not specify which cpu_type they want the cs_blob for
2357 */
2358 if (t_blob == NULL) {
2359 error = ENOENT; /* there is no codesigning blob for this process */
2360 goto outdrop;
2361 } else if (args.f_arch == 0 && t_blob->csb_next != NULL) {
2362 error = ENOENT; /* too many architectures and none specified */
2363 goto outdrop;
2364 }
2365
2366 /* If the user specified an architecture, find the right blob */
2367 if (args.f_arch != 0) {
2368 while (t_blob) {
2369 if (t_blob->csb_cpu_type == args.f_arch)
2370 break;
2371 t_blob = t_blob->csb_next;
2372 }
2373 /* The cpu_type the user requested could not be found */
2374 if (t_blob == NULL) {
2375 error = ENOENT;
2376 goto outdrop;
2377 }
2378 }
2379
2380 const CS_SuperBlob *super_blob = (void *)t_blob->csb_mem_kaddr;
2381 const CS_CodeDirectory *cd = findCodeDirectory(super_blob,
2382 (const char *) super_blob,
2383 (const char *) super_blob + t_blob->csb_mem_size);
2384 if (cd == NULL) {
2385 error = ENOENT;
2386 goto outdrop;
2387 }
2388
2389 uint64_t buffer_size = ntohl(cd->length);
2390
2391 if (buffer_size > UINT_MAX) {
2392 error = ERANGE;
2393 goto outdrop;
2394 }
2395
2396 error = copyout(&buffer_size, args.f_out_size, sizeof(unsigned int));
2397 if (error)
2398 goto outdrop;
2399
2400 if (sizeof(t_blob->csb_cdhash) > args.f_hash_size ||
2401 buffer_size > args.f_cd_size) {
2402 error = ERANGE;
2403 goto outdrop;
2404 }
2405
2406 error = copyout(t_blob->csb_cdhash, args.f_cd_hash, sizeof(t_blob->csb_cdhash));
2407 if (error)
2408 goto outdrop;
2409 error = copyout(cd, args.f_cd_buffer, buffer_size);
2410 if (error)
2411 goto outdrop;
2412
2413 break;
2414 }
2415
2416 /*
2417 * Set the vnode pointed to by 'fd'
2418 * and tag it as the (potentially future) backing store
2419 * for another filesystem
2420 */
2421 case F_SETBACKINGSTORE: {
2422 if (fp->f_type != DTYPE_VNODE) {
2423 error = EBADF;
2424 goto out;
2425 }
2426
2427 vp = (struct vnode *)fp->f_data;
2428
2429 if (vp->v_tag != VT_HFS) {
2430 error = EINVAL;
2431 goto out;
2432 }
2433 proc_fdunlock(p);
2434
2435 if (vnode_getwithref(vp)) {
2436 error = ENOENT;
2437 goto outdrop;
2438 }
2439
2440 /* only proceed if you have write access */
2441 vfs_context_t ctx = vfs_context_current();
2442 if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2443 vnode_put(vp);
2444 error = EBADF;
2445 goto outdrop;
2446 }
2447
2448
2449 /* If arg != 0, set, otherwise unset */
2450 if (uap->arg) {
2451 error = VNOP_IOCTL (vp, uap->cmd, (caddr_t)1, 0, &context);
2452 }
2453 else {
2454 error = VNOP_IOCTL (vp, uap->cmd, (caddr_t)NULL, 0, &context);
2455 }
2456
2457 vnode_put(vp);
2458 break;
2459 }
2460
2461 /*
2462 * like F_GETPATH, but special semantics for
2463 * the mobile time machine handler.
2464 */
2465 case F_GETPATH_MTMINFO: {
2466 char *pathbufp;
2467 int pathlen;
2468
2469 if (fp->f_type != DTYPE_VNODE) {
2470 error = EBADF;
2471 goto out;
2472 }
2473 vp = (struct vnode *)fp->f_data;
2474 proc_fdunlock(p);
2475
2476 pathlen = MAXPATHLEN;
2477 MALLOC(pathbufp, char *, pathlen, M_TEMP, M_WAITOK);
2478 if (pathbufp == NULL) {
2479 error = ENOMEM;
2480 goto outdrop;
2481 }
2482 if ( (error = vnode_getwithref(vp)) == 0 ) {
2483 int backingstore = 0;
2484
2485 /* Check for error from vn_getpath before moving on */
2486 if ((error = vn_getpath(vp, pathbufp, &pathlen)) == 0) {
2487 if (vp->v_tag == VT_HFS) {
2488 error = VNOP_IOCTL (vp, uap->cmd, (caddr_t) &backingstore, 0, &context);
2489 }
2490 (void)vnode_put(vp);
2491
2492 if (error == 0) {
2493 error = copyout((caddr_t)pathbufp, argp, pathlen);
2494 }
2495 if (error == 0) {
2496 /*
2497 * If the copyout was successful, now check to ensure
2498 * that this vnode is not a BACKINGSTORE vnode. mtmd
2499 * wants the path regardless.
2500 */
2501 if (backingstore) {
2502 error = EBUSY;
2503 }
2504 }
2505 } else
2506 (void)vnode_put(vp);
2507 }
2508 FREE(pathbufp, M_TEMP);
2509 goto outdrop;
2510 }
2511
2512 #if DEBUG || DEVELOPMENT
2513 case F_RECYCLE:
2514 if (fp->f_type != DTYPE_VNODE) {
2515 error = EBADF;
2516 goto out;
2517 }
2518 vp = (struct vnode *)fp->f_data;
2519 proc_fdunlock(p);
2520
2521 vnode_recycle(vp);
2522 break;
2523 #endif
2524
2525 default:
2526 /*
2527 * This is an fcntl() that we d not recognize at this level;
2528 * if this is a vnode, we send it down into the VNOP_IOCTL
2529 * for this vnode; this can include special devices, and will
2530 * effectively overload fcntl() to send ioctl()'s.
2531 */
2532 if((uap->cmd & IOC_VOID) && (uap->cmd & IOC_INOUT)){
2533 error = EINVAL;
2534 goto out;
2535 }
2536
2537 /* Catch any now-invalid fcntl() selectors */
2538 switch (uap->cmd) {
2539 case F_MARKDEPENDENCY:
2540 error = EINVAL;
2541 goto out;
2542 default:
2543 break;
2544 }
2545
2546 if (fp->f_type != DTYPE_VNODE) {
2547 error = EBADF;
2548 goto out;
2549 }
2550 vp = (struct vnode *)fp->f_data;
2551 proc_fdunlock(p);
2552
2553 if ( (error = vnode_getwithref(vp)) == 0 ) {
2554 #define STK_PARAMS 128
2555 char stkbuf[STK_PARAMS];
2556 unsigned int size;
2557 caddr_t data, memp;
2558 /*
2559 * For this to work properly, we have to copy in the
2560 * ioctl() cmd argument if there is one; we must also
2561 * check that a command parameter, if present, does
2562 * not exceed the maximum command length dictated by
2563 * the number of bits we have available in the command
2564 * to represent a structure length. Finally, we have
2565 * to copy the results back out, if it is that type of
2566 * ioctl().
2567 */
2568 size = IOCPARM_LEN(uap->cmd);
2569 if (size > IOCPARM_MAX) {
2570 (void)vnode_put(vp);
2571 error = EINVAL;
2572 break;
2573 }
2574
2575 memp = NULL;
2576 if (size > sizeof (stkbuf)) {
2577 if ((memp = (caddr_t)kalloc(size)) == 0) {
2578 (void)vnode_put(vp);
2579 error = ENOMEM;
2580 goto outdrop;
2581 }
2582 data = memp;
2583 } else {
2584 data = &stkbuf[0];
2585 }
2586
2587 if (uap->cmd & IOC_IN) {
2588 if (size) {
2589 /* structure */
2590 error = copyin(argp, data, size);
2591 if (error) {
2592 (void)vnode_put(vp);
2593 if (memp)
2594 kfree(memp, size);
2595 goto outdrop;
2596 }
2597
2598 /* Bzero the section beyond that which was needed */
2599 if (size <= sizeof(stkbuf)) {
2600 bzero ( (((uint8_t*)data) + size), (sizeof(stkbuf) - size));
2601 }
2602 } else {
2603 /* int */
2604 if (is64bit) {
2605 *(user_addr_t *)data = argp;
2606 } else {
2607 *(uint32_t *)data = (uint32_t)argp;
2608 }
2609 };
2610 } else if ((uap->cmd & IOC_OUT) && size) {
2611 /*
2612 * Zero the buffer so the user always
2613 * gets back something deterministic.
2614 */
2615 bzero(data, size);
2616 } else if (uap->cmd & IOC_VOID) {
2617 if (is64bit) {
2618 *(user_addr_t *)data = argp;
2619 } else {
2620 *(uint32_t *)data = (uint32_t)argp;
2621 }
2622 }
2623
2624 error = VNOP_IOCTL(vp, uap->cmd, CAST_DOWN(caddr_t, data), 0, &context);
2625
2626 (void)vnode_put(vp);
2627
2628 /* Copy any output data to user */
2629 if (error == 0 && (uap->cmd & IOC_OUT) && size)
2630 error = copyout(data, argp, size);
2631 if (memp)
2632 kfree(memp, size);
2633 }
2634 break;
2635 }
2636
2637 outdrop:
2638 AUDIT_ARG(vnpath_withref, vp, ARG_VNODE1);
2639 fp_drop(p, fd, fp, 0);
2640 return(error);
2641 out:
2642 fp_drop(p, fd, fp, 1);
2643 proc_fdunlock(p);
2644 return(error);
2645 }
2646
2647
2648 /*
2649 * finishdup
2650 *
2651 * Description: Common code for dup, dup2, and fcntl(F_DUPFD).
2652 *
2653 * Parameters: p Process performing the dup
2654 * old The fd to dup
2655 * new The fd to dup it to
2656 * fd_flags Flags to augment the new fd
2657 * retval Pointer to the call return area
2658 *
2659 * Returns: 0 Success
2660 * EBADF
2661 * ENOMEM
2662 *
2663 * Implicit returns:
2664 * *retval (modified) The new descriptor
2665 *
2666 * Locks: Assumes proc_fdlock for process pointing to fdp is held by
2667 * the caller
2668 *
2669 * Notes: This function may drop and reacquire this lock; it is unsafe
2670 * for a caller to assume that other state protected by the lock
2671 * has not been subsequently changed out from under it.
2672 */
2673 int
2674 finishdup(proc_t p,
2675 struct filedesc *fdp, int old, int new, int fd_flags, int32_t *retval)
2676 {
2677 struct fileproc *nfp;
2678 struct fileproc *ofp;
2679 #if CONFIG_MACF
2680 int error;
2681 #endif
2682
2683 #if DIAGNOSTIC
2684 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
2685 #endif
2686 if ((ofp = fdp->fd_ofiles[old]) == NULL ||
2687 (fdp->fd_ofileflags[old] & UF_RESERVED)) {
2688 fdrelse(p, new);
2689 return (EBADF);
2690 }
2691 fg_ref(ofp);
2692
2693 #if CONFIG_MACF
2694 error = mac_file_check_dup(proc_ucred(p), ofp->f_fglob, new);
2695 if (error) {
2696 fg_drop(ofp);
2697 fdrelse(p, new);
2698 return (error);
2699 }
2700 #endif
2701
2702 proc_fdunlock(p);
2703
2704 nfp = fileproc_alloc_init(NULL);
2705
2706 proc_fdlock(p);
2707
2708 if (nfp == NULL) {
2709 fg_drop(ofp);
2710 fdrelse(p, new);
2711 return (ENOMEM);
2712 }
2713
2714 nfp->f_fglob = ofp->f_fglob;
2715
2716 #if DIAGNOSTIC
2717 if (fdp->fd_ofiles[new] != 0)
2718 panic("finishdup: overwriting fd_ofiles with new %d", new);
2719 if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0)
2720 panic("finishdup: unreserved fileflags with new %d", new);
2721 #endif
2722
2723 if (new > fdp->fd_lastfile)
2724 fdp->fd_lastfile = new;
2725 *fdflags(p, new) |= fd_flags;
2726 procfdtbl_releasefd(p, new, nfp);
2727 *retval = new;
2728 return (0);
2729 }
2730
2731
2732 /*
2733 * close
2734 *
2735 * Description: The implementation of the close(2) system call
2736 *
2737 * Parameters: p Process in whose per process file table
2738 * the close is to occur
2739 * uap->fd fd to be closed
2740 * retval <unused>
2741 *
2742 * Returns: 0 Success
2743 * fp_lookup:EBADF Bad file descriptor
2744 * fp_guard_exception:??? Guarded file descriptor
2745 * close_internal:EBADF
2746 * close_internal:??? Anything returnable by a per-fileops
2747 * close function
2748 */
2749 int
2750 close(proc_t p, struct close_args *uap, int32_t *retval)
2751 {
2752 __pthread_testcancel(1);
2753 return(close_nocancel(p, (struct close_nocancel_args *)uap, retval));
2754 }
2755
2756
2757 int
2758 close_nocancel(proc_t p, struct close_nocancel_args *uap, __unused int32_t *retval)
2759 {
2760 struct fileproc *fp;
2761 int fd = uap->fd;
2762 int error;
2763
2764 AUDIT_SYSCLOSE(p, fd);
2765
2766 proc_fdlock(p);
2767
2768 if ( (error = fp_lookup(p,fd,&fp, 1)) ) {
2769 proc_fdunlock(p);
2770 return(error);
2771 }
2772
2773 if (FP_ISGUARDED(fp, GUARD_CLOSE)) {
2774 error = fp_guard_exception(p, fd, fp, kGUARD_EXC_CLOSE);
2775 (void) fp_drop(p, fd, fp, 1);
2776 proc_fdunlock(p);
2777 return (error);
2778 }
2779
2780 error = close_internal_locked(p, fd, fp, 0);
2781
2782 proc_fdunlock(p);
2783
2784 return (error);
2785 }
2786
2787
2788 /*
2789 * close_internal_locked
2790 *
2791 * Close a file descriptor.
2792 *
2793 * Parameters: p Process in whose per process file table
2794 * the close is to occur
2795 * fd fd to be closed
2796 * fp fileproc associated with the fd
2797 *
2798 * Returns: 0 Success
2799 * EBADF fd already in close wait state
2800 * closef_locked:??? Anything returnable by a per-fileops
2801 * close function
2802 *
2803 * Locks: Assumes proc_fdlock for process is held by the caller and returns
2804 * with lock held
2805 *
2806 * Notes: This function may drop and reacquire this lock; it is unsafe
2807 * for a caller to assume that other state protected by the lock
2808 * has not been subsequently changed out from under it.
2809 */
2810 int
2811 close_internal_locked(proc_t p, int fd, struct fileproc *fp, int flags)
2812 {
2813 struct filedesc *fdp = p->p_fd;
2814 int error =0;
2815 int resvfd = flags & FD_DUP2RESV;
2816
2817
2818 #if DIAGNOSTIC
2819 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
2820 #endif
2821
2822 /* Keep people from using the filedesc while we are closing it */
2823 procfdtbl_markclosefd(p, fd);
2824
2825
2826 if ((fp->f_flags & FP_CLOSING) == FP_CLOSING) {
2827 panic("close_internal_locked: being called on already closing fd");
2828 }
2829
2830
2831 #if DIAGNOSTIC
2832 if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0)
2833 panic("close_internal: unreserved fileflags with fd %d", fd);
2834 #endif
2835
2836 fp->f_flags |= FP_CLOSING;
2837
2838 if ( (fp->f_flags & FP_AIOISSUED) || kauth_authorize_fileop_has_listeners() ) {
2839
2840 proc_fdunlock(p);
2841
2842 if ( (fp->f_type == DTYPE_VNODE) && kauth_authorize_fileop_has_listeners() ) {
2843 /*
2844 * call out to allow 3rd party notification of close.
2845 * Ignore result of kauth_authorize_fileop call.
2846 */
2847 if (vnode_getwithref((vnode_t)fp->f_data) == 0) {
2848 u_int fileop_flags = 0;
2849 if ((fp->f_flags & FP_WRITTEN) != 0)
2850 fileop_flags |= KAUTH_FILEOP_CLOSE_MODIFIED;
2851 kauth_authorize_fileop(fp->f_fglob->fg_cred, KAUTH_FILEOP_CLOSE,
2852 (uintptr_t)fp->f_data, (uintptr_t)fileop_flags);
2853 vnode_put((vnode_t)fp->f_data);
2854 }
2855 }
2856 if (fp->f_flags & FP_AIOISSUED)
2857 /*
2858 * cancel all async IO requests that can be cancelled.
2859 */
2860 _aio_close( p, fd );
2861
2862 proc_fdlock(p);
2863 }
2864
2865 if (fd < fdp->fd_knlistsize)
2866 knote_fdclose(p, fd);
2867
2868 if (fp->f_flags & FP_WAITEVENT)
2869 (void)waitevent_close(p, fp);
2870
2871 fileproc_drain(p, fp);
2872
2873 if (resvfd == 0) {
2874 _fdrelse(p, fd);
2875 } else {
2876 procfdtbl_reservefd(p, fd);
2877 }
2878
2879 if (ENTR_SHOULDTRACE && fp->f_type == DTYPE_SOCKET)
2880 KERNEL_ENERGYTRACE(kEnTrActKernSocket, DBG_FUNC_END,
2881 fd, 0, (int64_t)VM_KERNEL_ADDRPERM(fp->f_data));
2882
2883 error = closef_locked(fp, fp->f_fglob, p);
2884 if ((fp->f_flags & FP_WAITCLOSE) == FP_WAITCLOSE)
2885 wakeup(&fp->f_flags);
2886 fp->f_flags &= ~(FP_WAITCLOSE | FP_CLOSING);
2887
2888 proc_fdunlock(p);
2889
2890 fileproc_free(fp);
2891
2892 proc_fdlock(p);
2893
2894 #if DIAGNOSTIC
2895 if (resvfd != 0) {
2896 if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0)
2897 panic("close with reserved fd returns with freed fd:%d: proc: %p", fd, p);
2898 }
2899 #endif
2900
2901 return(error);
2902 }
2903
2904
2905 /*
2906 * fstat1
2907 *
2908 * Description: Return status information about a file descriptor.
2909 *
2910 * Parameters: p The process doing the fstat
2911 * fd The fd to stat
2912 * ub The user stat buffer
2913 * xsecurity The user extended security
2914 * buffer, or 0 if none
2915 * xsecurity_size The size of xsecurity, or 0
2916 * if no xsecurity
2917 * isstat64 Flag to indicate 64 bit version
2918 * for inode size, etc.
2919 *
2920 * Returns: 0 Success
2921 * EBADF
2922 * EFAULT
2923 * fp_lookup:EBADF Bad file descriptor
2924 * vnode_getwithref:???
2925 * copyout:EFAULT
2926 * vnode_getwithref:???
2927 * vn_stat:???
2928 * soo_stat:???
2929 * pipe_stat:???
2930 * pshm_stat:???
2931 * kqueue_stat:???
2932 *
2933 * Notes: Internal implementation for all other fstat() related
2934 * functions
2935 *
2936 * XXX switch on node type is bogus; need a stat in struct
2937 * XXX fileops instead.
2938 */
2939 static int
2940 fstat1(proc_t p, int fd, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsecurity_size, int isstat64)
2941 {
2942 struct fileproc *fp;
2943 union {
2944 struct stat sb;
2945 struct stat64 sb64;
2946 } source;
2947 union {
2948 struct user64_stat user64_sb;
2949 struct user32_stat user32_sb;
2950 struct user64_stat64 user64_sb64;
2951 struct user32_stat64 user32_sb64;
2952 } dest;
2953 int error, my_size;
2954 file_type_t type;
2955 caddr_t data;
2956 kauth_filesec_t fsec;
2957 user_size_t xsecurity_bufsize;
2958 vfs_context_t ctx = vfs_context_current();
2959 void * sbptr;
2960
2961
2962 AUDIT_ARG(fd, fd);
2963
2964 if ((error = fp_lookup(p, fd, &fp, 0)) != 0) {
2965 return(error);
2966 }
2967 type = fp->f_type;
2968 data = fp->f_data;
2969 fsec = KAUTH_FILESEC_NONE;
2970
2971 sbptr = (void *)&source;
2972
2973 switch (type) {
2974
2975 case DTYPE_VNODE:
2976 if ((error = vnode_getwithref((vnode_t)data)) == 0) {
2977 /*
2978 * If the caller has the file open, and is not
2979 * requesting extended security information, we are
2980 * going to let them get the basic stat information.
2981 */
2982 if (xsecurity == USER_ADDR_NULL) {
2983 error = vn_stat_noauth((vnode_t)data, sbptr, NULL, isstat64, ctx);
2984 } else {
2985 error = vn_stat((vnode_t)data, sbptr, &fsec, isstat64, ctx);
2986 }
2987
2988 AUDIT_ARG(vnpath, (struct vnode *)data, ARG_VNODE1);
2989 (void)vnode_put((vnode_t)data);
2990 }
2991 break;
2992
2993 #if SOCKETS
2994 case DTYPE_SOCKET:
2995 error = soo_stat((struct socket *)data, sbptr, isstat64);
2996 break;
2997 #endif /* SOCKETS */
2998
2999 case DTYPE_PIPE:
3000 error = pipe_stat((void *)data, sbptr, isstat64);
3001 break;
3002
3003 case DTYPE_PSXSHM:
3004 error = pshm_stat((void *)data, sbptr, isstat64);
3005 break;
3006
3007 case DTYPE_KQUEUE:
3008 error = kqueue_stat((void *)data, sbptr, isstat64, p);
3009 break;
3010
3011 default:
3012 error = EBADF;
3013 goto out;
3014 }
3015 if (error == 0) {
3016 caddr_t sbp;
3017
3018 if (isstat64 != 0) {
3019 source.sb64.st_lspare = 0;
3020 source.sb64.st_qspare[0] = 0LL;
3021 source.sb64.st_qspare[1] = 0LL;
3022
3023 if (IS_64BIT_PROCESS(current_proc())) {
3024 munge_user64_stat64(&source.sb64, &dest.user64_sb64);
3025 my_size = sizeof(dest.user64_sb64);
3026 sbp = (caddr_t)&dest.user64_sb64;
3027 } else {
3028 munge_user32_stat64(&source.sb64, &dest.user32_sb64);
3029 my_size = sizeof(dest.user32_sb64);
3030 sbp = (caddr_t)&dest.user32_sb64;
3031 }
3032 } else {
3033 source.sb.st_lspare = 0;
3034 source.sb.st_qspare[0] = 0LL;
3035 source.sb.st_qspare[1] = 0LL;
3036 if (IS_64BIT_PROCESS(current_proc())) {
3037 munge_user64_stat(&source.sb, &dest.user64_sb);
3038 my_size = sizeof(dest.user64_sb);
3039 sbp = (caddr_t)&dest.user64_sb;
3040 } else {
3041 munge_user32_stat(&source.sb, &dest.user32_sb);
3042 my_size = sizeof(dest.user32_sb);
3043 sbp = (caddr_t)&dest.user32_sb;
3044 }
3045 }
3046
3047 error = copyout(sbp, ub, my_size);
3048 }
3049
3050 /* caller wants extended security information? */
3051 if (xsecurity != USER_ADDR_NULL) {
3052
3053 /* did we get any? */
3054 if (fsec == KAUTH_FILESEC_NONE) {
3055 if (susize(xsecurity_size, 0) != 0) {
3056 error = EFAULT;
3057 goto out;
3058 }
3059 } else {
3060 /* find the user buffer size */
3061 xsecurity_bufsize = fusize(xsecurity_size);
3062
3063 /* copy out the actual data size */
3064 if (susize(xsecurity_size, KAUTH_FILESEC_COPYSIZE(fsec)) != 0) {
3065 error = EFAULT;
3066 goto out;
3067 }
3068
3069 /* if the caller supplied enough room, copy out to it */
3070 if (xsecurity_bufsize >= KAUTH_FILESEC_COPYSIZE(fsec))
3071 error = copyout(fsec, xsecurity, KAUTH_FILESEC_COPYSIZE(fsec));
3072 }
3073 }
3074 out:
3075 fp_drop(p, fd, fp, 0);
3076 if (fsec != NULL)
3077 kauth_filesec_free(fsec);
3078 return (error);
3079 }
3080
3081
3082 /*
3083 * fstat_extended
3084 *
3085 * Description: Extended version of fstat supporting returning extended
3086 * security information
3087 *
3088 * Parameters: p The process doing the fstat
3089 * uap->fd The fd to stat
3090 * uap->ub The user stat buffer
3091 * uap->xsecurity The user extended security
3092 * buffer, or 0 if none
3093 * uap->xsecurity_size The size of xsecurity, or 0
3094 *
3095 * Returns: 0 Success
3096 * !0 Errno (see fstat1)
3097 */
3098 int
3099 fstat_extended(proc_t p, struct fstat_extended_args *uap, __unused int32_t *retval)
3100 {
3101 return(fstat1(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 0));
3102 }
3103
3104
3105 /*
3106 * fstat
3107 *
3108 * Description: Get file status for the file associated with fd
3109 *
3110 * Parameters: p The process doing the fstat
3111 * uap->fd The fd to stat
3112 * uap->ub The user stat buffer
3113 *
3114 * Returns: 0 Success
3115 * !0 Errno (see fstat1)
3116 */
3117 int
3118 fstat(proc_t p, register struct fstat_args *uap, __unused int32_t *retval)
3119 {
3120 return(fstat1(p, uap->fd, uap->ub, 0, 0, 0));
3121 }
3122
3123
3124 /*
3125 * fstat64_extended
3126 *
3127 * Description: Extended version of fstat64 supporting returning extended
3128 * security information
3129 *
3130 * Parameters: p The process doing the fstat
3131 * uap->fd The fd to stat
3132 * uap->ub The user stat buffer
3133 * uap->xsecurity The user extended security
3134 * buffer, or 0 if none
3135 * uap->xsecurity_size The size of xsecurity, or 0
3136 *
3137 * Returns: 0 Success
3138 * !0 Errno (see fstat1)
3139 */
3140 int
3141 fstat64_extended(proc_t p, struct fstat64_extended_args *uap, __unused int32_t *retval)
3142 {
3143 return(fstat1(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 1));
3144 }
3145
3146
3147 /*
3148 * fstat64
3149 *
3150 * Description: Get 64 bit version of the file status for the file associated
3151 * with fd
3152 *
3153 * Parameters: p The process doing the fstat
3154 * uap->fd The fd to stat
3155 * uap->ub The user stat buffer
3156 *
3157 * Returns: 0 Success
3158 * !0 Errno (see fstat1)
3159 */
3160 int
3161 fstat64(proc_t p, register struct fstat64_args *uap, __unused int32_t *retval)
3162 {
3163 return(fstat1(p, uap->fd, uap->ub, 0, 0, 1));
3164 }
3165
3166
3167 /*
3168 * fpathconf
3169 *
3170 * Description: Return pathconf information about a file descriptor.
3171 *
3172 * Parameters: p Process making the request
3173 * uap->fd fd to get information about
3174 * uap->name Name of information desired
3175 * retval Pointer to the call return area
3176 *
3177 * Returns: 0 Success
3178 * EINVAL
3179 * fp_lookup:EBADF Bad file descriptor
3180 * vnode_getwithref:???
3181 * vn_pathconf:???
3182 *
3183 * Implicit returns:
3184 * *retval (modified) Returned information (numeric)
3185 */
3186 int
3187 fpathconf(proc_t p, struct fpathconf_args *uap, int32_t *retval)
3188 {
3189 int fd = uap->fd;
3190 struct fileproc *fp;
3191 struct vnode *vp;
3192 int error = 0;
3193 file_type_t type;
3194 caddr_t data;
3195
3196
3197 AUDIT_ARG(fd, uap->fd);
3198 if ( (error = fp_lookup(p, fd, &fp, 0)) )
3199 return(error);
3200 type = fp->f_type;
3201 data = fp->f_data;
3202
3203 switch (type) {
3204
3205 case DTYPE_SOCKET:
3206 if (uap->name != _PC_PIPE_BUF) {
3207 error = EINVAL;
3208 goto out;
3209 }
3210 *retval = PIPE_BUF;
3211 error = 0;
3212 goto out;
3213
3214 case DTYPE_PIPE:
3215 if (uap->name != _PC_PIPE_BUF) {
3216 error = EINVAL;
3217 goto out;
3218 }
3219 *retval = PIPE_BUF;
3220 error = 0;
3221 goto out;
3222
3223 case DTYPE_VNODE:
3224 vp = (struct vnode *)data;
3225
3226 if ( (error = vnode_getwithref(vp)) == 0) {
3227 AUDIT_ARG(vnpath, vp, ARG_VNODE1);
3228
3229 error = vn_pathconf(vp, uap->name, retval, vfs_context_current());
3230
3231 (void)vnode_put(vp);
3232 }
3233 goto out;
3234
3235 default:
3236 error = EINVAL;
3237 goto out;
3238
3239 }
3240 /*NOTREACHED*/
3241 out:
3242 fp_drop(p, fd, fp, 0);
3243 return(error);
3244 }
3245
3246 /*
3247 * Statistics counter for the number of times a process calling fdalloc()
3248 * has resulted in an expansion of the per process open file table.
3249 *
3250 * XXX This would likely be of more use if it were per process
3251 */
3252 int fdexpand;
3253
3254
3255 /*
3256 * fdalloc
3257 *
3258 * Description: Allocate a file descriptor for the process.
3259 *
3260 * Parameters: p Process to allocate the fd in
3261 * want The fd we would prefer to get
3262 * result Pointer to fd we got
3263 *
3264 * Returns: 0 Success
3265 * EMFILE
3266 * ENOMEM
3267 *
3268 * Implicit returns:
3269 * *result (modified) The fd which was allocated
3270 */
3271 int
3272 fdalloc(proc_t p, int want, int *result)
3273 {
3274 struct filedesc *fdp = p->p_fd;
3275 int i;
3276 int lim, last, numfiles, oldnfiles;
3277 struct fileproc **newofiles, **ofiles;
3278 char *newofileflags;
3279
3280 /*
3281 * Search for a free descriptor starting at the higher
3282 * of want or fd_freefile. If that fails, consider
3283 * expanding the ofile array.
3284 */
3285 #if DIAGNOSTIC
3286 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
3287 #endif
3288
3289 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
3290 for (;;) {
3291 last = min(fdp->fd_nfiles, lim);
3292 if ((i = want) < fdp->fd_freefile)
3293 i = fdp->fd_freefile;
3294 for (; i < last; i++) {
3295 if (fdp->fd_ofiles[i] == NULL && !(fdp->fd_ofileflags[i] & UF_RESERVED)) {
3296 procfdtbl_reservefd(p, i);
3297 if (i > fdp->fd_lastfile)
3298 fdp->fd_lastfile = i;
3299 if (want <= fdp->fd_freefile)
3300 fdp->fd_freefile = i;
3301 *result = i;
3302 return (0);
3303 }
3304 }
3305
3306 /*
3307 * No space in current array. Expand?
3308 */
3309 if (fdp->fd_nfiles >= lim)
3310 return (EMFILE);
3311 if (fdp->fd_nfiles < NDEXTENT)
3312 numfiles = NDEXTENT;
3313 else
3314 numfiles = 2 * fdp->fd_nfiles;
3315 /* Enforce lim */
3316 if (numfiles > lim)
3317 numfiles = lim;
3318 proc_fdunlock(p);
3319 MALLOC_ZONE(newofiles, struct fileproc **,
3320 numfiles * OFILESIZE, M_OFILETABL, M_WAITOK);
3321 proc_fdlock(p);
3322 if (newofiles == NULL) {
3323 return (ENOMEM);
3324 }
3325 if (fdp->fd_nfiles >= numfiles) {
3326 FREE_ZONE(newofiles, numfiles * OFILESIZE, M_OFILETABL);
3327 continue;
3328 }
3329 newofileflags = (char *) &newofiles[numfiles];
3330 /*
3331 * Copy the existing ofile and ofileflags arrays
3332 * and zero the new portion of each array.
3333 */
3334 oldnfiles = fdp->fd_nfiles;
3335 (void) memcpy(newofiles, fdp->fd_ofiles,
3336 oldnfiles * sizeof(*fdp->fd_ofiles));
3337 (void) memset(&newofiles[oldnfiles], 0,
3338 (numfiles - oldnfiles) * sizeof(*fdp->fd_ofiles));
3339
3340 (void) memcpy(newofileflags, fdp->fd_ofileflags,
3341 oldnfiles * sizeof(*fdp->fd_ofileflags));
3342 (void) memset(&newofileflags[oldnfiles], 0,
3343 (numfiles - oldnfiles) *
3344 sizeof(*fdp->fd_ofileflags));
3345 ofiles = fdp->fd_ofiles;
3346 fdp->fd_ofiles = newofiles;
3347 fdp->fd_ofileflags = newofileflags;
3348 fdp->fd_nfiles = numfiles;
3349 FREE_ZONE(ofiles, oldnfiles * OFILESIZE, M_OFILETABL);
3350 fdexpand++;
3351 }
3352 }
3353
3354
3355 /*
3356 * fdavail
3357 *
3358 * Description: Check to see whether n user file descriptors are available
3359 * to the process p.
3360 *
3361 * Parameters: p Process to check in
3362 * n The number of fd's desired
3363 *
3364 * Returns: 0 No
3365 * 1 Yes
3366 *
3367 * Locks: Assumes proc_fdlock for process is held by the caller
3368 *
3369 * Notes: The answer only remains valid so long as the proc_fdlock is
3370 * held by the caller.
3371 */
3372 int
3373 fdavail(proc_t p, int n)
3374 {
3375 struct filedesc *fdp = p->p_fd;
3376 struct fileproc **fpp;
3377 char *flags;
3378 int i, lim;
3379
3380 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
3381 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
3382 return (1);
3383 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
3384 flags = &fdp->fd_ofileflags[fdp->fd_freefile];
3385 for (i = fdp->fd_nfiles - fdp->fd_freefile; --i >= 0; fpp++, flags++)
3386 if (*fpp == NULL && !(*flags & UF_RESERVED) && --n <= 0)
3387 return (1);
3388 return (0);
3389 }
3390
3391
3392 /*
3393 * fdrelse
3394 *
3395 * Description: Legacy KPI wrapper function for _fdrelse
3396 *
3397 * Parameters: p Process in which fd lives
3398 * fd fd to free
3399 *
3400 * Returns: void
3401 *
3402 * Locks: Assumes proc_fdlock for process is held by the caller
3403 */
3404 void
3405 fdrelse(proc_t p, int fd)
3406 {
3407 _fdrelse(p, fd);
3408 }
3409
3410
3411 /*
3412 * fdgetf_noref
3413 *
3414 * Description: Get the fileproc pointer for the given fd from the per process
3415 * open file table without taking an explicit reference on it.
3416 *
3417 * Parameters: p Process containing fd
3418 * fd fd to obtain fileproc for
3419 * resultfp Pointer to pointer return area
3420 *
3421 * Returns: 0 Success
3422 * EBADF
3423 *
3424 * Implicit returns:
3425 * *resultfp (modified) Pointer to fileproc pointer
3426 *
3427 * Locks: Assumes proc_fdlock for process is held by the caller
3428 *
3429 * Notes: Because there is no reference explicitly taken, the returned
3430 * fileproc pointer is only valid so long as the proc_fdlock
3431 * remains held by the caller.
3432 */
3433 int
3434 fdgetf_noref(proc_t p, int fd, struct fileproc **resultfp)
3435 {
3436 struct filedesc *fdp = p->p_fd;
3437 struct fileproc *fp;
3438
3439 if (fd < 0 || fd >= fdp->fd_nfiles ||
3440 (fp = fdp->fd_ofiles[fd]) == NULL ||
3441 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3442 return (EBADF);
3443 }
3444 if (resultfp)
3445 *resultfp = fp;
3446 return (0);
3447 }
3448
3449
3450 /*
3451 * fp_getfvp
3452 *
3453 * Description: Get fileproc and vnode pointer for a given fd from the per
3454 * process open file table of the specified process, and if
3455 * successful, increment the f_iocount
3456 *
3457 * Parameters: p Process in which fd lives
3458 * fd fd to get information for
3459 * resultfp Pointer to result fileproc
3460 * pointer area, or 0 if none
3461 * resultvp Pointer to result vnode pointer
3462 * area, or 0 if none
3463 *
3464 * Returns: 0 Success
3465 * EBADF Bad file descriptor
3466 * ENOTSUP fd does not refer to a vnode
3467 *
3468 * Implicit returns:
3469 * *resultfp (modified) Fileproc pointer
3470 * *resultvp (modified) vnode pointer
3471 *
3472 * Notes: The resultfp and resultvp fields are optional, and may be
3473 * independently specified as NULL to skip returning information
3474 *
3475 * Locks: Internally takes and releases proc_fdlock
3476 */
3477 int
3478 fp_getfvp(proc_t p, int fd, struct fileproc **resultfp, struct vnode **resultvp)
3479 {
3480 struct filedesc *fdp = p->p_fd;
3481 struct fileproc *fp;
3482
3483 proc_fdlock_spin(p);
3484 if (fd < 0 || fd >= fdp->fd_nfiles ||
3485 (fp = fdp->fd_ofiles[fd]) == NULL ||
3486 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3487 proc_fdunlock(p);
3488 return (EBADF);
3489 }
3490 if (fp->f_type != DTYPE_VNODE) {
3491 proc_fdunlock(p);
3492 return(ENOTSUP);
3493 }
3494 fp->f_iocount++;
3495
3496 if (resultfp)
3497 *resultfp = fp;
3498 if (resultvp)
3499 *resultvp = (struct vnode *)fp->f_data;
3500 proc_fdunlock(p);
3501
3502 return (0);
3503 }
3504
3505
3506 /*
3507 * fp_getfvpandvid
3508 *
3509 * Description: Get fileproc, vnode pointer, and vid for a given fd from the
3510 * per process open file table of the specified process, and if
3511 * successful, increment the f_iocount
3512 *
3513 * Parameters: p Process in which fd lives
3514 * fd fd to get information for
3515 * resultfp Pointer to result fileproc
3516 * pointer area, or 0 if none
3517 * resultvp Pointer to result vnode pointer
3518 * area, or 0 if none
3519 * vidp Pointer to resuld vid area
3520 *
3521 * Returns: 0 Success
3522 * EBADF Bad file descriptor
3523 * ENOTSUP fd does not refer to a vnode
3524 *
3525 * Implicit returns:
3526 * *resultfp (modified) Fileproc pointer
3527 * *resultvp (modified) vnode pointer
3528 * *vidp vid value
3529 *
3530 * Notes: The resultfp and resultvp fields are optional, and may be
3531 * independently specified as NULL to skip returning information
3532 *
3533 * Locks: Internally takes and releases proc_fdlock
3534 */
3535 int
3536 fp_getfvpandvid(proc_t p, int fd, struct fileproc **resultfp,
3537 struct vnode **resultvp, uint32_t *vidp)
3538 {
3539 struct filedesc *fdp = p->p_fd;
3540 struct fileproc *fp;
3541
3542 proc_fdlock_spin(p);
3543 if (fd < 0 || fd >= fdp->fd_nfiles ||
3544 (fp = fdp->fd_ofiles[fd]) == NULL ||
3545 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3546 proc_fdunlock(p);
3547 return (EBADF);
3548 }
3549 if (fp->f_type != DTYPE_VNODE) {
3550 proc_fdunlock(p);
3551 return(ENOTSUP);
3552 }
3553 fp->f_iocount++;
3554
3555 if (resultfp)
3556 *resultfp = fp;
3557 if (resultvp)
3558 *resultvp = (struct vnode *)fp->f_data;
3559 if (vidp)
3560 *vidp = (uint32_t)vnode_vid((struct vnode *)fp->f_data);
3561 proc_fdunlock(p);
3562
3563 return (0);
3564 }
3565
3566
3567 /*
3568 * fp_getfsock
3569 *
3570 * Description: Get fileproc and socket pointer for a given fd from the
3571 * per process open file table of the specified process, and if
3572 * successful, increment the f_iocount
3573 *
3574 * Parameters: p Process in which fd lives
3575 * fd fd to get information for
3576 * resultfp Pointer to result fileproc
3577 * pointer area, or 0 if none
3578 * results Pointer to result socket
3579 * pointer area, or 0 if none
3580 *
3581 * Returns: EBADF The file descriptor is invalid
3582 * EOPNOTSUPP The file descriptor is not a socket
3583 * 0 Success
3584 *
3585 * Implicit returns:
3586 * *resultfp (modified) Fileproc pointer
3587 * *results (modified) socket pointer
3588 *
3589 * Notes: EOPNOTSUPP should probably be ENOTSOCK; this function is only
3590 * ever called from accept1().
3591 */
3592 int
3593 fp_getfsock(proc_t p, int fd, struct fileproc **resultfp,
3594 struct socket **results)
3595 {
3596 struct filedesc *fdp = p->p_fd;
3597 struct fileproc *fp;
3598
3599 proc_fdlock_spin(p);
3600 if (fd < 0 || fd >= fdp->fd_nfiles ||
3601 (fp = fdp->fd_ofiles[fd]) == NULL ||
3602 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3603 proc_fdunlock(p);
3604 return (EBADF);
3605 }
3606 if (fp->f_type != DTYPE_SOCKET) {
3607 proc_fdunlock(p);
3608 return(EOPNOTSUPP);
3609 }
3610 fp->f_iocount++;
3611
3612 if (resultfp)
3613 *resultfp = fp;
3614 if (results)
3615 *results = (struct socket *)fp->f_data;
3616 proc_fdunlock(p);
3617
3618 return (0);
3619 }
3620
3621
3622 /*
3623 * fp_getfkq
3624 *
3625 * Description: Get fileproc and kqueue pointer for a given fd from the
3626 * per process open file table of the specified process, and if
3627 * successful, increment the f_iocount
3628 *
3629 * Parameters: p Process in which fd lives
3630 * fd fd to get information for
3631 * resultfp Pointer to result fileproc
3632 * pointer area, or 0 if none
3633 * resultkq Pointer to result kqueue
3634 * pointer area, or 0 if none
3635 *
3636 * Returns: EBADF The file descriptor is invalid
3637 * EBADF The file descriptor is not a socket
3638 * 0 Success
3639 *
3640 * Implicit returns:
3641 * *resultfp (modified) Fileproc pointer
3642 * *resultkq (modified) kqueue pointer
3643 *
3644 * Notes: The second EBADF should probably be something else to make
3645 * the error condition distinct.
3646 */
3647 int
3648 fp_getfkq(proc_t p, int fd, struct fileproc **resultfp,
3649 struct kqueue **resultkq)
3650 {
3651 struct filedesc *fdp = p->p_fd;
3652 struct fileproc *fp;
3653
3654 proc_fdlock_spin(p);
3655 if ( fd < 0 || fd >= fdp->fd_nfiles ||
3656 (fp = fdp->fd_ofiles[fd]) == NULL ||
3657 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3658 proc_fdunlock(p);
3659 return (EBADF);
3660 }
3661 if (fp->f_type != DTYPE_KQUEUE) {
3662 proc_fdunlock(p);
3663 return(EBADF);
3664 }
3665 fp->f_iocount++;
3666
3667 if (resultfp)
3668 *resultfp = fp;
3669 if (resultkq)
3670 *resultkq = (struct kqueue *)fp->f_data;
3671 proc_fdunlock(p);
3672
3673 return (0);
3674 }
3675
3676
3677 /*
3678 * fp_getfpshm
3679 *
3680 * Description: Get fileproc and POSIX shared memory pointer for a given fd
3681 * from the per process open file table of the specified process
3682 * and if successful, increment the f_iocount
3683 *
3684 * Parameters: p Process in which fd lives
3685 * fd fd to get information for
3686 * resultfp Pointer to result fileproc
3687 * pointer area, or 0 if none
3688 * resultpshm Pointer to result POSIX
3689 * shared memory pointer
3690 * pointer area, or 0 if none
3691 *
3692 * Returns: EBADF The file descriptor is invalid
3693 * EBADF The file descriptor is not a POSIX
3694 * shared memory area
3695 * 0 Success
3696 *
3697 * Implicit returns:
3698 * *resultfp (modified) Fileproc pointer
3699 * *resultpshm (modified) POSIX shared memory pointer
3700 *
3701 * Notes: The second EBADF should probably be something else to make
3702 * the error condition distinct.
3703 */
3704 int
3705 fp_getfpshm(proc_t p, int fd, struct fileproc **resultfp,
3706 struct pshmnode **resultpshm)
3707 {
3708 struct filedesc *fdp = p->p_fd;
3709 struct fileproc *fp;
3710
3711 proc_fdlock_spin(p);
3712 if (fd < 0 || fd >= fdp->fd_nfiles ||
3713 (fp = fdp->fd_ofiles[fd]) == NULL ||
3714 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3715 proc_fdunlock(p);
3716 return (EBADF);
3717 }
3718 if (fp->f_type != DTYPE_PSXSHM) {
3719
3720 proc_fdunlock(p);
3721 return(EBADF);
3722 }
3723 fp->f_iocount++;
3724
3725 if (resultfp)
3726 *resultfp = fp;
3727 if (resultpshm)
3728 *resultpshm = (struct pshmnode *)fp->f_data;
3729 proc_fdunlock(p);
3730
3731 return (0);
3732 }
3733
3734
3735 /*
3736 * fp_getfsem
3737 *
3738 * Description: Get fileproc and POSIX semaphore pointer for a given fd from
3739 * the per process open file table of the specified process
3740 * and if successful, increment the f_iocount
3741 *
3742 * Parameters: p Process in which fd lives
3743 * fd fd to get information for
3744 * resultfp Pointer to result fileproc
3745 * pointer area, or 0 if none
3746 * resultpsem Pointer to result POSIX
3747 * semaphore pointer area, or
3748 * 0 if none
3749 *
3750 * Returns: EBADF The file descriptor is invalid
3751 * EBADF The file descriptor is not a POSIX
3752 * semaphore
3753 * 0 Success
3754 *
3755 * Implicit returns:
3756 * *resultfp (modified) Fileproc pointer
3757 * *resultpsem (modified) POSIX semaphore pointer
3758 *
3759 * Notes: The second EBADF should probably be something else to make
3760 * the error condition distinct.
3761 *
3762 * In order to support unnamed POSIX semaphores, the named
3763 * POSIX semaphores will have to move out of the per-process
3764 * open filetable, and into a global table that is shared with
3765 * unnamed POSIX semaphores, since unnamed POSIX semaphores
3766 * are typically used by declaring instances in shared memory,
3767 * and there's no other way to do this without changing the
3768 * underlying type, which would introduce binary compatibility
3769 * issues.
3770 */
3771 int
3772 fp_getfpsem(proc_t p, int fd, struct fileproc **resultfp,
3773 struct psemnode **resultpsem)
3774 {
3775 struct filedesc *fdp = p->p_fd;
3776 struct fileproc *fp;
3777
3778 proc_fdlock_spin(p);
3779 if (fd < 0 || fd >= fdp->fd_nfiles ||
3780 (fp = fdp->fd_ofiles[fd]) == NULL ||
3781 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3782 proc_fdunlock(p);
3783 return (EBADF);
3784 }
3785 if (fp->f_type != DTYPE_PSXSEM) {
3786 proc_fdunlock(p);
3787 return(EBADF);
3788 }
3789 fp->f_iocount++;
3790
3791 if (resultfp)
3792 *resultfp = fp;
3793 if (resultpsem)
3794 *resultpsem = (struct psemnode *)fp->f_data;
3795 proc_fdunlock(p);
3796
3797 return (0);
3798 }
3799
3800
3801 /*
3802 * fp_getfpipe
3803 *
3804 * Description: Get fileproc and pipe pointer for a given fd from the
3805 * per process open file table of the specified process
3806 * and if successful, increment the f_iocount
3807 *
3808 * Parameters: p Process in which fd lives
3809 * fd fd to get information for
3810 * resultfp Pointer to result fileproc
3811 * pointer area, or 0 if none
3812 * resultpipe Pointer to result pipe
3813 * pointer area, or 0 if none
3814 *
3815 * Returns: EBADF The file descriptor is invalid
3816 * EBADF The file descriptor is not a socket
3817 * 0 Success
3818 *
3819 * Implicit returns:
3820 * *resultfp (modified) Fileproc pointer
3821 * *resultpipe (modified) pipe pointer
3822 *
3823 * Notes: The second EBADF should probably be something else to make
3824 * the error condition distinct.
3825 */
3826 int
3827 fp_getfpipe(proc_t p, int fd, struct fileproc **resultfp,
3828 struct pipe **resultpipe)
3829 {
3830 struct filedesc *fdp = p->p_fd;
3831 struct fileproc *fp;
3832
3833 proc_fdlock_spin(p);
3834 if (fd < 0 || fd >= fdp->fd_nfiles ||
3835 (fp = fdp->fd_ofiles[fd]) == NULL ||
3836 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3837 proc_fdunlock(p);
3838 return (EBADF);
3839 }
3840 if (fp->f_type != DTYPE_PIPE) {
3841 proc_fdunlock(p);
3842 return(EBADF);
3843 }
3844 fp->f_iocount++;
3845
3846 if (resultfp)
3847 *resultfp = fp;
3848 if (resultpipe)
3849 *resultpipe = (struct pipe *)fp->f_data;
3850 proc_fdunlock(p);
3851
3852 return (0);
3853 }
3854
3855 /*
3856 * fp_lookup
3857 *
3858 * Description: Get fileproc pointer for a given fd from the per process
3859 * open file table of the specified process and if successful,
3860 * increment the f_iocount
3861 *
3862 * Parameters: p Process in which fd lives
3863 * fd fd to get information for
3864 * resultfp Pointer to result fileproc
3865 * pointer area, or 0 if none
3866 * locked !0 if the caller holds the
3867 * proc_fdlock, 0 otherwise
3868 *
3869 * Returns: 0 Success
3870 * EBADF Bad file descriptor
3871 *
3872 * Implicit returns:
3873 * *resultfp (modified) Fileproc pointer
3874 *
3875 * Locks: If the argument 'locked' is non-zero, then the caller is
3876 * expected to have taken and held the proc_fdlock; if it is
3877 * zero, than this routine internally takes and drops this lock.
3878 */
3879 int
3880 fp_lookup(proc_t p, int fd, struct fileproc **resultfp, int locked)
3881 {
3882 struct filedesc *fdp = p->p_fd;
3883 struct fileproc *fp;
3884
3885 if (!locked)
3886 proc_fdlock_spin(p);
3887 if (fd < 0 || fdp == NULL || fd >= fdp->fd_nfiles ||
3888 (fp = fdp->fd_ofiles[fd]) == NULL ||
3889 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3890 if (!locked)
3891 proc_fdunlock(p);
3892 return (EBADF);
3893 }
3894 fp->f_iocount++;
3895
3896 if (resultfp)
3897 *resultfp = fp;
3898 if (!locked)
3899 proc_fdunlock(p);
3900
3901 return (0);
3902 }
3903
3904
3905 /*
3906 * fp_tryswap
3907 *
3908 * Description: Swap the fileproc pointer for a given fd with a new
3909 * fileproc pointer in the per-process open file table of
3910 * the specified process. The fdlock must be held at entry.
3911 *
3912 * Parameters: p Process containing the fd
3913 * fd The fd of interest
3914 * nfp Pointer to the newfp
3915 *
3916 * Returns: 0 Success
3917 * EBADF Bad file descriptor
3918 * EINTR Interrupted
3919 * EKEEPLOOKING f_iocount changed while lock was dropped.
3920 */
3921 int
3922 fp_tryswap(proc_t p, int fd, struct fileproc *nfp)
3923 {
3924 struct fileproc *fp;
3925 int error;
3926
3927 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
3928
3929 if (0 != (error = fp_lookup(p, fd, &fp, 1)))
3930 return (error);
3931 /*
3932 * At this point, our caller (change_guardedfd_np) has
3933 * one f_iocount reference, and we just took another
3934 * one to begin the replacement.
3935 */
3936 if (fp->f_iocount < 2) {
3937 panic("f_iocount too small %d", fp->f_iocount);
3938 } else if (2 == fp->f_iocount) {
3939
3940 /* Copy the contents of *fp, preserving the "type" of *nfp */
3941
3942 nfp->f_flags = (nfp->f_flags & FP_TYPEMASK) |
3943 (fp->f_flags & ~FP_TYPEMASK);
3944 nfp->f_iocount = fp->f_iocount;
3945 nfp->f_fglob = fp->f_fglob;
3946 nfp->f_wset = fp->f_wset;
3947
3948 p->p_fd->fd_ofiles[fd] = nfp;
3949 (void) fp_drop(p, fd, nfp, 1);
3950 } else {
3951 /*
3952 * Wait for all other active references to evaporate.
3953 */
3954 p->p_fpdrainwait = 1;
3955 error = msleep(&p->p_fpdrainwait, &p->p_fdmlock,
3956 PRIBIO | PCATCH, "tryswap fpdrain", NULL);
3957 if (0 == error) {
3958 /*
3959 * Return an "internal" errno to trigger a full
3960 * reevaluation of the change-guard attempt.
3961 */
3962 error = EKEEPLOOKING;
3963 printf("%s: lookup collision fd %d\n", __func__, fd);
3964 }
3965 (void) fp_drop(p, fd, fp, 1);
3966 }
3967 return (error);
3968 }
3969
3970
3971 /*
3972 * fp_drop_written
3973 *
3974 * Description: Set the FP_WRITTEN flag on the fileproc and drop the I/O
3975 * reference previously taken by calling fp_lookup et. al.
3976 *
3977 * Parameters: p Process in which the fd lives
3978 * fd fd associated with the fileproc
3979 * fp fileproc on which to set the
3980 * flag and drop the reference
3981 *
3982 * Returns: 0 Success
3983 * fp_drop:EBADF Bad file descriptor
3984 *
3985 * Locks: This function internally takes and drops the proc_fdlock for
3986 * the supplied process
3987 *
3988 * Notes: The fileproc must correspond to the fd in the supplied proc
3989 */
3990 int
3991 fp_drop_written(proc_t p, int fd, struct fileproc *fp)
3992 {
3993 int error;
3994
3995 proc_fdlock_spin(p);
3996
3997 fp->f_flags |= FP_WRITTEN;
3998
3999 error = fp_drop(p, fd, fp, 1);
4000
4001 proc_fdunlock(p);
4002
4003 return (error);
4004 }
4005
4006
4007 /*
4008 * fp_drop_event
4009 *
4010 * Description: Set the FP_WAITEVENT flag on the fileproc and drop the I/O
4011 * reference previously taken by calling fp_lookup et. al.
4012 *
4013 * Parameters: p Process in which the fd lives
4014 * fd fd associated with the fileproc
4015 * fp fileproc on which to set the
4016 * flag and drop the reference
4017 *
4018 * Returns: 0 Success
4019 * fp_drop:EBADF Bad file descriptor
4020 *
4021 * Locks: This function internally takes and drops the proc_fdlock for
4022 * the supplied process
4023 *
4024 * Notes: The fileproc must correspond to the fd in the supplied proc
4025 */
4026 int
4027 fp_drop_event(proc_t p, int fd, struct fileproc *fp)
4028 {
4029 int error;
4030
4031 proc_fdlock_spin(p);
4032
4033 fp->f_flags |= FP_WAITEVENT;
4034
4035 error = fp_drop(p, fd, fp, 1);
4036
4037 proc_fdunlock(p);
4038
4039 return (error);
4040 }
4041
4042
4043 /*
4044 * fp_drop
4045 *
4046 * Description: Drop the I/O reference previously taken by calling fp_lookup
4047 * et. al.
4048 *
4049 * Parameters: p Process in which the fd lives
4050 * fd fd associated with the fileproc
4051 * fp fileproc on which to set the
4052 * flag and drop the reference
4053 * locked flag to internally take and
4054 * drop proc_fdlock if it is not
4055 * already held by the caller
4056 *
4057 * Returns: 0 Success
4058 * EBADF Bad file descriptor
4059 *
4060 * Locks: This function internally takes and drops the proc_fdlock for
4061 * the supplied process if 'locked' is non-zero, and assumes that
4062 * the caller already holds this lock if 'locked' is non-zero.
4063 *
4064 * Notes: The fileproc must correspond to the fd in the supplied proc
4065 */
4066 int
4067 fp_drop(proc_t p, int fd, struct fileproc *fp, int locked)
4068 {
4069 struct filedesc *fdp = p->p_fd;
4070 int needwakeup = 0;
4071
4072 if (!locked)
4073 proc_fdlock_spin(p);
4074 if ((fp == FILEPROC_NULL) && (fd < 0 || fd >= fdp->fd_nfiles ||
4075 (fp = fdp->fd_ofiles[fd]) == NULL ||
4076 ((fdp->fd_ofileflags[fd] & UF_RESERVED) &&
4077 !(fdp->fd_ofileflags[fd] & UF_CLOSING)))) {
4078 if (!locked)
4079 proc_fdunlock(p);
4080 return (EBADF);
4081 }
4082 fp->f_iocount--;
4083
4084 if (fp->f_iocount == 0) {
4085 if (fp->f_flags & FP_SELCONFLICT)
4086 fp->f_flags &= ~FP_SELCONFLICT;
4087
4088 if (p->p_fpdrainwait) {
4089 p->p_fpdrainwait = 0;
4090 needwakeup = 1;
4091 }
4092 }
4093 if (!locked)
4094 proc_fdunlock(p);
4095 if (needwakeup)
4096 wakeup(&p->p_fpdrainwait);
4097
4098 return (0);
4099 }
4100
4101
4102 /*
4103 * file_vnode
4104 *
4105 * Description: Given an fd, look it up in the current process's per process
4106 * open file table, and return its internal vnode pointer.
4107 *
4108 * Parameters: fd fd to obtain vnode from
4109 * vpp pointer to vnode return area
4110 *
4111 * Returns: 0 Success
4112 * EINVAL The fd does not refer to a
4113 * vnode fileproc entry
4114 * fp_lookup:EBADF Bad file descriptor
4115 *
4116 * Implicit returns:
4117 * *vpp (modified) Returned vnode pointer
4118 *
4119 * Locks: This function internally takes and drops the proc_fdlock for
4120 * the current process
4121 *
4122 * Notes: If successful, this function increments the f_iocount on the
4123 * fd's corresponding fileproc.
4124 *
4125 * The fileproc referenced is not returned; because of this, care
4126 * must be taken to not drop the last reference (e.g. by closing
4127 * the file). This is inherently unsafe, since the reference may
4128 * not be recoverable from the vnode, if there is a subsequent
4129 * close that destroys the associate fileproc. The caller should
4130 * therefore retain their own reference on the fileproc so that
4131 * the f_iocount can be dropped subsequently. Failure to do this
4132 * can result in the returned pointer immediately becoming invalid
4133 * following the call.
4134 *
4135 * Use of this function is discouraged.
4136 */
4137 int
4138 file_vnode(int fd, struct vnode **vpp)
4139 {
4140 proc_t p = current_proc();
4141 struct fileproc *fp;
4142 int error;
4143
4144 proc_fdlock_spin(p);
4145 if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
4146 proc_fdunlock(p);
4147 return(error);
4148 }
4149 if (fp->f_type != DTYPE_VNODE) {
4150 fp_drop(p, fd, fp,1);
4151 proc_fdunlock(p);
4152 return(EINVAL);
4153 }
4154 if (vpp != NULL)
4155 *vpp = (struct vnode *)fp->f_data;
4156 proc_fdunlock(p);
4157
4158 return(0);
4159 }
4160
4161
4162 /*
4163 * file_vnode_withvid
4164 *
4165 * Description: Given an fd, look it up in the current process's per process
4166 * open file table, and return its internal vnode pointer.
4167 *
4168 * Parameters: fd fd to obtain vnode from
4169 * vpp pointer to vnode return area
4170 * vidp pointer to vid of the returned vnode
4171 *
4172 * Returns: 0 Success
4173 * EINVAL The fd does not refer to a
4174 * vnode fileproc entry
4175 * fp_lookup:EBADF Bad file descriptor
4176 *
4177 * Implicit returns:
4178 * *vpp (modified) Returned vnode pointer
4179 *
4180 * Locks: This function internally takes and drops the proc_fdlock for
4181 * the current process
4182 *
4183 * Notes: If successful, this function increments the f_iocount on the
4184 * fd's corresponding fileproc.
4185 *
4186 * The fileproc referenced is not returned; because of this, care
4187 * must be taken to not drop the last reference (e.g. by closing
4188 * the file). This is inherently unsafe, since the reference may
4189 * not be recoverable from the vnode, if there is a subsequent
4190 * close that destroys the associate fileproc. The caller should
4191 * therefore retain their own reference on the fileproc so that
4192 * the f_iocount can be dropped subsequently. Failure to do this
4193 * can result in the returned pointer immediately becoming invalid
4194 * following the call.
4195 *
4196 * Use of this function is discouraged.
4197 */
4198 int
4199 file_vnode_withvid(int fd, struct vnode **vpp, uint32_t * vidp)
4200 {
4201 proc_t p = current_proc();
4202 struct fileproc *fp;
4203 vnode_t vp;
4204 int error;
4205
4206 proc_fdlock_spin(p);
4207 if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
4208 proc_fdunlock(p);
4209 return(error);
4210 }
4211 if (fp->f_type != DTYPE_VNODE) {
4212 fp_drop(p, fd, fp,1);
4213 proc_fdunlock(p);
4214 return(EINVAL);
4215 }
4216 vp = (struct vnode *)fp->f_data;
4217 if (vpp != NULL)
4218 *vpp = vp;
4219
4220 if ((vidp != NULL) && (vp != NULLVP))
4221 *vidp = (uint32_t)vp->v_id;
4222
4223 proc_fdunlock(p);
4224
4225 return(0);
4226 }
4227
4228
4229 /*
4230 * file_socket
4231 *
4232 * Description: Given an fd, look it up in the current process's per process
4233 * open file table, and return its internal socket pointer.
4234 *
4235 * Parameters: fd fd to obtain vnode from
4236 * sp pointer to socket return area
4237 *
4238 * Returns: 0 Success
4239 * ENOTSOCK Not a socket
4240 * fp_lookup:EBADF Bad file descriptor
4241 *
4242 * Implicit returns:
4243 * *sp (modified) Returned socket pointer
4244 *
4245 * Locks: This function internally takes and drops the proc_fdlock for
4246 * the current process
4247 *
4248 * Notes: If successful, this function increments the f_iocount on the
4249 * fd's corresponding fileproc.
4250 *
4251 * The fileproc referenced is not returned; because of this, care
4252 * must be taken to not drop the last reference (e.g. by closing
4253 * the file). This is inherently unsafe, since the reference may
4254 * not be recoverable from the socket, if there is a subsequent
4255 * close that destroys the associate fileproc. The caller should
4256 * therefore retain their own reference on the fileproc so that
4257 * the f_iocount can be dropped subsequently. Failure to do this
4258 * can result in the returned pointer immediately becoming invalid
4259 * following the call.
4260 *
4261 * Use of this function is discouraged.
4262 */
4263 int
4264 file_socket(int fd, struct socket **sp)
4265 {
4266 proc_t p = current_proc();
4267 struct fileproc *fp;
4268 int error;
4269
4270 proc_fdlock_spin(p);
4271 if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
4272 proc_fdunlock(p);
4273 return(error);
4274 }
4275 if (fp->f_type != DTYPE_SOCKET) {
4276 fp_drop(p, fd, fp,1);
4277 proc_fdunlock(p);
4278 return(ENOTSOCK);
4279 }
4280 *sp = (struct socket *)fp->f_data;
4281 proc_fdunlock(p);
4282
4283 return(0);
4284 }
4285
4286
4287 /*
4288 * file_flags
4289 *
4290 * Description: Given an fd, look it up in the current process's per process
4291 * open file table, and return its fileproc's flags field.
4292 *
4293 * Parameters: fd fd whose flags are to be
4294 * retrieved
4295 * flags pointer to flags data area
4296 *
4297 * Returns: 0 Success
4298 * ENOTSOCK Not a socket
4299 * fp_lookup:EBADF Bad file descriptor
4300 *
4301 * Implicit returns:
4302 * *flags (modified) Returned flags field
4303 *
4304 * Locks: This function internally takes and drops the proc_fdlock for
4305 * the current process
4306 *
4307 * Notes: This function will internally increment and decrement the
4308 * f_iocount of the fileproc as part of its operation.
4309 */
4310 int
4311 file_flags(int fd, int *flags)
4312 {
4313
4314 proc_t p = current_proc();
4315 struct fileproc *fp;
4316 int error;
4317
4318 proc_fdlock_spin(p);
4319 if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
4320 proc_fdunlock(p);
4321 return(error);
4322 }
4323 *flags = (int)fp->f_flag;
4324 fp_drop(p, fd, fp,1);
4325 proc_fdunlock(p);
4326
4327 return(0);
4328 }
4329
4330
4331 /*
4332 * file_drop
4333 *
4334 * Description: Drop an iocount reference on an fd, and wake up any waiters
4335 * for draining (i.e. blocked in fileproc_drain() called during
4336 * the last attempt to close a file).
4337 *
4338 * Parameters: fd fd on which an ioreference is
4339 * to be dropped
4340 *
4341 * Returns: 0 Success
4342 * EBADF Bad file descriptor
4343 *
4344 * Description: Given an fd, look it up in the current process's per process
4345 * open file table, and drop it's fileproc's f_iocount by one
4346 *
4347 * Notes: This is intended as a corresponding operation to the functions
4348 * file_vnode() and file_socket() operations.
4349 *
4350 * Technically, the close reference is supposed to be protected
4351 * by a fileproc_drain(), however, a drain will only block if
4352 * the fd refers to a character device, and that device has had
4353 * preparefileread() called on it. If it refers to something
4354 * other than a character device, then the drain will occur and
4355 * block each close attempt, rather than merely the last close.
4356 *
4357 * Since it's possible for an fd that refers to a character
4358 * device to have an intermediate close followed by an open to
4359 * cause a different file to correspond to that descriptor,
4360 * unless there was a cautionary reference taken on the fileproc,
4361 * this is an inherently unsafe function. This happens in the
4362 * case where multiple fd's in a process refer to the same
4363 * character device (e.g. stdin/out/err pointing to a tty, etc.).
4364 *
4365 * Use of this function is discouraged.
4366 */
4367 int
4368 file_drop(int fd)
4369 {
4370 struct fileproc *fp;
4371 proc_t p = current_proc();
4372 int needwakeup = 0;
4373
4374 proc_fdlock_spin(p);
4375 if (fd < 0 || fd >= p->p_fd->fd_nfiles ||
4376 (fp = p->p_fd->fd_ofiles[fd]) == NULL ||
4377 ((p->p_fd->fd_ofileflags[fd] & UF_RESERVED) &&
4378 !(p->p_fd->fd_ofileflags[fd] & UF_CLOSING))) {
4379 proc_fdunlock(p);
4380 return (EBADF);
4381 }
4382 fp->f_iocount --;
4383
4384 if (fp->f_iocount == 0) {
4385 if (fp->f_flags & FP_SELCONFLICT)
4386 fp->f_flags &= ~FP_SELCONFLICT;
4387
4388 if (p->p_fpdrainwait) {
4389 p->p_fpdrainwait = 0;
4390 needwakeup = 1;
4391 }
4392 }
4393 proc_fdunlock(p);
4394
4395 if (needwakeup)
4396 wakeup(&p->p_fpdrainwait);
4397 return(0);
4398 }
4399
4400
4401 static int falloc_withalloc_locked(proc_t, struct fileproc **, int *,
4402 vfs_context_t, struct fileproc * (*)(void *), void *, int);
4403
4404 /*
4405 * falloc
4406 *
4407 * Description: Allocate an entry in the per process open file table and
4408 * return the corresponding fileproc and fd.
4409 *
4410 * Parameters: p The process in whose open file
4411 * table the fd is to be allocated
4412 * resultfp Pointer to fileproc pointer
4413 * return area
4414 * resultfd Pointer to fd return area
4415 * ctx VFS context
4416 *
4417 * Returns: 0 Success
4418 * falloc:ENFILE Too many open files in system
4419 * falloc:EMFILE Too many open files in process
4420 * falloc:ENOMEM M_FILEPROC or M_FILEGLOB zone
4421 * exhausted
4422 *
4423 * Implicit returns:
4424 * *resultfd (modified) Returned fileproc pointer
4425 * *resultfd (modified) Returned fd
4426 *
4427 * Locks: This function takes and drops the proc_fdlock; if this lock
4428 * is already held, use falloc_locked() instead.
4429 *
4430 * Notes: This function takes separate process and context arguments
4431 * solely to support kern_exec.c; otherwise, it would take
4432 * neither, and expect falloc_locked() to use the
4433 * vfs_context_current() routine internally.
4434 */
4435 int
4436 falloc(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx)
4437 {
4438 return (falloc_withalloc(p, resultfp, resultfd, ctx,
4439 fileproc_alloc_init, NULL));
4440 }
4441
4442 /*
4443 * Like falloc, but including the fileproc allocator and create-args
4444 */
4445 int
4446 falloc_withalloc(proc_t p, struct fileproc **resultfp, int *resultfd,
4447 vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *arg)
4448 {
4449 int error;
4450
4451 proc_fdlock(p);
4452 error = falloc_withalloc_locked(p,
4453 resultfp, resultfd, ctx, fp_zalloc, arg, 1);
4454 proc_fdunlock(p);
4455
4456 return (error);
4457 }
4458
4459 /*
4460 * "uninitialized" ops -- ensure fg->fg_ops->fo_type always exists
4461 */
4462 static const struct fileops uninitops;
4463
4464 /*
4465 * falloc_locked
4466 *
4467 * Create a new open file structure and allocate
4468 * a file descriptor for the process that refers to it.
4469 *
4470 * Returns: 0 Success
4471 *
4472 * Description: Allocate an entry in the per process open file table and
4473 * return the corresponding fileproc and fd.
4474 *
4475 * Parameters: p The process in whose open file
4476 * table the fd is to be allocated
4477 * resultfp Pointer to fileproc pointer
4478 * return area
4479 * resultfd Pointer to fd return area
4480 * ctx VFS context
4481 * locked Flag to indicate whether the
4482 * caller holds proc_fdlock
4483 *
4484 * Returns: 0 Success
4485 * ENFILE Too many open files in system
4486 * fdalloc:EMFILE Too many open files in process
4487 * ENOMEM M_FILEPROC or M_FILEGLOB zone
4488 * exhausted
4489 * fdalloc:ENOMEM
4490 *
4491 * Implicit returns:
4492 * *resultfd (modified) Returned fileproc pointer
4493 * *resultfd (modified) Returned fd
4494 *
4495 * Locks: If the parameter 'locked' is zero, this function takes and
4496 * drops the proc_fdlock; if non-zero, the caller must hold the
4497 * lock.
4498 *
4499 * Notes: If you intend to use a non-zero 'locked' parameter, use the
4500 * utility function falloc() instead.
4501 *
4502 * This function takes separate process and context arguments
4503 * solely to support kern_exec.c; otherwise, it would take
4504 * neither, and use the vfs_context_current() routine internally.
4505 */
4506 int
4507 falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd,
4508 vfs_context_t ctx, int locked)
4509 {
4510 return (falloc_withalloc_locked(p, resultfp, resultfd, ctx,
4511 fileproc_alloc_init, NULL, locked));
4512 }
4513
4514 static int
4515 falloc_withalloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd,
4516 vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *crarg,
4517 int locked)
4518 {
4519 struct fileproc *fp;
4520 struct fileglob *fg;
4521 int error, nfd;
4522
4523 if (!locked)
4524 proc_fdlock(p);
4525 if ( (error = fdalloc(p, 0, &nfd)) ) {
4526 if (!locked)
4527 proc_fdunlock(p);
4528 return (error);
4529 }
4530 if (nfiles >= maxfiles) {
4531 if (!locked)
4532 proc_fdunlock(p);
4533 tablefull("file");
4534 return (ENFILE);
4535 }
4536 #if CONFIG_MACF
4537 error = mac_file_check_create(proc_ucred(p));
4538 if (error) {
4539 if (!locked)
4540 proc_fdunlock(p);
4541 return (error);
4542 }
4543 #endif
4544
4545 /*
4546 * Allocate a new file descriptor.
4547 * If the process has file descriptor zero open, add to the list
4548 * of open files at that point, otherwise put it at the front of
4549 * the list of open files.
4550 */
4551 proc_fdunlock(p);
4552
4553 fp = (*fp_zalloc)(crarg);
4554 if (fp == NULL) {
4555 if (locked)
4556 proc_fdlock(p);
4557 return (ENOMEM);
4558 }
4559 MALLOC_ZONE(fg, struct fileglob *, sizeof(struct fileglob), M_FILEGLOB, M_WAITOK);
4560 if (fg == NULL) {
4561 fileproc_free(fp);
4562 if (locked)
4563 proc_fdlock(p);
4564 return (ENOMEM);
4565 }
4566 bzero(fg, sizeof(struct fileglob));
4567 lck_mtx_init(&fg->fg_lock, file_lck_grp, file_lck_attr);
4568
4569 fp->f_iocount = 1;
4570 fg->fg_count = 1;
4571 fg->fg_ops = &uninitops;
4572 fp->f_fglob = fg;
4573 #if CONFIG_MACF
4574 mac_file_label_init(fg);
4575 #endif
4576
4577 kauth_cred_ref(ctx->vc_ucred);
4578
4579 proc_fdlock(p);
4580
4581 fp->f_cred = ctx->vc_ucred;
4582
4583 #if CONFIG_MACF
4584 mac_file_label_associate(fp->f_cred, fg);
4585 #endif
4586
4587 OSAddAtomic(1, &nfiles);
4588
4589 p->p_fd->fd_ofiles[nfd] = fp;
4590
4591 if (!locked)
4592 proc_fdunlock(p);
4593
4594 if (resultfp)
4595 *resultfp = fp;
4596 if (resultfd)
4597 *resultfd = nfd;
4598
4599 return (0);
4600 }
4601
4602
4603 /*
4604 * fg_free
4605 *
4606 * Description: Free a file structure; drop the global open file count, and
4607 * drop the credential reference, if the fileglob has one, and
4608 * destroy the instance mutex before freeing
4609 *
4610 * Parameters: fg Pointer to fileglob to be
4611 * freed
4612 *
4613 * Returns: void
4614 */
4615 void
4616 fg_free(struct fileglob *fg)
4617 {
4618 OSAddAtomic(-1, &nfiles);
4619
4620 if (fg->fg_vn_data) {
4621 fg_vn_data_free(fg->fg_vn_data);
4622 fg->fg_vn_data = NULL;
4623 }
4624
4625 if (IS_VALID_CRED(fg->fg_cred)) {
4626 kauth_cred_unref(&fg->fg_cred);
4627 }
4628 lck_mtx_destroy(&fg->fg_lock, file_lck_grp);
4629
4630 #if CONFIG_MACF
4631 mac_file_label_destroy(fg);
4632 #endif
4633 FREE_ZONE(fg, sizeof *fg, M_FILEGLOB);
4634 }
4635
4636
4637 /*
4638 * fdexec
4639 *
4640 * Description: Perform close-on-exec processing for all files in a process
4641 * that are either marked as close-on-exec, or which were in the
4642 * process of being opened at the time of the execve
4643 *
4644 * Also handles the case (via posix_spawn()) where -all-
4645 * files except those marked with "inherit" as treated as
4646 * close-on-exec.
4647 *
4648 * Parameters: p Pointer to process calling
4649 * execve
4650 *
4651 * Returns: void
4652 *
4653 * Locks: This function internally takes and drops proc_fdlock()
4654 *
4655 */
4656 void
4657 fdexec(proc_t p, short flags)
4658 {
4659 struct filedesc *fdp = p->p_fd;
4660 int i;
4661 boolean_t cloexec_default = (flags & POSIX_SPAWN_CLOEXEC_DEFAULT) != 0;
4662
4663 proc_fdlock(p);
4664 for (i = fdp->fd_lastfile; i >= 0; i--) {
4665
4666 struct fileproc *fp = fdp->fd_ofiles[i];
4667 char *flagp = &fdp->fd_ofileflags[i];
4668
4669 if (fp && cloexec_default) {
4670 /*
4671 * Reverse the usual semantics of file descriptor
4672 * inheritance - all of them should be closed
4673 * except files marked explicitly as "inherit" and
4674 * not marked close-on-exec.
4675 */
4676 if ((*flagp & (UF_EXCLOSE|UF_INHERIT)) != UF_INHERIT)
4677 *flagp |= UF_EXCLOSE;
4678 *flagp &= ~UF_INHERIT;
4679 }
4680
4681 if (
4682 ((*flagp & (UF_RESERVED|UF_EXCLOSE)) == UF_EXCLOSE)
4683 #if CONFIG_MACF
4684 || (fp && mac_file_check_inherit(proc_ucred(p), fp->f_fglob))
4685 #endif
4686 ) {
4687 if (i < fdp->fd_knlistsize)
4688 knote_fdclose(p, i);
4689 procfdtbl_clearfd(p, i);
4690 if (i == fdp->fd_lastfile && i > 0)
4691 fdp->fd_lastfile--;
4692 if (i < fdp->fd_freefile)
4693 fdp->fd_freefile = i;
4694
4695 /*
4696 * Wait for any third party viewers (e.g., lsof)
4697 * to release their references to this fileproc.
4698 */
4699 while (fp->f_iocount > 0) {
4700 p->p_fpdrainwait = 1;
4701 msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO,
4702 "fpdrain", NULL);
4703 }
4704
4705 closef_locked(fp, fp->f_fglob, p);
4706
4707 fileproc_free(fp);
4708 }
4709 }
4710 proc_fdunlock(p);
4711 }
4712
4713
4714 /*
4715 * fdcopy
4716 *
4717 * Description: Copy a filedesc structure. This is normally used as part of
4718 * forkproc() when forking a new process, to copy the per process
4719 * open file table over to the new process.
4720 *
4721 * Parameters: p Process whose open file table
4722 * is to be copied (parent)
4723 * uth_cdir Per thread current working
4724 * cirectory, or NULL
4725 *
4726 * Returns: NULL Copy failed
4727 * !NULL Pointer to new struct filedesc
4728 *
4729 * Locks: This function internally takes and drops proc_fdlock()
4730 *
4731 * Notes: Files are copied directly, ignoring the new resource limits
4732 * for the process that's being copied into. Since the descriptor
4733 * references are just additional references, this does not count
4734 * against the number of open files on the system.
4735 *
4736 * The struct filedesc includes the current working directory,
4737 * and the current root directory, if the process is chroot'ed.
4738 *
4739 * If the exec was called by a thread using a per thread current
4740 * working directory, we inherit the working directory from the
4741 * thread making the call, rather than from the process.
4742 *
4743 * In the case of a failure to obtain a reference, for most cases,
4744 * the file entry will be silently dropped. There's an exception
4745 * for the case of a chroot dir, since a failure to to obtain a
4746 * reference there would constitute an "escape" from the chroot
4747 * environment, which must not be allowed. In that case, we will
4748 * deny the execve() operation, rather than allowing the escape.
4749 */
4750 struct filedesc *
4751 fdcopy(proc_t p, vnode_t uth_cdir)
4752 {
4753 struct filedesc *newfdp, *fdp = p->p_fd;
4754 int i;
4755 struct fileproc *ofp, *fp;
4756 vnode_t v_dir;
4757
4758 MALLOC_ZONE(newfdp, struct filedesc *,
4759 sizeof(*newfdp), M_FILEDESC, M_WAITOK);
4760 if (newfdp == NULL)
4761 return(NULL);
4762
4763 proc_fdlock(p);
4764
4765 /*
4766 * the FD_CHROOT flag will be inherited via this copy
4767 */
4768 (void) memcpy(newfdp, fdp, sizeof(*newfdp));
4769
4770 /*
4771 * If we are running with per-thread current working directories,
4772 * inherit the new current working directory from the current thread
4773 * instead, before we take our references.
4774 */
4775 if (uth_cdir != NULLVP)
4776 newfdp->fd_cdir = uth_cdir;
4777
4778 /*
4779 * For both fd_cdir and fd_rdir make sure we get
4780 * a valid reference... if we can't, than set
4781 * set the pointer(s) to NULL in the child... this
4782 * will keep us from using a non-referenced vp
4783 * and allows us to do the vnode_rele only on
4784 * a properly referenced vp
4785 */
4786 if ( (v_dir = newfdp->fd_cdir) ) {
4787 if (vnode_getwithref(v_dir) == 0) {
4788 if ( (vnode_ref(v_dir)) )
4789 newfdp->fd_cdir = NULL;
4790 vnode_put(v_dir);
4791 } else
4792 newfdp->fd_cdir = NULL;
4793 }
4794 if (newfdp->fd_cdir == NULL && fdp->fd_cdir) {
4795 /*
4796 * we couldn't get a new reference on
4797 * the current working directory being
4798 * inherited... we might as well drop
4799 * our reference from the parent also
4800 * since the vnode has gone DEAD making
4801 * it useless... by dropping it we'll
4802 * be that much closer to recycling it
4803 */
4804 vnode_rele(fdp->fd_cdir);
4805 fdp->fd_cdir = NULL;
4806 }
4807
4808 if ( (v_dir = newfdp->fd_rdir) ) {
4809 if (vnode_getwithref(v_dir) == 0) {
4810 if ( (vnode_ref(v_dir)) )
4811 newfdp->fd_rdir = NULL;
4812 vnode_put(v_dir);
4813 } else {
4814 newfdp->fd_rdir = NULL;
4815 }
4816 }
4817 /* Coming from a chroot environment and unable to get a reference... */
4818 if (newfdp->fd_rdir == NULL && fdp->fd_rdir) {
4819 /*
4820 * We couldn't get a new reference on
4821 * the chroot directory being
4822 * inherited... this is fatal, since
4823 * otherwise it would constitute an
4824 * escape from a chroot environment by
4825 * the new process.
4826 */
4827 if (newfdp->fd_cdir)
4828 vnode_rele(newfdp->fd_cdir);
4829 FREE_ZONE(newfdp, sizeof *newfdp, M_FILEDESC);
4830 return(NULL);
4831 }
4832
4833 /*
4834 * If the number of open files fits in the internal arrays
4835 * of the open file structure, use them, otherwise allocate
4836 * additional memory for the number of descriptors currently
4837 * in use.
4838 */
4839 if (newfdp->fd_lastfile < NDFILE)
4840 i = NDFILE;
4841 else {
4842 /*
4843 * Compute the smallest multiple of NDEXTENT needed
4844 * for the file descriptors currently in use,
4845 * allowing the table to shrink.
4846 */
4847 i = newfdp->fd_nfiles;
4848 while (i > 1 + 2 * NDEXTENT && i > 1 + newfdp->fd_lastfile * 2)
4849 i /= 2;
4850 }
4851 proc_fdunlock(p);
4852
4853 MALLOC_ZONE(newfdp->fd_ofiles, struct fileproc **,
4854 i * OFILESIZE, M_OFILETABL, M_WAITOK);
4855 if (newfdp->fd_ofiles == NULL) {
4856 if (newfdp->fd_cdir)
4857 vnode_rele(newfdp->fd_cdir);
4858 if (newfdp->fd_rdir)
4859 vnode_rele(newfdp->fd_rdir);
4860
4861 FREE_ZONE(newfdp, sizeof(*newfdp), M_FILEDESC);
4862 return(NULL);
4863 }
4864 (void) memset(newfdp->fd_ofiles, 0, i * OFILESIZE);
4865 proc_fdlock(p);
4866
4867 newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
4868 newfdp->fd_nfiles = i;
4869
4870 if (fdp->fd_nfiles > 0) {
4871 struct fileproc **fpp;
4872 char *flags;
4873
4874 (void) memcpy(newfdp->fd_ofiles, fdp->fd_ofiles,
4875 (newfdp->fd_lastfile + 1) * sizeof(*fdp->fd_ofiles));
4876 (void) memcpy(newfdp->fd_ofileflags, fdp->fd_ofileflags,
4877 (newfdp->fd_lastfile + 1) * sizeof(*fdp->fd_ofileflags));
4878
4879 /*
4880 * kq descriptors cannot be copied.
4881 */
4882 if (newfdp->fd_knlistsize != -1) {
4883 fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile];
4884 flags = &newfdp->fd_ofileflags[newfdp->fd_lastfile];
4885 for (i = newfdp->fd_lastfile;
4886 i >= 0; i--, fpp--, flags--) {
4887 if (*flags & UF_RESERVED)
4888 continue; /* (removed below) */
4889 if (*fpp != NULL && (*fpp)->f_type == DTYPE_KQUEUE) {
4890 *fpp = NULL;
4891 *flags = 0;
4892 if (i < newfdp->fd_freefile)
4893 newfdp->fd_freefile = i;
4894 }
4895 if (*fpp == NULL && i == newfdp->fd_lastfile && i > 0)
4896 newfdp->fd_lastfile--;
4897 }
4898 newfdp->fd_knlist = NULL;
4899 newfdp->fd_knlistsize = -1;
4900 newfdp->fd_knhash = NULL;
4901 newfdp->fd_knhashmask = 0;
4902 }
4903 fpp = newfdp->fd_ofiles;
4904 flags = newfdp->fd_ofileflags;
4905
4906 for (i = newfdp->fd_lastfile + 1; --i >= 0; fpp++, flags++)
4907 if ((ofp = *fpp) != NULL &&
4908 0 == (ofp->f_fglob->fg_lflags & FG_CONFINED) &&
4909 0 == (*flags & (UF_FORKCLOSE|UF_RESERVED))) {
4910 #if DEBUG
4911 if (FILEPROC_TYPE(ofp) != FTYPE_SIMPLE)
4912 panic("complex fileproc");
4913 #endif
4914 fp = fileproc_alloc_init(NULL);
4915 if (fp == NULL) {
4916 /*
4917 * XXX no room to copy, unable to
4918 * XXX safely unwind state at present
4919 */
4920 *fpp = NULL;
4921 } else {
4922 fp->f_flags |=
4923 (ofp->f_flags & ~FP_TYPEMASK);
4924 fp->f_fglob = ofp->f_fglob;
4925 (void)fg_ref(fp);
4926 *fpp = fp;
4927 }
4928 } else {
4929 if (i < newfdp->fd_freefile)
4930 newfdp->fd_freefile = i;
4931 *fpp = NULL;
4932 *flags = 0;
4933 }
4934 }
4935
4936 proc_fdunlock(p);
4937 return (newfdp);
4938 }
4939
4940
4941 /*
4942 * fdfree
4943 *
4944 * Description: Release a filedesc (per process open file table) structure;
4945 * this is done on process exit(), or from forkproc_free() if
4946 * the fork fails for some reason subsequent to a successful
4947 * call to fdcopy()
4948 *
4949 * Parameters: p Pointer to process going away
4950 *
4951 * Returns: void
4952 *
4953 * Locks: This function internally takes and drops proc_fdlock()
4954 */
4955 void
4956 fdfree(proc_t p)
4957 {
4958 struct filedesc *fdp;
4959 struct fileproc *fp;
4960 int i;
4961
4962 proc_fdlock(p);
4963
4964 if (p == kernproc || NULL == (fdp = p->p_fd)) {
4965 proc_fdunlock(p);
4966 return;
4967 }
4968
4969 extern struct filedesc filedesc0;
4970
4971 if (&filedesc0 == fdp)
4972 panic("filedesc0");
4973
4974 if (fdp->fd_nfiles > 0 && fdp->fd_ofiles) {
4975 for (i = fdp->fd_lastfile; i >= 0; i--) {
4976 if ((fp = fdp->fd_ofiles[i]) != NULL) {
4977
4978 if (fdp->fd_ofileflags[i] & UF_RESERVED)
4979 panic("fdfree: found fp with UF_RESERVED");
4980
4981 procfdtbl_reservefd(p, i);
4982
4983 if (i < fdp->fd_knlistsize)
4984 knote_fdclose(p, i);
4985 if (fp->f_flags & FP_WAITEVENT)
4986 (void)waitevent_close(p, fp);
4987 (void) closef_locked(fp, fp->f_fglob, p);
4988 fileproc_free(fp);
4989 }
4990 }
4991 FREE_ZONE(fdp->fd_ofiles, fdp->fd_nfiles * OFILESIZE, M_OFILETABL);
4992 fdp->fd_ofiles = NULL;
4993 fdp->fd_nfiles = 0;
4994 }
4995
4996 proc_fdunlock(p);
4997
4998 if (fdp->fd_cdir)
4999 vnode_rele(fdp->fd_cdir);
5000 if (fdp->fd_rdir)
5001 vnode_rele(fdp->fd_rdir);
5002
5003 proc_fdlock_spin(p);
5004 p->p_fd = NULL;
5005 proc_fdunlock(p);
5006
5007 if (fdp->fd_knlist)
5008 FREE(fdp->fd_knlist, M_KQUEUE);
5009 if (fdp->fd_knhash)
5010 FREE(fdp->fd_knhash, M_KQUEUE);
5011
5012 FREE_ZONE(fdp, sizeof(*fdp), M_FILEDESC);
5013 }
5014
5015 /*
5016 * closef_locked
5017 *
5018 * Description: Internal form of closef; called with proc_fdlock held
5019 *
5020 * Parameters: fp Pointer to fileproc for fd
5021 * fg Pointer to fileglob for fd
5022 * p Pointer to proc structure
5023 *
5024 * Returns: 0 Success
5025 * closef_finish:??? Anything returnable by a per-fileops
5026 * close function
5027 *
5028 * Note: Decrements reference count on file structure; if this was the
5029 * last reference, then closef_finish() is called
5030 *
5031 * p and fp are allowed to be NULL when closing a file that was
5032 * being passed in a message (but only if we are called when this
5033 * is NOT the last reference).
5034 */
5035 int
5036 closef_locked(struct fileproc *fp, struct fileglob *fg, proc_t p)
5037 {
5038 struct vnode *vp;
5039 struct flock lf;
5040 struct vfs_context context;
5041 int error;
5042
5043 if (fg == NULL) {
5044 return (0);
5045 }
5046
5047 /* Set up context with cred stashed in fg */
5048 if (p == current_proc())
5049 context.vc_thread = current_thread();
5050 else
5051 context.vc_thread = NULL;
5052 context.vc_ucred = fg->fg_cred;
5053
5054 /*
5055 * POSIX record locking dictates that any close releases ALL
5056 * locks owned by this process. This is handled by setting
5057 * a flag in the unlock to free ONLY locks obeying POSIX
5058 * semantics, and not to free BSD-style file locks.
5059 * If the descriptor was in a message, POSIX-style locks
5060 * aren't passed with the descriptor.
5061 */
5062 if (p && (p->p_ladvflag & P_LADVLOCK) &&
5063 DTYPE_VNODE == FILEGLOB_DTYPE(fg)) {
5064 proc_fdunlock(p);
5065
5066 lf.l_whence = SEEK_SET;
5067 lf.l_start = 0;
5068 lf.l_len = 0;
5069 lf.l_type = F_UNLCK;
5070 vp = (struct vnode *)fg->fg_data;
5071
5072 if ( (error = vnode_getwithref(vp)) == 0 ) {
5073 (void) VNOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX, &context, NULL);
5074 (void)vnode_put(vp);
5075 }
5076 proc_fdlock(p);
5077 }
5078 lck_mtx_lock_spin(&fg->fg_lock);
5079 fg->fg_count--;
5080
5081 if (fg->fg_count > 0) {
5082 lck_mtx_unlock(&fg->fg_lock);
5083 return (0);
5084 }
5085 #if DIAGNOSTIC
5086 if (fg->fg_count != 0)
5087 panic("fg %p: being freed with bad fg_count (%d)", fg, fg->fg_count);
5088 #endif
5089
5090 if (fp && (fp->f_flags & FP_WRITTEN))
5091 fg->fg_flag |= FWASWRITTEN;
5092
5093 fg->fg_lflags |= FG_TERM;
5094 lck_mtx_unlock(&fg->fg_lock);
5095
5096 if (p)
5097 proc_fdunlock(p);
5098
5099 /* Since we ensure that fg->fg_ops is always initialized,
5100 * it is safe to invoke fo_close on the fg */
5101 error = fo_close(fg, &context);
5102
5103 fg_free(fg);
5104
5105 if (p)
5106 proc_fdlock(p);
5107
5108 return(error);
5109 }
5110
5111
5112 /*
5113 * fileproc_drain
5114 *
5115 * Description: Drain out pending I/O operations
5116 *
5117 * Parameters: p Process closing this file
5118 * fp fileproc struct for the open
5119 * instance on the file
5120 *
5121 * Returns: void
5122 *
5123 * Locks: Assumes the caller holds the proc_fdlock
5124 *
5125 * Notes: For character devices, this occurs on the last close of the
5126 * device; for all other file descriptors, this occurs on each
5127 * close to prevent fd's from being closed out from under
5128 * operations currently in progress and blocked
5129 *
5130 * See Also: file_vnode(), file_socket(), file_drop(), and the cautions
5131 * regarding their use and interaction with this function.
5132 */
5133 void
5134 fileproc_drain(proc_t p, struct fileproc * fp)
5135 {
5136 struct vfs_context context;
5137
5138 context.vc_thread = proc_thread(p); /* XXX */
5139 context.vc_ucred = fp->f_fglob->fg_cred;
5140
5141 fp->f_iocount-- ; /* (the one the close holds) */
5142
5143 while (fp->f_iocount) {
5144
5145 lck_mtx_convert_spin(&p->p_fdmlock);
5146
5147 if (fp->f_fglob->fg_ops->fo_drain) {
5148 (*fp->f_fglob->fg_ops->fo_drain)(fp, &context);
5149 }
5150 if ((fp->f_flags & FP_INSELECT) == FP_INSELECT) {
5151 if (waitq_wakeup64_all((struct waitq *)fp->f_wset, NO_EVENT64,
5152 THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT)
5153 panic("bad wait queue for waitq_wakeup64_all %p (fp:%p)", fp->f_wset, fp);
5154 }
5155 if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
5156 if (waitq_wakeup64_all(&select_conflict_queue, NO_EVENT64,
5157 THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT)
5158 panic("bad select_conflict_queue");
5159 }
5160 p->p_fpdrainwait = 1;
5161
5162 msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO, "fpdrain", NULL);
5163
5164 }
5165 #if DIAGNOSTIC
5166 if ((fp->f_flags & FP_INSELECT) != 0)
5167 panic("FP_INSELECT set on drained fp");
5168 #endif
5169 if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT)
5170 fp->f_flags &= ~FP_SELCONFLICT;
5171 }
5172
5173
5174 /*
5175 * fp_free
5176 *
5177 * Description: Release the fd and free the fileproc associated with the fd
5178 * in the per process open file table of the specified process;
5179 * these values must correspond.
5180 *
5181 * Parameters: p Process containing fd
5182 * fd fd to be released
5183 * fp fileproc to be freed
5184 *
5185 * Returns: 0 Success
5186 *
5187 * Notes: XXX function should be void - no one interprets the returns
5188 * XXX code
5189 */
5190 int
5191 fp_free(proc_t p, int fd, struct fileproc * fp)
5192 {
5193 proc_fdlock_spin(p);
5194 fdrelse(p, fd);
5195 proc_fdunlock(p);
5196
5197 fg_free(fp->f_fglob);
5198 fileproc_free(fp);
5199 return(0);
5200 }
5201
5202
5203 /*
5204 * flock
5205 *
5206 * Description: Apply an advisory lock on a file descriptor.
5207 *
5208 * Parameters: p Process making request
5209 * uap->fd fd on which the lock is to be
5210 * attempted
5211 * uap->how (Un)Lock bits, including type
5212 * retval Pointer to the call return area
5213 *
5214 * Returns: 0 Success
5215 * fp_getfvp:EBADF Bad file descriptor
5216 * fp_getfvp:ENOTSUP fd does not refer to a vnode
5217 * vnode_getwithref:???
5218 * VNOP_ADVLOCK:???
5219 *
5220 * Implicit returns:
5221 * *retval (modified) Size of dtable
5222 *
5223 * Notes: Just attempt to get a record lock of the requested type on
5224 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
5225 */
5226 int
5227 flock(proc_t p, struct flock_args *uap, __unused int32_t *retval)
5228 {
5229 int fd = uap->fd;
5230 int how = uap->how;
5231 struct fileproc *fp;
5232 struct vnode *vp;
5233 struct flock lf;
5234 vfs_context_t ctx = vfs_context_current();
5235 int error=0;
5236
5237 AUDIT_ARG(fd, uap->fd);
5238 if ( (error = fp_getfvp(p, fd, &fp, &vp)) ) {
5239 return(error);
5240 }
5241 if ( (error = vnode_getwithref(vp)) ) {
5242 goto out1;
5243 }
5244 AUDIT_ARG(vnpath, vp, ARG_VNODE1);
5245
5246 lf.l_whence = SEEK_SET;
5247 lf.l_start = 0;
5248 lf.l_len = 0;
5249 if (how & LOCK_UN) {
5250 lf.l_type = F_UNLCK;
5251 fp->f_flag &= ~FHASLOCK;
5252 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_UNLCK, &lf, F_FLOCK, ctx, NULL);
5253 goto out;
5254 }
5255 if (how & LOCK_EX)
5256 lf.l_type = F_WRLCK;
5257 else if (how & LOCK_SH)
5258 lf.l_type = F_RDLCK;
5259 else {
5260 error = EBADF;
5261 goto out;
5262 }
5263 #if CONFIG_MACF
5264 error = mac_file_check_lock(proc_ucred(p), fp->f_fglob, F_SETLK, &lf);
5265 if (error)
5266 goto out;
5267 #endif
5268 fp->f_flag |= FHASLOCK;
5269 if (how & LOCK_NB) {
5270 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, F_FLOCK, ctx, NULL);
5271 goto out;
5272 }
5273 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, F_FLOCK|F_WAIT, ctx, NULL);
5274 out:
5275 (void)vnode_put(vp);
5276 out1:
5277 fp_drop(p, fd, fp, 0);
5278 return(error);
5279
5280 }
5281
5282 /*
5283 * fileport_makeport
5284 *
5285 * Description: Obtain a Mach send right for a given file descriptor.
5286 *
5287 * Parameters: p Process calling fileport
5288 * uap->fd The fd to reference
5289 * uap->portnamep User address at which to place port name.
5290 *
5291 * Returns: 0 Success.
5292 * EBADF Bad file descriptor.
5293 * EINVAL File descriptor had type that cannot be sent, misc. other errors.
5294 * EFAULT Address at which to store port name is not valid.
5295 * EAGAIN Resource shortage.
5296 *
5297 * Implicit returns:
5298 * On success, name of send right is stored at user-specified address.
5299 */
5300 int
5301 fileport_makeport(proc_t p, struct fileport_makeport_args *uap,
5302 __unused int *retval)
5303 {
5304 int err;
5305 int fd = uap->fd;
5306 user_addr_t user_portaddr = uap->portnamep;
5307 struct fileproc *fp = FILEPROC_NULL;
5308 struct fileglob *fg = NULL;
5309 ipc_port_t fileport;
5310 mach_port_name_t name = MACH_PORT_NULL;
5311
5312 proc_fdlock(p);
5313 err = fp_lookup(p, fd, &fp, 1);
5314 if (err != 0) {
5315 goto out_unlock;
5316 }
5317
5318 if (!file_issendable(p, fp)) {
5319 err = EINVAL;
5320 goto out_unlock;
5321 }
5322
5323 if (FP_ISGUARDED(fp, GUARD_FILEPORT)) {
5324 err = fp_guard_exception(p, fd, fp, kGUARD_EXC_FILEPORT);
5325 goto out_unlock;
5326 }
5327
5328 /* Dropped when port is deallocated */
5329 fg = fp->f_fglob;
5330 fg_ref(fp);
5331
5332 proc_fdunlock(p);
5333
5334 /* Allocate and initialize a port */
5335 fileport = fileport_alloc(fg);
5336 if (fileport == IPC_PORT_NULL) {
5337 err = EAGAIN;
5338 fg_drop(fp);
5339 goto out;
5340 }
5341
5342 /* Add an entry. Deallocates port on failure. */
5343 name = ipc_port_copyout_send(fileport, get_task_ipcspace(p->task));
5344 if (!MACH_PORT_VALID(name)) {
5345 err = EINVAL;
5346 goto out;
5347 }
5348
5349 err = copyout(&name, user_portaddr, sizeof(mach_port_name_t));
5350 if (err != 0) {
5351 goto out;
5352 }
5353
5354 /* Tag the fileglob for debugging purposes */
5355 lck_mtx_lock_spin(&fg->fg_lock);
5356 fg->fg_lflags |= FG_PORTMADE;
5357 lck_mtx_unlock(&fg->fg_lock);
5358
5359 fp_drop(p, fd, fp, 0);
5360
5361 return 0;
5362
5363 out_unlock:
5364 proc_fdunlock(p);
5365 out:
5366 if (MACH_PORT_VALID(name)) {
5367 /* Don't care if another thread races us to deallocate the entry */
5368 (void) mach_port_deallocate(get_task_ipcspace(p->task), name);
5369 }
5370
5371 if (fp != FILEPROC_NULL) {
5372 fp_drop(p, fd, fp, 0);
5373 }
5374
5375 return err;
5376 }
5377
5378 void
5379 fileport_releasefg(struct fileglob *fg)
5380 {
5381 (void)closef_locked(NULL, fg, PROC_NULL);
5382
5383 return;
5384 }
5385
5386
5387 /*
5388 * fileport_makefd
5389 *
5390 * Description: Obtain the file descriptor for a given Mach send right.
5391 *
5392 * Parameters: p Process calling fileport
5393 * uap->port Name of send right to file port.
5394 *
5395 * Returns: 0 Success
5396 * EINVAL Invalid Mach port name, or port is not for a file.
5397 * fdalloc:EMFILE
5398 * fdalloc:ENOMEM Unable to allocate fileproc or extend file table.
5399 *
5400 * Implicit returns:
5401 * *retval (modified) The new descriptor
5402 */
5403 int
5404 fileport_makefd(proc_t p, struct fileport_makefd_args *uap, int32_t *retval)
5405 {
5406 struct fileglob *fg;
5407 struct fileproc *fp = FILEPROC_NULL;
5408 ipc_port_t port = IPC_PORT_NULL;
5409 mach_port_name_t send = uap->port;
5410 kern_return_t res;
5411 int fd;
5412 int err;
5413
5414 res = ipc_object_copyin(get_task_ipcspace(p->task),
5415 send, MACH_MSG_TYPE_COPY_SEND, &port);
5416
5417 if (res != KERN_SUCCESS) {
5418 err = EINVAL;
5419 goto out;
5420 }
5421
5422 fg = fileport_port_to_fileglob(port);
5423 if (fg == NULL) {
5424 err = EINVAL;
5425 goto out;
5426 }
5427
5428 fp = fileproc_alloc_init(NULL);
5429 if (fp == FILEPROC_NULL) {
5430 err = ENOMEM;
5431 goto out;
5432 }
5433
5434 fp->f_fglob = fg;
5435 fg_ref(fp);
5436
5437 proc_fdlock(p);
5438 err = fdalloc(p, 0, &fd);
5439 if (err != 0) {
5440 proc_fdunlock(p);
5441 goto out;
5442 }
5443 *fdflags(p, fd) |= UF_EXCLOSE;
5444
5445 procfdtbl_releasefd(p, fd, fp);
5446 proc_fdunlock(p);
5447
5448 *retval = fd;
5449 err = 0;
5450 out:
5451 if ((fp != NULL) && (0 != err)) {
5452 fileproc_free(fp);
5453 }
5454
5455 if (IPC_PORT_NULL != port) {
5456 ipc_port_release_send(port);
5457 }
5458
5459 return err;
5460 }
5461
5462
5463 /*
5464 * dupfdopen
5465 *
5466 * Description: Duplicate the specified descriptor to a free descriptor;
5467 * this is the second half of fdopen(), above.
5468 *
5469 * Parameters: fdp filedesc pointer to fill in
5470 * indx fd to dup to
5471 * dfd fd to dup from
5472 * mode mode to set on new fd
5473 * error command code
5474 *
5475 * Returns: 0 Success
5476 * EBADF Source fd is bad
5477 * EACCES Requested mode not allowed
5478 * !0 'error', if not ENODEV or
5479 * ENXIO
5480 *
5481 * Notes: XXX This is not thread safe; see fdopen() above
5482 */
5483 int
5484 dupfdopen(struct filedesc *fdp, int indx, int dfd, int flags, int error)
5485 {
5486 struct fileproc *wfp;
5487 struct fileproc *fp;
5488 #if CONFIG_MACF
5489 int myerror;
5490 #endif
5491 proc_t p = current_proc();
5492
5493 /*
5494 * If the to-be-dup'd fd number is greater than the allowed number
5495 * of file descriptors, or the fd to be dup'd has already been
5496 * closed, reject. Note, check for new == old is necessary as
5497 * falloc could allocate an already closed to-be-dup'd descriptor
5498 * as the new descriptor.
5499 */
5500 proc_fdlock(p);
5501
5502 fp = fdp->fd_ofiles[indx];
5503 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
5504 (wfp = fdp->fd_ofiles[dfd]) == NULL || wfp == fp ||
5505 (fdp->fd_ofileflags[dfd] & UF_RESERVED)) {
5506
5507 proc_fdunlock(p);
5508 return (EBADF);
5509 }
5510 #if CONFIG_MACF
5511 myerror = mac_file_check_dup(proc_ucred(p), wfp->f_fglob, dfd);
5512 if (myerror) {
5513 proc_fdunlock(p);
5514 return (myerror);
5515 }
5516 #endif
5517 /*
5518 * There are two cases of interest here.
5519 *
5520 * For ENODEV simply dup (dfd) to file descriptor
5521 * (indx) and return.
5522 *
5523 * For ENXIO steal away the file structure from (dfd) and
5524 * store it in (indx). (dfd) is effectively closed by
5525 * this operation.
5526 *
5527 * Any other error code is just returned.
5528 */
5529 switch (error) {
5530 case ENODEV:
5531 if (FP_ISGUARDED(wfp, GUARD_DUP)) {
5532 proc_fdunlock(p);
5533 return (EPERM);
5534 }
5535
5536 /*
5537 * Check that the mode the file is being opened for is a
5538 * subset of the mode of the existing descriptor.
5539 */
5540 if (((flags & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
5541 proc_fdunlock(p);
5542 return (EACCES);
5543 }
5544 if (indx > fdp->fd_lastfile)
5545 fdp->fd_lastfile = indx;
5546 (void)fg_ref(wfp);
5547
5548 if (fp->f_fglob)
5549 fg_free(fp->f_fglob);
5550 fp->f_fglob = wfp->f_fglob;
5551
5552 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd] |
5553 (flags & O_CLOEXEC) ? UF_EXCLOSE : 0;
5554
5555 proc_fdunlock(p);
5556 return (0);
5557
5558 default:
5559 proc_fdunlock(p);
5560 return (error);
5561 }
5562 /* NOTREACHED */
5563 }
5564
5565
5566 /*
5567 * fg_ref
5568 *
5569 * Description: Add a reference to a fileglob by fileproc
5570 *
5571 * Parameters: fp fileproc containing fileglob
5572 * pointer
5573 *
5574 * Returns: void
5575 *
5576 * Notes: XXX Should use OSAddAtomic?
5577 */
5578 void
5579 fg_ref(struct fileproc * fp)
5580 {
5581 struct fileglob *fg;
5582
5583 fg = fp->f_fglob;
5584
5585 lck_mtx_lock_spin(&fg->fg_lock);
5586
5587 #if DIAGNOSTIC
5588 if ((fp->f_flags & ~((unsigned int)FP_VALID_FLAGS)) != 0)
5589 panic("fg_ref: invalid bits on fp %p", fp);
5590
5591 if (fg->fg_count == 0)
5592 panic("fg_ref: adding fgcount to zeroed fg: fp %p fg %p",
5593 fp, fg);
5594 #endif
5595 fg->fg_count++;
5596 lck_mtx_unlock(&fg->fg_lock);
5597 }
5598
5599
5600 /*
5601 * fg_drop
5602 *
5603 * Description: Remove a reference to a fileglob by fileproc
5604 *
5605 * Parameters: fp fileproc containing fileglob
5606 * pointer
5607 *
5608 * Returns: void
5609 *
5610 * Notes: XXX Should use OSAddAtomic?
5611 */
5612 void
5613 fg_drop(struct fileproc * fp)
5614 {
5615 struct fileglob *fg;
5616
5617 fg = fp->f_fglob;
5618 lck_mtx_lock_spin(&fg->fg_lock);
5619 fg->fg_count--;
5620 lck_mtx_unlock(&fg->fg_lock);
5621 }
5622
5623 #if SOCKETS
5624 /*
5625 * fg_insertuipc_mark
5626 *
5627 * Description: Mark fileglob for insertion onto message queue if needed
5628 * Also takes fileglob reference
5629 *
5630 * Parameters: fg Fileglob pointer to insert
5631 *
5632 * Returns: true, if the fileglob needs to be inserted onto msg queue
5633 *
5634 * Locks: Takes and drops fg_lock, potentially many times
5635 */
5636 boolean_t
5637 fg_insertuipc_mark(struct fileglob * fg)
5638 {
5639 boolean_t insert = FALSE;
5640
5641 lck_mtx_lock_spin(&fg->fg_lock);
5642 while (fg->fg_lflags & FG_RMMSGQ) {
5643 lck_mtx_convert_spin(&fg->fg_lock);
5644
5645 fg->fg_lflags |= FG_WRMMSGQ;
5646 msleep(&fg->fg_lflags, &fg->fg_lock, 0, "fg_insertuipc", NULL);
5647 }
5648
5649 fg->fg_count++;
5650 fg->fg_msgcount++;
5651 if (fg->fg_msgcount == 1) {
5652 fg->fg_lflags |= FG_INSMSGQ;
5653 insert = TRUE;
5654 }
5655 lck_mtx_unlock(&fg->fg_lock);
5656 return (insert);
5657 }
5658
5659 /*
5660 * fg_insertuipc
5661 *
5662 * Description: Insert marked fileglob onto message queue
5663 *
5664 * Parameters: fg Fileglob pointer to insert
5665 *
5666 * Returns: void
5667 *
5668 * Locks: Takes and drops fg_lock & uipc_lock
5669 * DO NOT call this function with proc_fdlock held as unp_gc()
5670 * can potentially try to acquire proc_fdlock, which can result
5671 * in a deadlock if this function is in unp_gc_wait().
5672 */
5673 void
5674 fg_insertuipc(struct fileglob * fg)
5675 {
5676 if (fg->fg_lflags & FG_INSMSGQ) {
5677 lck_mtx_lock_spin(uipc_lock);
5678 unp_gc_wait();
5679 LIST_INSERT_HEAD(&fmsghead, fg, f_msglist);
5680 lck_mtx_unlock(uipc_lock);
5681 lck_mtx_lock(&fg->fg_lock);
5682 fg->fg_lflags &= ~FG_INSMSGQ;
5683 if (fg->fg_lflags & FG_WINSMSGQ) {
5684 fg->fg_lflags &= ~FG_WINSMSGQ;
5685 wakeup(&fg->fg_lflags);
5686 }
5687 lck_mtx_unlock(&fg->fg_lock);
5688 }
5689 }
5690
5691 /*
5692 * fg_removeuipc_mark
5693 *
5694 * Description: Mark the fileglob for removal from message queue if needed
5695 * Also releases fileglob message queue reference
5696 *
5697 * Parameters: fg Fileglob pointer to remove
5698 *
5699 * Returns: true, if the fileglob needs to be removed from msg queue
5700 *
5701 * Locks: Takes and drops fg_lock, potentially many times
5702 */
5703 boolean_t
5704 fg_removeuipc_mark(struct fileglob * fg)
5705 {
5706 boolean_t remove = FALSE;
5707
5708 lck_mtx_lock_spin(&fg->fg_lock);
5709 while (fg->fg_lflags & FG_INSMSGQ) {
5710 lck_mtx_convert_spin(&fg->fg_lock);
5711
5712 fg->fg_lflags |= FG_WINSMSGQ;
5713 msleep(&fg->fg_lflags, &fg->fg_lock, 0, "fg_removeuipc", NULL);
5714 }
5715 fg->fg_msgcount--;
5716 if (fg->fg_msgcount == 0) {
5717 fg->fg_lflags |= FG_RMMSGQ;
5718 remove = TRUE;
5719 }
5720 lck_mtx_unlock(&fg->fg_lock);
5721 return (remove);
5722 }
5723
5724 /*
5725 * fg_removeuipc
5726 *
5727 * Description: Remove marked fileglob from message queue
5728 *
5729 * Parameters: fg Fileglob pointer to remove
5730 *
5731 * Returns: void
5732 *
5733 * Locks: Takes and drops fg_lock & uipc_lock
5734 * DO NOT call this function with proc_fdlock held as unp_gc()
5735 * can potentially try to acquire proc_fdlock, which can result
5736 * in a deadlock if this function is in unp_gc_wait().
5737 */
5738 void
5739 fg_removeuipc(struct fileglob * fg)
5740 {
5741 if (fg->fg_lflags & FG_RMMSGQ) {
5742 lck_mtx_lock_spin(uipc_lock);
5743 unp_gc_wait();
5744 LIST_REMOVE(fg, f_msglist);
5745 lck_mtx_unlock(uipc_lock);
5746 lck_mtx_lock(&fg->fg_lock);
5747 fg->fg_lflags &= ~FG_RMMSGQ;
5748 if (fg->fg_lflags & FG_WRMMSGQ) {
5749 fg->fg_lflags &= ~FG_WRMMSGQ;
5750 wakeup(&fg->fg_lflags);
5751 }
5752 lck_mtx_unlock(&fg->fg_lock);
5753 }
5754 }
5755 #endif /* SOCKETS */
5756
5757 /*
5758 * fo_read
5759 *
5760 * Description: Generic fileops read indirected through the fileops pointer
5761 * in the fileproc structure
5762 *
5763 * Parameters: fp fileproc structure pointer
5764 * uio user I/O structure pointer
5765 * flags FOF_ flags
5766 * ctx VFS context for operation
5767 *
5768 * Returns: 0 Success
5769 * !0 Errno from read
5770 */
5771 int
5772 fo_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
5773 {
5774 return ((*fp->f_ops->fo_read)(fp, uio, flags, ctx));
5775 }
5776
5777
5778 /*
5779 * fo_write
5780 *
5781 * Description: Generic fileops write indirected through the fileops pointer
5782 * in the fileproc structure
5783 *
5784 * Parameters: fp fileproc structure pointer
5785 * uio user I/O structure pointer
5786 * flags FOF_ flags
5787 * ctx VFS context for operation
5788 *
5789 * Returns: 0 Success
5790 * !0 Errno from write
5791 */
5792 int
5793 fo_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
5794 {
5795 return((*fp->f_ops->fo_write)(fp, uio, flags, ctx));
5796 }
5797
5798
5799 /*
5800 * fo_ioctl
5801 *
5802 * Description: Generic fileops ioctl indirected through the fileops pointer
5803 * in the fileproc structure
5804 *
5805 * Parameters: fp fileproc structure pointer
5806 * com ioctl command
5807 * data pointer to internalized copy
5808 * of user space ioctl command
5809 * parameter data in kernel space
5810 * ctx VFS context for operation
5811 *
5812 * Returns: 0 Success
5813 * !0 Errno from ioctl
5814 *
5815 * Locks: The caller is assumed to have held the proc_fdlock; this
5816 * function releases and reacquires this lock. If the caller
5817 * accesses data protected by this lock prior to calling this
5818 * function, it will need to revalidate/reacquire any cached
5819 * protected data obtained prior to the call.
5820 */
5821 int
5822 fo_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
5823 {
5824 int error;
5825
5826 proc_fdunlock(vfs_context_proc(ctx));
5827 error = (*fp->f_ops->fo_ioctl)(fp, com, data, ctx);
5828 proc_fdlock(vfs_context_proc(ctx));
5829 return(error);
5830 }
5831
5832
5833 /*
5834 * fo_select
5835 *
5836 * Description: Generic fileops select indirected through the fileops pointer
5837 * in the fileproc structure
5838 *
5839 * Parameters: fp fileproc structure pointer
5840 * which select which
5841 * wql pointer to wait queue list
5842 * ctx VFS context for operation
5843 *
5844 * Returns: 0 Success
5845 * !0 Errno from select
5846 */
5847 int
5848 fo_select(struct fileproc *fp, int which, void *wql, vfs_context_t ctx)
5849 {
5850 return((*fp->f_ops->fo_select)(fp, which, wql, ctx));
5851 }
5852
5853
5854 /*
5855 * fo_close
5856 *
5857 * Description: Generic fileops close indirected through the fileops pointer
5858 * in the fileproc structure
5859 *
5860 * Parameters: fp fileproc structure pointer for
5861 * file to close
5862 * ctx VFS context for operation
5863 *
5864 * Returns: 0 Success
5865 * !0 Errno from close
5866 */
5867 int
5868 fo_close(struct fileglob *fg, vfs_context_t ctx)
5869 {
5870 return((*fg->fg_ops->fo_close)(fg, ctx));
5871 }
5872
5873
5874 /*
5875 * fo_kqfilter
5876 *
5877 * Description: Generic fileops kqueue filter indirected through the fileops
5878 * pointer in the fileproc structure
5879 *
5880 * Parameters: fp fileproc structure pointer
5881 * kn pointer to knote to filter on
5882 * ctx VFS context for operation
5883 *
5884 * Returns: 0 Success
5885 * !0 Errno from kqueue filter
5886 */
5887 int
5888 fo_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx)
5889 {
5890 return ((*fp->f_ops->fo_kqfilter)(fp, kn, ctx));
5891 }
5892
5893 /*
5894 * The ability to send a file descriptor to another
5895 * process is opt-in by file type.
5896 */
5897 boolean_t
5898 file_issendable(proc_t p, struct fileproc *fp)
5899 {
5900 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
5901
5902 switch (fp->f_type) {
5903 case DTYPE_VNODE:
5904 case DTYPE_SOCKET:
5905 case DTYPE_PIPE:
5906 case DTYPE_PSXSHM:
5907 return (0 == (fp->f_fglob->fg_lflags & FG_CONFINED));
5908 default:
5909 /* DTYPE_KQUEUE, DTYPE_FSEVENTS, DTYPE_PSXSEM */
5910 return FALSE;
5911 }
5912 }
5913
5914
5915 struct fileproc *
5916 fileproc_alloc_init(__unused void *arg)
5917 {
5918 struct fileproc *fp;
5919
5920 MALLOC_ZONE(fp, struct fileproc *, sizeof (*fp), M_FILEPROC, M_WAITOK);
5921 if (fp)
5922 bzero(fp, sizeof (*fp));
5923
5924 return (fp);
5925 }
5926
5927 void
5928 fileproc_free(struct fileproc *fp)
5929 {
5930 switch (FILEPROC_TYPE(fp)) {
5931 case FTYPE_SIMPLE:
5932 FREE_ZONE(fp, sizeof (*fp), M_FILEPROC);
5933 break;
5934 case FTYPE_GUARDED:
5935 guarded_fileproc_free(fp);
5936 break;
5937 default:
5938 panic("%s: corrupt fp %p flags %x", __func__, fp, fp->f_flags);
5939 }
5940 }
mirror of XNU