]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/kern_descrip.c
projects / apple / xnu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xnu-6153.81.5.tar.gz
[apple/xnu.git] / bsd / kern / kern_descrip.c
1 /*
2 * Copyright (c) 2000-2016 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/fsctl.h>
91 #include <sys/malloc.h>
92 #include <sys/mman.h>
93 #include <sys/syslog.h>
94 #include <sys/unistd.h>
95 #include <sys/resourcevar.h>
96 #include <sys/aio_kern.h>
97 #include <sys/ev.h>
98 #include <kern/locks.h>
99 #include <sys/uio_internal.h>
100 #include <sys/codesign.h>
101 #include <sys/codedir_internal.h>
102
103 #include <security/audit/audit.h>
104
105 #include <sys/mount_internal.h>
106 #include <sys/kdebug.h>
107 #include <sys/sysproto.h>
108 #include <sys/pipe.h>
109 #include <sys/spawn.h>
110 #include <sys/cprotect.h>
111 #include <kern/kern_types.h>
112 #include <kern/kalloc.h>
113 #include <kern/waitq.h>
114 #include <libkern/OSAtomic.h>
115
116 #include <sys/ubc_internal.h>
117
118 #include <kern/ipc_misc.h>
119 #include <vm/vm_protos.h>
120
121 #include <mach/mach_port.h>
122 #include <stdbool.h>
123
124 #if CONFIG_MACF
125 #include <security/mac_framework.h>
126 #endif
127
128 #define IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND 0x1
129 kern_return_t ipc_object_copyin(ipc_space_t, mach_port_name_t,
130 mach_msg_type_name_t, ipc_port_t *, mach_port_context_t, mach_msg_guard_flags_t *, uint32_t);
131 void ipc_port_release_send(ipc_port_t);
132
133 struct psemnode;
134 struct pshmnode;
135
136 static int finishdup(proc_t p,
137 struct filedesc *fdp, int old, int new, int flags, int32_t *retval);
138
139 int falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx, int locked);
140 void fg_drop(struct fileproc * fp);
141 void fg_free(struct fileglob *fg);
142 void fg_ref(struct fileproc * fp);
143 void fileport_releasefg(struct fileglob *fg);
144
145 /* flags for close_internal_locked */
146 #define FD_DUP2RESV 1
147
148 /* We don't want these exported */
149
150 __private_extern__
151 int unlink1(vfs_context_t, vnode_t, user_addr_t, enum uio_seg, int);
152
153 static void _fdrelse(struct proc * p, int fd);
154
155
156 extern void file_lock_init(void);
157
158 extern kauth_scope_t kauth_scope_fileop;
159
160 /* Conflict wait queue for when selects collide (opaque type) */
161 extern struct waitq select_conflict_queue;
162
163 #ifndef HFS_GET_BOOT_INFO
164 #define HFS_GET_BOOT_INFO (FCNTL_FS_SPECIFIC_BASE + 0x00004)
165 #endif
166
167 #ifndef HFS_SET_BOOT_INFO
168 #define HFS_SET_BOOT_INFO (FCNTL_FS_SPECIFIC_BASE + 0x00005)
169 #endif
170
171 #ifndef APFSIOC_REVERT_TO_SNAPSHOT
172 #define APFSIOC_REVERT_TO_SNAPSHOT _IOW('J', 1, u_int64_t)
173 #endif
174
175 #define f_flag f_fglob->fg_flag
176 #define f_type f_fglob->fg_ops->fo_type
177 #define f_msgcount f_fglob->fg_msgcount
178 #define f_cred f_fglob->fg_cred
179 #define f_ops f_fglob->fg_ops
180 #define f_offset f_fglob->fg_offset
181 #define f_data f_fglob->fg_data
182 #define CHECK_ADD_OVERFLOW_INT64L(x, y) \
183 (((((x) > 0) && ((y) > 0) && ((x) > LLONG_MAX - (y))) || \
184 (((x) < 0) && ((y) < 0) && ((x) < LLONG_MIN - (y)))) \
185 ? 1 : 0)
186 /*
187 * Descriptor management.
188 */
189 struct fmsglist fmsghead; /* head of list of open files */
190 struct fmsglist fmsg_ithead; /* head of list of open files */
191 int nfiles; /* actual number of open files */
192
193
194 lck_grp_attr_t * file_lck_grp_attr;
195 lck_grp_t * file_lck_grp;
196 lck_attr_t * file_lck_attr;
197
198 lck_mtx_t * uipc_lock;
199
200
201 /*
202 * check_file_seek_range
203 *
204 * Description: Checks if seek offsets are in the range of 0 to LLONG_MAX.
205 *
206 * Parameters: fl Flock structure.
207 * cur_file_offset Current offset in the file.
208 *
209 * Returns: 0 on Success.
210 * EOVERFLOW on overflow.
211 * EINVAL on offset less than zero.
212 */
213
214 static int
215 check_file_seek_range(struct flock *fl, off_t cur_file_offset)
216 {
217 if (fl->l_whence == SEEK_CUR) {
218 /* Check if the start marker is beyond LLONG_MAX. */
219 if (CHECK_ADD_OVERFLOW_INT64L(fl->l_start, cur_file_offset)) {
220 /* Check if start marker is negative */
221 if (fl->l_start < 0) {
222 return EINVAL;
223 }
224 return EOVERFLOW;
225 }
226 /* Check if the start marker is negative. */
227 if (fl->l_start + cur_file_offset < 0) {
228 return EINVAL;
229 }
230 /* Check if end marker is beyond LLONG_MAX. */
231 if ((fl->l_len > 0) && (CHECK_ADD_OVERFLOW_INT64L(fl->l_start +
232 cur_file_offset, fl->l_len - 1))) {
233 return EOVERFLOW;
234 }
235 /* Check if the end marker is negative. */
236 if ((fl->l_len <= 0) && (fl->l_start + cur_file_offset +
237 fl->l_len < 0)) {
238 return EINVAL;
239 }
240 } else if (fl->l_whence == SEEK_SET) {
241 /* Check if the start marker is negative. */
242 if (fl->l_start < 0) {
243 return EINVAL;
244 }
245 /* Check if the end marker is beyond LLONG_MAX. */
246 if ((fl->l_len > 0) &&
247 CHECK_ADD_OVERFLOW_INT64L(fl->l_start, fl->l_len - 1)) {
248 return EOVERFLOW;
249 }
250 /* Check if the end marker is negative. */
251 if ((fl->l_len < 0) && fl->l_start + fl->l_len < 0) {
252 return EINVAL;
253 }
254 }
255 return 0;
256 }
257
258
259 /*
260 * file_lock_init
261 *
262 * Description: Initialize the file lock group and the uipc and flist locks
263 *
264 * Parameters: (void)
265 *
266 * Returns: void
267 *
268 * Notes: Called at system startup from bsd_init().
269 */
270 void
271 file_lock_init(void)
272 {
273 /* allocate file lock group attribute and group */
274 file_lck_grp_attr = lck_grp_attr_alloc_init();
275
276 file_lck_grp = lck_grp_alloc_init("file", file_lck_grp_attr);
277
278 /* Allocate file lock attribute */
279 file_lck_attr = lck_attr_alloc_init();
280
281 uipc_lock = lck_mtx_alloc_init(file_lck_grp, file_lck_attr);
282 }
283
284
285 /*
286 * proc_fdlock, proc_fdlock_spin
287 *
288 * Description: Lock to control access to the per process struct fileproc
289 * and struct filedesc
290 *
291 * Parameters: p Process to take the lock on
292 *
293 * Returns: void
294 *
295 * Notes: The lock is initialized in forkproc() and destroyed in
296 * reap_child_process().
297 */
298 void
299 proc_fdlock(proc_t p)
300 {
301 lck_mtx_lock(&p->p_fdmlock);
302 }
303
304 void
305 proc_fdlock_spin(proc_t p)
306 {
307 lck_mtx_lock_spin(&p->p_fdmlock);
308 }
309
310 void
311 proc_fdlock_assert(proc_t p, int assertflags)
312 {
313 lck_mtx_assert(&p->p_fdmlock, assertflags);
314 }
315
316
317 /*
318 * proc_fdunlock
319 *
320 * Description: Unlock the lock previously locked by a call to proc_fdlock()
321 *
322 * Parameters: p Process to drop the lock on
323 *
324 * Returns: void
325 */
326 void
327 proc_fdunlock(proc_t p)
328 {
329 lck_mtx_unlock(&p->p_fdmlock);
330 }
331
332
333 /*
334 * System calls on descriptors.
335 */
336
337
338 /*
339 * getdtablesize
340 *
341 * Description: Returns the per process maximum size of the descriptor table
342 *
343 * Parameters: p Process being queried
344 * retval Pointer to the call return area
345 *
346 * Returns: 0 Success
347 *
348 * Implicit returns:
349 * *retval (modified) Size of dtable
350 */
351 int
352 getdtablesize(proc_t p, __unused struct getdtablesize_args *uap, int32_t *retval)
353 {
354 proc_fdlock_spin(p);
355 *retval = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
356 proc_fdunlock(p);
357
358 return 0;
359 }
360
361
362 void
363 procfdtbl_reservefd(struct proc * p, int fd)
364 {
365 p->p_fd->fd_ofiles[fd] = NULL;
366 p->p_fd->fd_ofileflags[fd] |= UF_RESERVED;
367 }
368
369 void
370 procfdtbl_markclosefd(struct proc * p, int fd)
371 {
372 p->p_fd->fd_ofileflags[fd] |= (UF_RESERVED | UF_CLOSING);
373 }
374
375 void
376 procfdtbl_releasefd(struct proc * p, int fd, struct fileproc * fp)
377 {
378 if (fp != NULL) {
379 p->p_fd->fd_ofiles[fd] = fp;
380 }
381 p->p_fd->fd_ofileflags[fd] &= ~UF_RESERVED;
382 if ((p->p_fd->fd_ofileflags[fd] & UF_RESVWAIT) == UF_RESVWAIT) {
383 p->p_fd->fd_ofileflags[fd] &= ~UF_RESVWAIT;
384 wakeup(&p->p_fd);
385 }
386 }
387
388 void
389 procfdtbl_waitfd(struct proc * p, int fd)
390 {
391 p->p_fd->fd_ofileflags[fd] |= UF_RESVWAIT;
392 msleep(&p->p_fd, &p->p_fdmlock, PRIBIO, "ftbl_waitfd", NULL);
393 }
394
395
396 void
397 procfdtbl_clearfd(struct proc * p, int fd)
398 {
399 int waiting;
400
401 waiting = (p->p_fd->fd_ofileflags[fd] & UF_RESVWAIT);
402 p->p_fd->fd_ofiles[fd] = NULL;
403 p->p_fd->fd_ofileflags[fd] = 0;
404 if (waiting == UF_RESVWAIT) {
405 wakeup(&p->p_fd);
406 }
407 }
408
409 /*
410 * _fdrelse
411 *
412 * Description: Inline utility function to free an fd in a filedesc
413 *
414 * Parameters: fdp Pointer to filedesc fd lies in
415 * fd fd to free
416 * reserv fd should be reserved
417 *
418 * Returns: void
419 *
420 * Locks: Assumes proc_fdlock for process pointing to fdp is held by
421 * the caller
422 */
423 static void
424 _fdrelse(struct proc * p, int fd)
425 {
426 struct filedesc *fdp = p->p_fd;
427 int nfd = 0;
428
429 if (fd < fdp->fd_freefile) {
430 fdp->fd_freefile = fd;
431 }
432 #if DIAGNOSTIC
433 if (fd > fdp->fd_lastfile) {
434 panic("fdrelse: fd_lastfile inconsistent");
435 }
436 #endif
437 procfdtbl_clearfd(p, fd);
438
439 while ((nfd = fdp->fd_lastfile) > 0 &&
440 fdp->fd_ofiles[nfd] == NULL &&
441 !(fdp->fd_ofileflags[nfd] & UF_RESERVED)) {
442 /* JMM - What about files with lingering EV_VANISHED knotes? */
443 fdp->fd_lastfile--;
444 }
445 }
446
447
448 int
449 fd_rdwr(
450 int fd,
451 enum uio_rw rw,
452 uint64_t base,
453 int64_t len,
454 enum uio_seg segflg,
455 off_t offset,
456 int io_flg,
457 int64_t *aresid)
458 {
459 struct fileproc *fp;
460 proc_t p;
461 int error = 0;
462 int flags = 0;
463 int spacetype;
464 uio_t auio = NULL;
465 char uio_buf[UIO_SIZEOF(1)];
466 struct vfs_context context = *(vfs_context_current());
467 bool wrote_some = false;
468
469 p = current_proc();
470
471 error = fp_lookup(p, fd, &fp, 0);
472 if (error) {
473 return error;
474 }
475
476 if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_PIPE && fp->f_type != DTYPE_SOCKET) {
477 error = EINVAL;
478 goto out;
479 }
480 if (rw == UIO_WRITE && !(fp->f_flag & FWRITE)) {
481 error = EBADF;
482 goto out;
483 }
484
485 if (rw == UIO_READ && !(fp->f_flag & FREAD)) {
486 error = EBADF;
487 goto out;
488 }
489
490 context.vc_ucred = fp->f_fglob->fg_cred;
491
492 if (UIO_SEG_IS_USER_SPACE(segflg)) {
493 spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
494 } else {
495 spacetype = UIO_SYSSPACE;
496 }
497
498 auio = uio_createwithbuffer(1, offset, spacetype, rw, &uio_buf[0], sizeof(uio_buf));
499
500 uio_addiov(auio, base, len);
501
502 if (!(io_flg & IO_APPEND)) {
503 flags = FOF_OFFSET;
504 }
505
506 if (rw == UIO_WRITE) {
507 user_ssize_t orig_resid = uio_resid(auio);
508 error = fo_write(fp, auio, flags, &context);
509 wrote_some = uio_resid(auio) < orig_resid;
510 } else {
511 error = fo_read(fp, auio, flags, &context);
512 }
513
514 if (aresid) {
515 *aresid = uio_resid(auio);
516 } else {
517 if (uio_resid(auio) && error == 0) {
518 error = EIO;
519 }
520 }
521 out:
522 if (wrote_some) {
523 fp_drop_written(p, fd, fp);
524 } else {
525 fp_drop(p, fd, fp, 0);
526 }
527
528 return error;
529 }
530
531
532
533 /*
534 * dup
535 *
536 * Description: Duplicate a file descriptor.
537 *
538 * Parameters: p Process performing the dup
539 * uap->fd The fd to dup
540 * retval Pointer to the call return area
541 *
542 * Returns: 0 Success
543 * !0 Errno
544 *
545 * Implicit returns:
546 * *retval (modified) The new descriptor
547 */
548 int
549 dup(proc_t p, struct dup_args *uap, int32_t *retval)
550 {
551 struct filedesc *fdp = p->p_fd;
552 int old = uap->fd;
553 int new, error;
554 struct fileproc *fp;
555
556 proc_fdlock(p);
557 if ((error = fp_lookup(p, old, &fp, 1))) {
558 proc_fdunlock(p);
559 return error;
560 }
561 if (FP_ISGUARDED(fp, GUARD_DUP)) {
562 error = fp_guard_exception(p, old, fp, kGUARD_EXC_DUP);
563 (void) fp_drop(p, old, fp, 1);
564 proc_fdunlock(p);
565 return error;
566 }
567 if ((error = fdalloc(p, 0, &new))) {
568 fp_drop(p, old, fp, 1);
569 proc_fdunlock(p);
570 return error;
571 }
572 error = finishdup(p, fdp, old, new, 0, retval);
573 fp_drop(p, old, fp, 1);
574 proc_fdunlock(p);
575
576 if (ENTR_SHOULDTRACE && fp->f_type == DTYPE_SOCKET) {
577 KERNEL_ENERGYTRACE(kEnTrActKernSocket, DBG_FUNC_START,
578 new, 0, (int64_t)VM_KERNEL_ADDRPERM(fp->f_data));
579 }
580
581 return error;
582 }
583
584 /*
585 * dup2
586 *
587 * Description: Duplicate a file descriptor to a particular value.
588 *
589 * Parameters: p Process performing the dup
590 * uap->from The fd to dup
591 * uap->to The fd to dup it to
592 * retval Pointer to the call return area
593 *
594 * Returns: 0 Success
595 * !0 Errno
596 *
597 * Implicit returns:
598 * *retval (modified) The new descriptor
599 */
600 int
601 dup2(proc_t p, struct dup2_args *uap, int32_t *retval)
602 {
603 struct filedesc *fdp = p->p_fd;
604 int old = uap->from, new = uap->to;
605 int i, error;
606 struct fileproc *fp, *nfp;
607
608 proc_fdlock(p);
609
610 startover:
611 if ((error = fp_lookup(p, old, &fp, 1))) {
612 proc_fdunlock(p);
613 return error;
614 }
615 if (FP_ISGUARDED(fp, GUARD_DUP)) {
616 error = fp_guard_exception(p, old, fp, kGUARD_EXC_DUP);
617 (void) fp_drop(p, old, fp, 1);
618 proc_fdunlock(p);
619 return error;
620 }
621 if (new < 0 ||
622 (rlim_t)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
623 new >= maxfiles) {
624 fp_drop(p, old, fp, 1);
625 proc_fdunlock(p);
626 return EBADF;
627 }
628 if (old == new) {
629 fp_drop(p, old, fp, 1);
630 *retval = new;
631 proc_fdunlock(p);
632 return 0;
633 }
634 if (new < 0 || new >= fdp->fd_nfiles) {
635 if ((error = fdalloc(p, new, &i))) {
636 fp_drop(p, old, fp, 1);
637 proc_fdunlock(p);
638 return error;
639 }
640 if (new != i) {
641 fdrelse(p, i);
642 goto closeit;
643 }
644 } else {
645 closeit:
646 while ((fdp->fd_ofileflags[new] & UF_RESERVED) == UF_RESERVED) {
647 fp_drop(p, old, fp, 1);
648 procfdtbl_waitfd(p, new);
649 #if DIAGNOSTIC
650 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
651 #endif
652 goto startover;
653 }
654
655 if ((fdp->fd_ofiles[new] != NULL) &&
656 ((error = fp_lookup(p, new, &nfp, 1)) == 0)) {
657 fp_drop(p, old, fp, 1);
658 if (FP_ISGUARDED(nfp, GUARD_CLOSE)) {
659 error = fp_guard_exception(p,
660 new, nfp, kGUARD_EXC_CLOSE);
661 (void) fp_drop(p, new, nfp, 1);
662 proc_fdunlock(p);
663 return error;
664 }
665 (void)close_internal_locked(p, new, nfp, FD_DUP2RESV);
666 #if DIAGNOSTIC
667 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
668 #endif
669 procfdtbl_clearfd(p, new);
670 goto startover;
671 } else {
672 #if DIAGNOSTIC
673 if (fdp->fd_ofiles[new] != NULL) {
674 panic("dup2: no ref on fileproc %d", new);
675 }
676 #endif
677 procfdtbl_reservefd(p, new);
678 }
679
680 #if DIAGNOSTIC
681 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
682 #endif
683 }
684 #if DIAGNOSTIC
685 if (fdp->fd_ofiles[new] != 0) {
686 panic("dup2: overwriting fd_ofiles with new %d", new);
687 }
688 if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0) {
689 panic("dup2: unreserved fileflags with new %d", new);
690 }
691 #endif
692 error = finishdup(p, fdp, old, new, 0, retval);
693 fp_drop(p, old, fp, 1);
694 proc_fdunlock(p);
695
696 return error;
697 }
698
699
700 /*
701 * fcntl
702 *
703 * Description: The file control system call.
704 *
705 * Parameters: p Process performing the fcntl
706 * uap->fd The fd to operate against
707 * uap->cmd The command to perform
708 * uap->arg Pointer to the command argument
709 * retval Pointer to the call return area
710 *
711 * Returns: 0 Success
712 * !0 Errno (see fcntl_nocancel)
713 *
714 * Implicit returns:
715 * *retval (modified) fcntl return value (if any)
716 *
717 * Notes: This system call differs from fcntl_nocancel() in that it
718 * tests for cancellation prior to performing a potentially
719 * blocking operation.
720 */
721 int
722 fcntl(proc_t p, struct fcntl_args *uap, int32_t *retval)
723 {
724 __pthread_testcancel(1);
725 return fcntl_nocancel(p, (struct fcntl_nocancel_args *)uap, retval);
726 }
727
728
729 /*
730 * fcntl_nocancel
731 *
732 * Description: A non-cancel-testing file control system call.
733 *
734 * Parameters: p Process performing the fcntl
735 * uap->fd The fd to operate against
736 * uap->cmd The command to perform
737 * uap->arg Pointer to the command argument
738 * retval Pointer to the call return area
739 *
740 * Returns: 0 Success
741 * EINVAL
742 * fp_lookup:EBADF Bad file descriptor
743 * [F_DUPFD]
744 * fdalloc:EMFILE
745 * fdalloc:ENOMEM
746 * finishdup:EBADF
747 * finishdup:ENOMEM
748 * [F_SETOWN]
749 * ESRCH
750 * [F_SETLK]
751 * EBADF
752 * EOVERFLOW
753 * copyin:EFAULT
754 * vnode_getwithref:???
755 * VNOP_ADVLOCK:???
756 * msleep:ETIMEDOUT
757 * [F_GETLK]
758 * EBADF
759 * EOVERFLOW
760 * copyin:EFAULT
761 * copyout:EFAULT
762 * vnode_getwithref:???
763 * VNOP_ADVLOCK:???
764 * [F_PREALLOCATE]
765 * EBADF
766 * EINVAL
767 * copyin:EFAULT
768 * copyout:EFAULT
769 * vnode_getwithref:???
770 * VNOP_ALLOCATE:???
771 * [F_SETSIZE,F_RDADVISE]
772 * EBADF
773 * copyin:EFAULT
774 * vnode_getwithref:???
775 * [F_RDAHEAD,F_NOCACHE]
776 * EBADF
777 * vnode_getwithref:???
778 * [???]
779 *
780 * Implicit returns:
781 * *retval (modified) fcntl return value (if any)
782 */
783 int
784 fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval)
785 {
786 int fd = uap->fd;
787 struct filedesc *fdp = p->p_fd;
788 struct fileproc *fp;
789 char *pop;
790 struct vnode *vp = NULLVP; /* for AUDIT_ARG() at end */
791 int i, tmp, error, error2, flg = 0;
792 struct flock fl = {};
793 struct flocktimeout fltimeout;
794 struct timespec *timeout = NULL;
795 struct vfs_context context;
796 off_t offset;
797 int newmin;
798 daddr64_t lbn, bn;
799 unsigned int fflag;
800 user_addr_t argp;
801 boolean_t is64bit;
802
803 AUDIT_ARG(fd, uap->fd);
804 AUDIT_ARG(cmd, uap->cmd);
805
806 proc_fdlock(p);
807 if ((error = fp_lookup(p, fd, &fp, 1))) {
808 proc_fdunlock(p);
809 return error;
810 }
811 context.vc_thread = current_thread();
812 context.vc_ucred = fp->f_cred;
813
814 is64bit = proc_is64bit(p);
815 if (is64bit) {
816 argp = uap->arg;
817 } else {
818 /*
819 * Since the arg parameter is defined as a long but may be
820 * either a long or a pointer we must take care to handle
821 * sign extension issues. Our sys call munger will sign
822 * extend a long when we are called from a 32-bit process.
823 * Since we can never have an address greater than 32-bits
824 * from a 32-bit process we lop off the top 32-bits to avoid
825 * getting the wrong address
826 */
827 argp = CAST_USER_ADDR_T((uint32_t)uap->arg);
828 }
829
830 pop = &fdp->fd_ofileflags[fd];
831
832 #if CONFIG_MACF
833 error = mac_file_check_fcntl(proc_ucred(p), fp->f_fglob, uap->cmd,
834 uap->arg);
835 if (error) {
836 goto out;
837 }
838 #endif
839
840 switch (uap->cmd) {
841 case F_DUPFD:
842 case F_DUPFD_CLOEXEC:
843 if (FP_ISGUARDED(fp, GUARD_DUP)) {
844 error = fp_guard_exception(p, fd, fp, kGUARD_EXC_DUP);
845 goto out;
846 }
847 newmin = CAST_DOWN_EXPLICIT(int, uap->arg); /* arg is an int, so we won't lose bits */
848 AUDIT_ARG(value32, newmin);
849 if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
850 newmin >= maxfiles) {
851 error = EINVAL;
852 goto out;
853 }
854 if ((error = fdalloc(p, newmin, &i))) {
855 goto out;
856 }
857 error = finishdup(p, fdp, fd, i,
858 uap->cmd == F_DUPFD_CLOEXEC ? UF_EXCLOSE : 0, retval);
859 goto out;
860
861 case F_GETFD:
862 *retval = (*pop & UF_EXCLOSE)? FD_CLOEXEC : 0;
863 error = 0;
864 goto out;
865
866 case F_SETFD:
867 AUDIT_ARG(value32, uap->arg);
868 if (uap->arg & FD_CLOEXEC) {
869 *pop |= UF_EXCLOSE;
870 } else {
871 if (FILEPROC_TYPE(fp) == FTYPE_GUARDED) {
872 error = fp_guard_exception(p,
873 fd, fp, kGUARD_EXC_NOCLOEXEC);
874 goto out;
875 }
876 *pop &= ~UF_EXCLOSE;
877 }
878 error = 0;
879 goto out;
880
881 case F_GETFL:
882 *retval = OFLAGS(fp->f_flag);
883 error = 0;
884 goto out;
885
886 case F_SETFL:
887 fp->f_flag &= ~FCNTLFLAGS;
888 tmp = CAST_DOWN_EXPLICIT(int, uap->arg); /* arg is an int, so we won't lose bits */
889 AUDIT_ARG(value32, tmp);
890 fp->f_flag |= FFLAGS(tmp) & FCNTLFLAGS;
891 tmp = fp->f_flag & FNONBLOCK;
892 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
893 if (error) {
894 goto out;
895 }
896 tmp = fp->f_flag & FASYNC;
897 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, &context);
898 if (!error) {
899 goto out;
900 }
901 fp->f_flag &= ~FNONBLOCK;
902 tmp = 0;
903 (void)fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
904 goto out;
905
906 case F_GETOWN:
907 if (fp->f_type == DTYPE_SOCKET) {
908 *retval = ((struct socket *)fp->f_data)->so_pgid;
909 error = 0;
910 goto out;
911 }
912 error = fo_ioctl(fp, TIOCGPGRP, (caddr_t)retval, &context);
913 *retval = -*retval;
914 goto out;
915
916 case F_SETOWN:
917 tmp = CAST_DOWN_EXPLICIT(pid_t, uap->arg); /* arg is an int, so we won't lose bits */
918 AUDIT_ARG(value32, tmp);
919 if (fp->f_type == DTYPE_SOCKET) {
920 ((struct socket *)fp->f_data)->so_pgid = tmp;
921 error = 0;
922 goto out;
923 }
924 if (fp->f_type == DTYPE_PIPE) {
925 error = fo_ioctl(fp, TIOCSPGRP, (caddr_t)&tmp, &context);
926 goto out;
927 }
928
929 if (tmp <= 0) {
930 tmp = -tmp;
931 } else {
932 proc_t p1 = proc_find(tmp);
933 if (p1 == 0) {
934 error = ESRCH;
935 goto out;
936 }
937 tmp = (int)p1->p_pgrpid;
938 proc_rele(p1);
939 }
940 error = fo_ioctl(fp, TIOCSPGRP, (caddr_t)&tmp, &context);
941 goto out;
942
943 case F_SETNOSIGPIPE:
944 tmp = CAST_DOWN_EXPLICIT(int, uap->arg);
945 if (fp->f_type == DTYPE_SOCKET) {
946 #if SOCKETS
947 error = sock_setsockopt((struct socket *)fp->f_data,
948 SOL_SOCKET, SO_NOSIGPIPE, &tmp, sizeof(tmp));
949 #else
950 error = EINVAL;
951 #endif
952 } else {
953 struct fileglob *fg = fp->f_fglob;
954
955 lck_mtx_lock_spin(&fg->fg_lock);
956 if (tmp) {
957 fg->fg_lflags |= FG_NOSIGPIPE;
958 } else {
959 fg->fg_lflags &= ~FG_NOSIGPIPE;
960 }
961 lck_mtx_unlock(&fg->fg_lock);
962 error = 0;
963 }
964 goto out;
965
966 case F_GETNOSIGPIPE:
967 if (fp->f_type == DTYPE_SOCKET) {
968 #if SOCKETS
969 int retsize = sizeof(*retval);
970 error = sock_getsockopt((struct socket *)fp->f_data,
971 SOL_SOCKET, SO_NOSIGPIPE, retval, &retsize);
972 #else
973 error = EINVAL;
974 #endif
975 } else {
976 *retval = (fp->f_fglob->fg_lflags & FG_NOSIGPIPE) ?
977 1 : 0;
978 error = 0;
979 }
980 goto out;
981
982 case F_SETCONFINED:
983 /*
984 * If this is the only reference to this fglob in the process
985 * and it's already marked as close-on-fork then mark it as
986 * (immutably) "confined" i.e. any fd that points to it will
987 * forever be close-on-fork, and attempts to use an IPC
988 * mechanism to move the descriptor elsewhere will fail.
989 */
990 if (CAST_DOWN_EXPLICIT(int, uap->arg)) {
991 struct fileglob *fg = fp->f_fglob;
992
993 lck_mtx_lock_spin(&fg->fg_lock);
994 if (fg->fg_lflags & FG_CONFINED) {
995 error = 0;
996 } else if (1 != fg->fg_count) {
997 error = EAGAIN; /* go close the dup .. */
998 } else if (UF_FORKCLOSE == (*pop & UF_FORKCLOSE)) {
999 fg->fg_lflags |= FG_CONFINED;
1000 error = 0;
1001 } else {
1002 error = EBADF; /* open without O_CLOFORK? */
1003 }
1004 lck_mtx_unlock(&fg->fg_lock);
1005 } else {
1006 /*
1007 * Other subsystems may have built on the immutability
1008 * of FG_CONFINED; clearing it may be tricky.
1009 */
1010 error = EPERM; /* immutable */
1011 }
1012 goto out;
1013
1014 case F_GETCONFINED:
1015 *retval = (fp->f_fglob->fg_lflags & FG_CONFINED) ? 1 : 0;
1016 error = 0;
1017 goto out;
1018
1019 case F_SETLKWTIMEOUT:
1020 case F_SETLKW:
1021 case F_OFD_SETLKWTIMEOUT:
1022 case F_OFD_SETLKW:
1023 flg |= F_WAIT;
1024 /* Fall into F_SETLK */
1025
1026 case F_SETLK:
1027 case F_OFD_SETLK:
1028 if (fp->f_type != DTYPE_VNODE) {
1029 error = EBADF;
1030 goto out;
1031 }
1032 vp = (struct vnode *)fp->f_data;
1033
1034 fflag = fp->f_flag;
1035 offset = fp->f_offset;
1036 proc_fdunlock(p);
1037
1038 /* Copy in the lock structure */
1039 if (F_SETLKWTIMEOUT == uap->cmd ||
1040 F_OFD_SETLKWTIMEOUT == uap->cmd) {
1041 error = copyin(argp, (caddr_t) &fltimeout, sizeof(fltimeout));
1042 if (error) {
1043 goto outdrop;
1044 }
1045 fl = fltimeout.fl;
1046 timeout = &fltimeout.timeout;
1047 } else {
1048 error = copyin(argp, (caddr_t)&fl, sizeof(fl));
1049 if (error) {
1050 goto outdrop;
1051 }
1052 }
1053
1054 /* Check starting byte and ending byte for EOVERFLOW in SEEK_CUR */
1055 /* and ending byte for EOVERFLOW in SEEK_SET */
1056 error = check_file_seek_range(&fl, offset);
1057 if (error) {
1058 goto outdrop;
1059 }
1060
1061 if ((error = vnode_getwithref(vp))) {
1062 goto outdrop;
1063 }
1064 if (fl.l_whence == SEEK_CUR) {
1065 fl.l_start += offset;
1066 }
1067
1068 #if CONFIG_MACF
1069 error = mac_file_check_lock(proc_ucred(p), fp->f_fglob,
1070 F_SETLK, &fl);
1071 if (error) {
1072 (void)vnode_put(vp);
1073 goto outdrop;
1074 }
1075 #endif
1076 switch (uap->cmd) {
1077 case F_OFD_SETLK:
1078 case F_OFD_SETLKW:
1079 case F_OFD_SETLKWTIMEOUT:
1080 flg |= F_OFD_LOCK;
1081 switch (fl.l_type) {
1082 case F_RDLCK:
1083 if ((fflag & FREAD) == 0) {
1084 error = EBADF;
1085 break;
1086 }
1087 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1088 F_SETLK, &fl, flg, &context, timeout);
1089 break;
1090 case F_WRLCK:
1091 if ((fflag & FWRITE) == 0) {
1092 error = EBADF;
1093 break;
1094 }
1095 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1096 F_SETLK, &fl, flg, &context, timeout);
1097 break;
1098 case F_UNLCK:
1099 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1100 F_UNLCK, &fl, F_OFD_LOCK, &context,
1101 timeout);
1102 break;
1103 default:
1104 error = EINVAL;
1105 break;
1106 }
1107 if (0 == error &&
1108 (F_RDLCK == fl.l_type || F_WRLCK == fl.l_type)) {
1109 struct fileglob *fg = fp->f_fglob;
1110
1111 /*
1112 * arrange F_UNLCK on last close (once
1113 * set, FG_HAS_OFDLOCK is immutable)
1114 */
1115 if ((fg->fg_lflags & FG_HAS_OFDLOCK) == 0) {
1116 lck_mtx_lock_spin(&fg->fg_lock);
1117 fg->fg_lflags |= FG_HAS_OFDLOCK;
1118 lck_mtx_unlock(&fg->fg_lock);
1119 }
1120 }
1121 break;
1122 default:
1123 flg |= F_POSIX;
1124 switch (fl.l_type) {
1125 case F_RDLCK:
1126 if ((fflag & FREAD) == 0) {
1127 error = EBADF;
1128 break;
1129 }
1130 // XXX UInt32 unsafe for LP64 kernel
1131 OSBitOrAtomic(P_LADVLOCK, &p->p_ladvflag);
1132 error = VNOP_ADVLOCK(vp, (caddr_t)p,
1133 F_SETLK, &fl, flg, &context, timeout);
1134 break;
1135 case F_WRLCK:
1136 if ((fflag & FWRITE) == 0) {
1137 error = EBADF;
1138 break;
1139 }
1140 // XXX UInt32 unsafe for LP64 kernel
1141 OSBitOrAtomic(P_LADVLOCK, &p->p_ladvflag);
1142 error = VNOP_ADVLOCK(vp, (caddr_t)p,
1143 F_SETLK, &fl, flg, &context, timeout);
1144 break;
1145 case F_UNLCK:
1146 error = VNOP_ADVLOCK(vp, (caddr_t)p,
1147 F_UNLCK, &fl, F_POSIX, &context, timeout);
1148 break;
1149 default:
1150 error = EINVAL;
1151 break;
1152 }
1153 break;
1154 }
1155 (void) vnode_put(vp);
1156 goto outdrop;
1157
1158 case F_GETLK:
1159 case F_OFD_GETLK:
1160 case F_GETLKPID:
1161 case F_OFD_GETLKPID:
1162 if (fp->f_type != DTYPE_VNODE) {
1163 error = EBADF;
1164 goto out;
1165 }
1166 vp = (struct vnode *)fp->f_data;
1167
1168 offset = fp->f_offset;
1169 proc_fdunlock(p);
1170
1171 /* Copy in the lock structure */
1172 error = copyin(argp, (caddr_t)&fl, sizeof(fl));
1173 if (error) {
1174 goto outdrop;
1175 }
1176
1177 /* Check starting byte and ending byte for EOVERFLOW in SEEK_CUR */
1178 /* and ending byte for EOVERFLOW in SEEK_SET */
1179 error = check_file_seek_range(&fl, offset);
1180 if (error) {
1181 goto outdrop;
1182 }
1183
1184 if ((fl.l_whence == SEEK_SET) && (fl.l_start < 0)) {
1185 error = EINVAL;
1186 goto outdrop;
1187 }
1188
1189 switch (fl.l_type) {
1190 case F_RDLCK:
1191 case F_UNLCK:
1192 case F_WRLCK:
1193 break;
1194 default:
1195 error = EINVAL;
1196 goto outdrop;
1197 }
1198
1199 switch (fl.l_whence) {
1200 case SEEK_CUR:
1201 case SEEK_SET:
1202 case SEEK_END:
1203 break;
1204 default:
1205 error = EINVAL;
1206 goto outdrop;
1207 }
1208
1209 if ((error = vnode_getwithref(vp)) == 0) {
1210 if (fl.l_whence == SEEK_CUR) {
1211 fl.l_start += offset;
1212 }
1213
1214 #if CONFIG_MACF
1215 error = mac_file_check_lock(proc_ucred(p), fp->f_fglob,
1216 uap->cmd, &fl);
1217 if (error == 0)
1218 #endif
1219 switch (uap->cmd) {
1220 case F_OFD_GETLK:
1221 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1222 F_GETLK, &fl, F_OFD_LOCK, &context, NULL);
1223 break;
1224 case F_OFD_GETLKPID:
1225 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
1226 F_GETLKPID, &fl, F_OFD_LOCK, &context, NULL);
1227 break;
1228 default:
1229 error = VNOP_ADVLOCK(vp, (caddr_t)p,
1230 uap->cmd, &fl, F_POSIX, &context, NULL);
1231 break;
1232 }
1233
1234 (void)vnode_put(vp);
1235
1236 if (error == 0) {
1237 error = copyout((caddr_t)&fl, argp, sizeof(fl));
1238 }
1239 }
1240 goto outdrop;
1241
1242 case F_PREALLOCATE: {
1243 fstore_t alloc_struct; /* structure for allocate command */
1244 u_int32_t alloc_flags = 0;
1245
1246 if (fp->f_type != DTYPE_VNODE) {
1247 error = EBADF;
1248 goto out;
1249 }
1250
1251 vp = (struct vnode *)fp->f_data;
1252 proc_fdunlock(p);
1253
1254 /* make sure that we have write permission */
1255 if ((fp->f_flag & FWRITE) == 0) {
1256 error = EBADF;
1257 goto outdrop;
1258 }
1259
1260 error = copyin(argp, (caddr_t)&alloc_struct, sizeof(alloc_struct));
1261 if (error) {
1262 goto outdrop;
1263 }
1264
1265 /* now set the space allocated to 0 */
1266 alloc_struct.fst_bytesalloc = 0;
1267
1268 /*
1269 * Do some simple parameter checking
1270 */
1271
1272 /* set up the flags */
1273
1274 alloc_flags |= PREALLOCATE;
1275
1276 if (alloc_struct.fst_flags & F_ALLOCATECONTIG) {
1277 alloc_flags |= ALLOCATECONTIG;
1278 }
1279
1280 if (alloc_struct.fst_flags & F_ALLOCATEALL) {
1281 alloc_flags |= ALLOCATEALL;
1282 }
1283
1284 /*
1285 * Do any position mode specific stuff. The only
1286 * position mode supported now is PEOFPOSMODE
1287 */
1288
1289 switch (alloc_struct.fst_posmode) {
1290 case F_PEOFPOSMODE:
1291 if (alloc_struct.fst_offset != 0) {
1292 error = EINVAL;
1293 goto outdrop;
1294 }
1295
1296 alloc_flags |= ALLOCATEFROMPEOF;
1297 break;
1298
1299 case F_VOLPOSMODE:
1300 if (alloc_struct.fst_offset <= 0) {
1301 error = EINVAL;
1302 goto outdrop;
1303 }
1304
1305 alloc_flags |= ALLOCATEFROMVOL;
1306 break;
1307
1308 default: {
1309 error = EINVAL;
1310 goto outdrop;
1311 }
1312 }
1313 if ((error = vnode_getwithref(vp)) == 0) {
1314 /*
1315 * call allocate to get the space
1316 */
1317 error = VNOP_ALLOCATE(vp, alloc_struct.fst_length, alloc_flags,
1318 &alloc_struct.fst_bytesalloc, alloc_struct.fst_offset,
1319 &context);
1320 (void)vnode_put(vp);
1321
1322 error2 = copyout((caddr_t)&alloc_struct, argp, sizeof(alloc_struct));
1323
1324 if (error == 0) {
1325 error = error2;
1326 }
1327 }
1328 goto outdrop;
1329 }
1330 case F_PUNCHHOLE: {
1331 fpunchhole_t args;
1332
1333 if (fp->f_type != DTYPE_VNODE) {
1334 error = EBADF;
1335 goto out;
1336 }
1337
1338 vp = (struct vnode *)fp->f_data;
1339 proc_fdunlock(p);
1340
1341 /* need write permissions */
1342 if ((fp->f_flag & FWRITE) == 0) {
1343 error = EPERM;
1344 goto outdrop;
1345 }
1346
1347 if ((error = copyin(argp, (caddr_t)&args, sizeof(args)))) {
1348 goto outdrop;
1349 }
1350
1351 if ((error = vnode_getwithref(vp))) {
1352 goto outdrop;
1353 }
1354
1355 #if CONFIG_MACF
1356 if ((error = mac_vnode_check_write(&context, fp->f_fglob->fg_cred, vp))) {
1357 (void)vnode_put(vp);
1358 goto outdrop;
1359 }
1360 #endif
1361
1362 error = VNOP_IOCTL(vp, F_PUNCHHOLE, (caddr_t)&args, 0, &context);
1363 (void)vnode_put(vp);
1364
1365 goto outdrop;
1366 }
1367 case F_TRIM_ACTIVE_FILE: {
1368 ftrimactivefile_t args;
1369
1370 if (priv_check_cred(kauth_cred_get(), PRIV_TRIM_ACTIVE_FILE, 0)) {
1371 error = EACCES;
1372 goto out;
1373 }
1374
1375 if (fp->f_type != DTYPE_VNODE) {
1376 error = EBADF;
1377 goto out;
1378 }
1379
1380 vp = (struct vnode *)fp->f_data;
1381 proc_fdunlock(p);
1382
1383 /* need write permissions */
1384 if ((fp->f_flag & FWRITE) == 0) {
1385 error = EPERM;
1386 goto outdrop;
1387 }
1388
1389 if ((error = copyin(argp, (caddr_t)&args, sizeof(args)))) {
1390 goto outdrop;
1391 }
1392
1393 if ((error = vnode_getwithref(vp))) {
1394 goto outdrop;
1395 }
1396
1397 error = VNOP_IOCTL(vp, F_TRIM_ACTIVE_FILE, (caddr_t)&args, 0, &context);
1398 (void)vnode_put(vp);
1399
1400 goto outdrop;
1401 }
1402 case F_SPECULATIVE_READ: {
1403 fspecread_t args;
1404
1405 if (fp->f_type != DTYPE_VNODE) {
1406 error = EBADF;
1407 goto out;
1408 }
1409
1410 vp = (struct vnode *)fp->f_data;
1411 proc_fdunlock(p);
1412
1413 if ((error = copyin(argp, (caddr_t)&args, sizeof(args)))) {
1414 goto outdrop;
1415 }
1416
1417 /* Discard invalid offsets or lengths */
1418 if ((args.fsr_offset < 0) || (args.fsr_length < 0)) {
1419 error = EINVAL;
1420 goto outdrop;
1421 }
1422
1423 /*
1424 * Round the file offset down to a page-size boundary (or to 0).
1425 * The filesystem will need to round the length up to the end of the page boundary
1426 * or to the EOF of the file.
1427 */
1428 uint64_t foff = (((uint64_t)args.fsr_offset) & ~((uint64_t)PAGE_MASK));
1429 uint64_t foff_delta = args.fsr_offset - foff;
1430 args.fsr_offset = (off_t) foff;
1431
1432 /*
1433 * Now add in the delta to the supplied length. Since we may have adjusted the
1434 * offset, increase it by the amount that we adjusted.
1435 */
1436 args.fsr_length += foff_delta;
1437
1438 if ((error = vnode_getwithref(vp))) {
1439 goto outdrop;
1440 }
1441 error = VNOP_IOCTL(vp, F_SPECULATIVE_READ, (caddr_t)&args, 0, &context);
1442 (void)vnode_put(vp);
1443
1444 goto outdrop;
1445 }
1446 case F_SETSIZE:
1447 if (fp->f_type != DTYPE_VNODE) {
1448 error = EBADF;
1449 goto out;
1450 }
1451 vp = (struct vnode *)fp->f_data;
1452 proc_fdunlock(p);
1453
1454 error = copyin(argp, (caddr_t)&offset, sizeof(off_t));
1455 if (error) {
1456 goto outdrop;
1457 }
1458 AUDIT_ARG(value64, offset);
1459
1460 error = vnode_getwithref(vp);
1461 if (error) {
1462 goto outdrop;
1463 }
1464
1465 #if CONFIG_MACF
1466 error = mac_vnode_check_truncate(&context,
1467 fp->f_fglob->fg_cred, vp);
1468 if (error) {
1469 (void)vnode_put(vp);
1470 goto outdrop;
1471 }
1472 #endif
1473 /*
1474 * Make sure that we are root. Growing a file
1475 * without zero filling the data is a security hole
1476 * root would have access anyway so we'll allow it
1477 */
1478 if (!kauth_cred_issuser(kauth_cred_get())) {
1479 error = EACCES;
1480 } else {
1481 /*
1482 * set the file size
1483 */
1484 error = vnode_setsize(vp, offset, IO_NOZEROFILL,
1485 &context);
1486
1487 #if CONFIG_MACF
1488 if (error == 0) {
1489 mac_vnode_notify_truncate(&context, fp->f_fglob->fg_cred, vp);
1490 }
1491 #endif
1492 }
1493
1494 (void)vnode_put(vp);
1495 goto outdrop;
1496
1497 case F_RDAHEAD:
1498 if (fp->f_type != DTYPE_VNODE) {
1499 error = EBADF;
1500 goto out;
1501 }
1502 if (uap->arg) {
1503 fp->f_fglob->fg_flag &= ~FNORDAHEAD;
1504 } else {
1505 fp->f_fglob->fg_flag |= FNORDAHEAD;
1506 }
1507
1508 goto out;
1509
1510 case F_NOCACHE:
1511 if (fp->f_type != DTYPE_VNODE) {
1512 error = EBADF;
1513 goto out;
1514 }
1515 if (uap->arg) {
1516 fp->f_fglob->fg_flag |= FNOCACHE;
1517 } else {
1518 fp->f_fglob->fg_flag &= ~FNOCACHE;
1519 }
1520
1521 goto out;
1522
1523 case F_NODIRECT:
1524 if (fp->f_type != DTYPE_VNODE) {
1525 error = EBADF;
1526 goto out;
1527 }
1528 if (uap->arg) {
1529 fp->f_fglob->fg_flag |= FNODIRECT;
1530 } else {
1531 fp->f_fglob->fg_flag &= ~FNODIRECT;
1532 }
1533
1534 goto out;
1535
1536 case F_SINGLE_WRITER:
1537 if (fp->f_type != DTYPE_VNODE) {
1538 error = EBADF;
1539 goto out;
1540 }
1541 if (uap->arg) {
1542 fp->f_fglob->fg_flag |= FSINGLE_WRITER;
1543 } else {
1544 fp->f_fglob->fg_flag &= ~FSINGLE_WRITER;
1545 }
1546
1547 goto out;
1548
1549 case F_GLOBAL_NOCACHE:
1550 if (fp->f_type != DTYPE_VNODE) {
1551 error = EBADF;
1552 goto out;
1553 }
1554 vp = (struct vnode *)fp->f_data;
1555 proc_fdunlock(p);
1556
1557 if ((error = vnode_getwithref(vp)) == 0) {
1558 *retval = vnode_isnocache(vp);
1559
1560 if (uap->arg) {
1561 vnode_setnocache(vp);
1562 } else {
1563 vnode_clearnocache(vp);
1564 }
1565
1566 (void)vnode_put(vp);
1567 }
1568 goto outdrop;
1569
1570 case F_CHECK_OPENEVT:
1571 if (fp->f_type != DTYPE_VNODE) {
1572 error = EBADF;
1573 goto out;
1574 }
1575 vp = (struct vnode *)fp->f_data;
1576 proc_fdunlock(p);
1577
1578 if ((error = vnode_getwithref(vp)) == 0) {
1579 *retval = vnode_is_openevt(vp);
1580
1581 if (uap->arg) {
1582 vnode_set_openevt(vp);
1583 } else {
1584 vnode_clear_openevt(vp);
1585 }
1586
1587 (void)vnode_put(vp);
1588 }
1589 goto outdrop;
1590
1591 case F_RDADVISE: {
1592 struct radvisory ra_struct;
1593
1594 if (fp->f_type != DTYPE_VNODE) {
1595 error = EBADF;
1596 goto out;
1597 }
1598 vp = (struct vnode *)fp->f_data;
1599 proc_fdunlock(p);
1600
1601 if ((error = copyin(argp, (caddr_t)&ra_struct, sizeof(ra_struct)))) {
1602 goto outdrop;
1603 }
1604 if ((error = vnode_getwithref(vp)) == 0) {
1605 error = VNOP_IOCTL(vp, F_RDADVISE, (caddr_t)&ra_struct, 0, &context);
1606
1607 (void)vnode_put(vp);
1608 }
1609 goto outdrop;
1610 }
1611
1612 case F_FLUSH_DATA:
1613
1614 if (fp->f_type != DTYPE_VNODE) {
1615 error = EBADF;
1616 goto out;
1617 }
1618 vp = (struct vnode *)fp->f_data;
1619 proc_fdunlock(p);
1620
1621 if ((error = vnode_getwithref(vp)) == 0) {
1622 error = VNOP_FSYNC(vp, MNT_NOWAIT, &context);
1623
1624 (void)vnode_put(vp);
1625 }
1626 goto outdrop;
1627
1628 case F_LOG2PHYS:
1629 case F_LOG2PHYS_EXT: {
1630 struct log2phys l2p_struct = {}; /* structure for allocate command */
1631 int devBlockSize;
1632
1633 off_t file_offset = 0;
1634 size_t a_size = 0;
1635 size_t run = 0;
1636
1637 if (uap->cmd == F_LOG2PHYS_EXT) {
1638 error = copyin(argp, (caddr_t)&l2p_struct, sizeof(l2p_struct));
1639 if (error) {
1640 goto out;
1641 }
1642 file_offset = l2p_struct.l2p_devoffset;
1643 } else {
1644 file_offset = fp->f_offset;
1645 }
1646 if (fp->f_type != DTYPE_VNODE) {
1647 error = EBADF;
1648 goto out;
1649 }
1650 vp = (struct vnode *)fp->f_data;
1651 proc_fdunlock(p);
1652 if ((error = vnode_getwithref(vp))) {
1653 goto outdrop;
1654 }
1655 error = VNOP_OFFTOBLK(vp, file_offset, &lbn);
1656 if (error) {
1657 (void)vnode_put(vp);
1658 goto outdrop;
1659 }
1660 error = VNOP_BLKTOOFF(vp, lbn, &offset);
1661 if (error) {
1662 (void)vnode_put(vp);
1663 goto outdrop;
1664 }
1665 devBlockSize = vfs_devblocksize(vnode_mount(vp));
1666 if (uap->cmd == F_LOG2PHYS_EXT) {
1667 if (l2p_struct.l2p_contigbytes < 0) {
1668 vnode_put(vp);
1669 error = EINVAL;
1670 goto outdrop;
1671 }
1672
1673 a_size = MIN((uint64_t)l2p_struct.l2p_contigbytes, SIZE_MAX);
1674 } else {
1675 a_size = devBlockSize;
1676 }
1677
1678 error = VNOP_BLOCKMAP(vp, offset, a_size, &bn, &run, NULL, 0, &context);
1679
1680 (void)vnode_put(vp);
1681
1682 if (!error) {
1683 l2p_struct.l2p_flags = 0; /* for now */
1684 if (uap->cmd == F_LOG2PHYS_EXT) {
1685 l2p_struct.l2p_contigbytes = run - (file_offset - offset);
1686 } else {
1687 l2p_struct.l2p_contigbytes = 0; /* for now */
1688 }
1689
1690 /*
1691 * The block number being -1 suggests that the file offset is not backed
1692 * by any real blocks on-disk. As a result, just let it be passed back up wholesale.
1693 */
1694 if (bn == -1) {
1695 /* Don't multiply it by the block size */
1696 l2p_struct.l2p_devoffset = bn;
1697 } else {
1698 l2p_struct.l2p_devoffset = bn * devBlockSize;
1699 l2p_struct.l2p_devoffset += file_offset - offset;
1700 }
1701 error = copyout((caddr_t)&l2p_struct, argp, sizeof(l2p_struct));
1702 }
1703 goto outdrop;
1704 }
1705 case F_GETPATH:
1706 case F_GETPATH_NOFIRMLINK: {
1707 char *pathbufp;
1708 int pathlen;
1709
1710 if (fp->f_type != DTYPE_VNODE) {
1711 error = EBADF;
1712 goto out;
1713 }
1714 vp = (struct vnode *)fp->f_data;
1715 proc_fdunlock(p);
1716
1717 pathlen = MAXPATHLEN;
1718 MALLOC(pathbufp, char *, pathlen, M_TEMP, M_WAITOK);
1719 if (pathbufp == NULL) {
1720 error = ENOMEM;
1721 goto outdrop;
1722 }
1723 if ((error = vnode_getwithref(vp)) == 0) {
1724 if (uap->cmd == F_GETPATH_NOFIRMLINK) {
1725 error = vn_getpath_ext(vp, NULL, pathbufp, &pathlen, VN_GETPATH_NO_FIRMLINK);
1726 } else {
1727 error = vn_getpath(vp, pathbufp, &pathlen);
1728 }
1729 (void)vnode_put(vp);
1730
1731 if (error == 0) {
1732 error = copyout((caddr_t)pathbufp, argp, pathlen);
1733 }
1734 }
1735 FREE(pathbufp, M_TEMP);
1736 goto outdrop;
1737 }
1738
1739 case F_PATHPKG_CHECK: {
1740 char *pathbufp;
1741 size_t pathlen;
1742
1743 if (fp->f_type != DTYPE_VNODE) {
1744 error = EBADF;
1745 goto out;
1746 }
1747 vp = (struct vnode *)fp->f_data;
1748 proc_fdunlock(p);
1749
1750 pathlen = MAXPATHLEN;
1751 pathbufp = kalloc(MAXPATHLEN);
1752
1753 if ((error = copyinstr(argp, pathbufp, MAXPATHLEN, &pathlen)) == 0) {
1754 if ((error = vnode_getwithref(vp)) == 0) {
1755 AUDIT_ARG(text, pathbufp);
1756 error = vn_path_package_check(vp, pathbufp, pathlen, retval);
1757
1758 (void)vnode_put(vp);
1759 }
1760 }
1761 kfree(pathbufp, MAXPATHLEN);
1762 goto outdrop;
1763 }
1764
1765 case F_CHKCLEAN: // used by regression tests to see if all dirty pages got cleaned by fsync()
1766 case F_FULLFSYNC: // fsync + flush the journal + DKIOCSYNCHRONIZE
1767 case F_BARRIERFSYNC: // fsync + barrier
1768 case F_FREEZE_FS: // freeze all other fs operations for the fs of this fd
1769 case F_THAW_FS: { // thaw all frozen fs operations for the fs of this fd
1770 if (fp->f_type != DTYPE_VNODE) {
1771 error = EBADF;
1772 goto out;
1773 }
1774 vp = (struct vnode *)fp->f_data;
1775 proc_fdunlock(p);
1776
1777 if ((error = vnode_getwithref(vp)) == 0) {
1778 error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)NULL, 0, &context);
1779
1780 (void)vnode_put(vp);
1781 }
1782 break;
1783 }
1784
1785 /*
1786 * SPI (private) for opening a file starting from a dir fd
1787 */
1788 case F_OPENFROM: {
1789 struct user_fopenfrom fopen;
1790 struct vnode_attr va;
1791 struct nameidata nd;
1792 int cmode;
1793
1794 /* Check if this isn't a valid file descriptor */
1795 if ((fp->f_type != DTYPE_VNODE) ||
1796 (fp->f_flag & FREAD) == 0) {
1797 error = EBADF;
1798 goto out;
1799 }
1800 vp = (struct vnode *)fp->f_data;
1801 proc_fdunlock(p);
1802
1803 if (vnode_getwithref(vp)) {
1804 error = ENOENT;
1805 goto outdrop;
1806 }
1807
1808 /* Only valid for directories */
1809 if (vp->v_type != VDIR) {
1810 vnode_put(vp);
1811 error = ENOTDIR;
1812 goto outdrop;
1813 }
1814
1815 /* Get flags, mode and pathname arguments. */
1816 if (IS_64BIT_PROCESS(p)) {
1817 error = copyin(argp, &fopen, sizeof(fopen));
1818 } else {
1819 struct user32_fopenfrom fopen32;
1820
1821 error = copyin(argp, &fopen32, sizeof(fopen32));
1822 fopen.o_flags = fopen32.o_flags;
1823 fopen.o_mode = fopen32.o_mode;
1824 fopen.o_pathname = CAST_USER_ADDR_T(fopen32.o_pathname);
1825 }
1826 if (error) {
1827 vnode_put(vp);
1828 goto outdrop;
1829 }
1830 AUDIT_ARG(fflags, fopen.o_flags);
1831 AUDIT_ARG(mode, fopen.o_mode);
1832 VATTR_INIT(&va);
1833 /* Mask off all but regular access permissions */
1834 cmode = ((fopen.o_mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1835 VATTR_SET(&va, va_mode, cmode & ACCESSPERMS);
1836
1837 /* Start the lookup relative to the file descriptor's vnode. */
1838 NDINIT(&nd, LOOKUP, OP_OPEN, USEDVP | FOLLOW | AUDITVNPATH1, UIO_USERSPACE,
1839 fopen.o_pathname, &context);
1840 nd.ni_dvp = vp;
1841
1842 error = open1(&context, &nd, fopen.o_flags, &va,
1843 fileproc_alloc_init, NULL, retval);
1844
1845 vnode_put(vp);
1846 break;
1847 }
1848 /*
1849 * SPI (private) for unlinking a file starting from a dir fd
1850 */
1851 case F_UNLINKFROM: {
1852 user_addr_t pathname;
1853
1854 /* Check if this isn't a valid file descriptor */
1855 if ((fp->f_type != DTYPE_VNODE) ||
1856 (fp->f_flag & FREAD) == 0) {
1857 error = EBADF;
1858 goto out;
1859 }
1860 vp = (struct vnode *)fp->f_data;
1861 proc_fdunlock(p);
1862
1863 if (vnode_getwithref(vp)) {
1864 error = ENOENT;
1865 goto outdrop;
1866 }
1867
1868 /* Only valid for directories */
1869 if (vp->v_type != VDIR) {
1870 vnode_put(vp);
1871 error = ENOTDIR;
1872 goto outdrop;
1873 }
1874
1875 /* Get flags, mode and pathname arguments. */
1876 if (IS_64BIT_PROCESS(p)) {
1877 pathname = (user_addr_t)argp;
1878 } else {
1879 pathname = CAST_USER_ADDR_T(argp);
1880 }
1881
1882 /* Start the lookup relative to the file descriptor's vnode. */
1883 error = unlink1(&context, vp, pathname, UIO_USERSPACE, 0);
1884
1885 vnode_put(vp);
1886 break;
1887 }
1888
1889 case F_ADDSIGS:
1890 case F_ADDFILESIGS:
1891 case F_ADDFILESIGS_FOR_DYLD_SIM:
1892 case F_ADDFILESIGS_RETURN:
1893 {
1894 struct cs_blob *blob = NULL;
1895 struct user_fsignatures fs;
1896 kern_return_t kr;
1897 vm_offset_t kernel_blob_addr;
1898 vm_size_t kernel_blob_size;
1899 int blob_add_flags = 0;
1900
1901 if (fp->f_type != DTYPE_VNODE) {
1902 error = EBADF;
1903 goto out;
1904 }
1905 vp = (struct vnode *)fp->f_data;
1906 proc_fdunlock(p);
1907
1908 if (uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM) {
1909 blob_add_flags |= MAC_VNODE_CHECK_DYLD_SIM;
1910 if ((p->p_csflags & CS_KILL) == 0) {
1911 proc_lock(p);
1912 p->p_csflags |= CS_KILL;
1913 proc_unlock(p);
1914 }
1915 }
1916
1917 error = vnode_getwithref(vp);
1918 if (error) {
1919 goto outdrop;
1920 }
1921
1922 if (IS_64BIT_PROCESS(p)) {
1923 error = copyin(argp, &fs, sizeof(fs));
1924 } else {
1925 struct user32_fsignatures fs32;
1926
1927 error = copyin(argp, &fs32, sizeof(fs32));
1928 fs.fs_file_start = fs32.fs_file_start;
1929 fs.fs_blob_start = CAST_USER_ADDR_T(fs32.fs_blob_start);
1930 fs.fs_blob_size = fs32.fs_blob_size;
1931 }
1932
1933 if (error) {
1934 vnode_put(vp);
1935 goto outdrop;
1936 }
1937
1938 /*
1939 * First check if we have something loaded a this offset
1940 */
1941 blob = ubc_cs_blob_get(vp, CPU_TYPE_ANY, fs.fs_file_start);
1942 if (blob != NULL) {
1943 /* If this is for dyld_sim revalidate the blob */
1944 if (uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM) {
1945 error = ubc_cs_blob_revalidate(vp, blob, NULL, blob_add_flags);
1946 if (error) {
1947 blob = NULL;
1948 if (error != EAGAIN) {
1949 vnode_put(vp);
1950 goto outdrop;
1951 }
1952 }
1953 }
1954 }
1955
1956 if (blob == NULL) {
1957 /*
1958 * An arbitrary limit, to prevent someone from mapping in a 20GB blob. This should cover
1959 * our use cases for the immediate future, but note that at the time of this commit, some
1960 * platforms are nearing 2MB blob sizes (with a prior soft limit of 2.5MB).
1961 *
1962 * We should consider how we can manage this more effectively; the above means that some
1963 * platforms are using megabytes of memory for signing data; it merely hasn't crossed the
1964 * threshold considered ridiculous at the time of this change.
1965 */
1966 #define CS_MAX_BLOB_SIZE (40ULL * 1024ULL * 1024ULL)
1967 if (fs.fs_blob_size > CS_MAX_BLOB_SIZE) {
1968 error = E2BIG;
1969 vnode_put(vp);
1970 goto outdrop;
1971 }
1972
1973 kernel_blob_size = CAST_DOWN(vm_size_t, fs.fs_blob_size);
1974 kr = ubc_cs_blob_allocate(&kernel_blob_addr, &kernel_blob_size);
1975 if (kr != KERN_SUCCESS) {
1976 error = ENOMEM;
1977 vnode_put(vp);
1978 goto outdrop;
1979 }
1980
1981 if (uap->cmd == F_ADDSIGS) {
1982 error = copyin(fs.fs_blob_start,
1983 (void *) kernel_blob_addr,
1984 kernel_blob_size);
1985 } else { /* F_ADDFILESIGS || F_ADDFILESIGS_RETURN || F_ADDFILESIGS_FOR_DYLD_SIM */
1986 int resid;
1987
1988 error = vn_rdwr(UIO_READ,
1989 vp,
1990 (caddr_t) kernel_blob_addr,
1991 kernel_blob_size,
1992 fs.fs_file_start + fs.fs_blob_start,
1993 UIO_SYSSPACE,
1994 0,
1995 kauth_cred_get(),
1996 &resid,
1997 p);
1998 if ((error == 0) && resid) {
1999 /* kernel_blob_size rounded to a page size, but signature may be at end of file */
2000 memset((void *)(kernel_blob_addr + (kernel_blob_size - resid)), 0x0, resid);
2001 }
2002 }
2003
2004 if (error) {
2005 ubc_cs_blob_deallocate(kernel_blob_addr,
2006 kernel_blob_size);
2007 vnode_put(vp);
2008 goto outdrop;
2009 }
2010
2011 blob = NULL;
2012 error = ubc_cs_blob_add(vp,
2013 CPU_TYPE_ANY, /* not for a specific architecture */
2014 fs.fs_file_start,
2015 &kernel_blob_addr,
2016 kernel_blob_size,
2017 NULL,
2018 blob_add_flags,
2019 &blob);
2020
2021 /* ubc_blob_add() has consumed "kernel_blob_addr" if it is zeroed */
2022 if (error) {
2023 if (kernel_blob_addr) {
2024 ubc_cs_blob_deallocate(kernel_blob_addr,
2025 kernel_blob_size);
2026 }
2027 vnode_put(vp);
2028 goto outdrop;
2029 } else {
2030 #if CHECK_CS_VALIDATION_BITMAP
2031 ubc_cs_validation_bitmap_allocate( vp );
2032 #endif
2033 }
2034 }
2035
2036 if (uap->cmd == F_ADDFILESIGS_RETURN || uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM) {
2037 /*
2038 * The first element of the structure is a
2039 * off_t that happen to have the same size for
2040 * all archs. Lets overwrite that.
2041 */
2042 off_t end_offset = 0;
2043 if (blob) {
2044 end_offset = blob->csb_end_offset;
2045 }
2046 error = copyout(&end_offset, argp, sizeof(end_offset));
2047 }
2048
2049 (void) vnode_put(vp);
2050 break;
2051 }
2052 case F_GETCODEDIR:
2053 case F_FINDSIGS: {
2054 error = ENOTSUP;
2055 goto out;
2056 }
2057 case F_CHECK_LV: {
2058 struct fileglob *fg;
2059 fchecklv_t lv = {};
2060
2061 if (fp->f_type != DTYPE_VNODE) {
2062 error = EBADF;
2063 goto out;
2064 }
2065 fg = fp->f_fglob;
2066 proc_fdunlock(p);
2067
2068 if (IS_64BIT_PROCESS(p)) {
2069 error = copyin(argp, &lv, sizeof(lv));
2070 } else {
2071 struct user32_fchecklv lv32 = {};
2072
2073 error = copyin(argp, &lv32, sizeof(lv32));
2074 lv.lv_file_start = lv32.lv_file_start;
2075 lv.lv_error_message = (void *)(uintptr_t)lv32.lv_error_message;
2076 lv.lv_error_message_size = lv32.lv_error_message_size;
2077 }
2078 if (error) {
2079 goto outdrop;
2080 }
2081
2082 #if CONFIG_MACF
2083 error = mac_file_check_library_validation(p, fg, lv.lv_file_start,
2084 (user_long_t)lv.lv_error_message, lv.lv_error_message_size);
2085 #endif
2086
2087 break;
2088 }
2089 #if CONFIG_PROTECT
2090 case F_GETPROTECTIONCLASS: {
2091 if (fp->f_type != DTYPE_VNODE) {
2092 error = EBADF;
2093 goto out;
2094 }
2095 vp = (struct vnode *)fp->f_data;
2096
2097 proc_fdunlock(p);
2098
2099 if (vnode_getwithref(vp)) {
2100 error = ENOENT;
2101 goto outdrop;
2102 }
2103
2104 struct vnode_attr va;
2105
2106 VATTR_INIT(&va);
2107 VATTR_WANTED(&va, va_dataprotect_class);
2108 error = VNOP_GETATTR(vp, &va, &context);
2109 if (!error) {
2110 if (VATTR_IS_SUPPORTED(&va, va_dataprotect_class)) {
2111 *retval = va.va_dataprotect_class;
2112 } else {
2113 error = ENOTSUP;
2114 }
2115 }
2116
2117 vnode_put(vp);
2118 break;
2119 }
2120
2121 case F_SETPROTECTIONCLASS: {
2122 /* tmp must be a valid PROTECTION_CLASS_* */
2123 tmp = CAST_DOWN_EXPLICIT(uint32_t, uap->arg);
2124
2125 if (fp->f_type != DTYPE_VNODE) {
2126 error = EBADF;
2127 goto out;
2128 }
2129 vp = (struct vnode *)fp->f_data;
2130
2131 proc_fdunlock(p);
2132
2133 if (vnode_getwithref(vp)) {
2134 error = ENOENT;
2135 goto outdrop;
2136 }
2137
2138 /* Only go forward if you have write access */
2139 vfs_context_t ctx = vfs_context_current();
2140 if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2141 vnode_put(vp);
2142 error = EBADF;
2143 goto outdrop;
2144 }
2145
2146 struct vnode_attr va;
2147
2148 VATTR_INIT(&va);
2149 VATTR_SET(&va, va_dataprotect_class, tmp);
2150
2151 error = VNOP_SETATTR(vp, &va, ctx);
2152
2153 vnode_put(vp);
2154 break;
2155 }
2156
2157 case F_TRANSCODEKEY: {
2158 if (fp->f_type != DTYPE_VNODE) {
2159 error = EBADF;
2160 goto out;
2161 }
2162
2163 vp = (struct vnode *)fp->f_data;
2164 proc_fdunlock(p);
2165
2166 if (vnode_getwithref(vp)) {
2167 error = ENOENT;
2168 goto outdrop;
2169 }
2170
2171 cp_key_t k = {
2172 .len = CP_MAX_WRAPPEDKEYSIZE,
2173 };
2174
2175 MALLOC(k.key, char *, k.len, M_TEMP, M_WAITOK | M_ZERO);
2176
2177 error = VNOP_IOCTL(vp, F_TRANSCODEKEY, (caddr_t)&k, 1, &context);
2178
2179 vnode_put(vp);
2180
2181 if (error == 0) {
2182 error = copyout(k.key, argp, k.len);
2183 *retval = k.len;
2184 }
2185
2186 FREE(k.key, M_TEMP);
2187
2188 break;
2189 }
2190
2191 case F_GETPROTECTIONLEVEL: {
2192 if (fp->f_type != DTYPE_VNODE) {
2193 error = EBADF;
2194 goto out;
2195 }
2196
2197 vp = (struct vnode*) fp->f_data;
2198 proc_fdunlock(p);
2199
2200 if (vnode_getwithref(vp)) {
2201 error = ENOENT;
2202 goto outdrop;
2203 }
2204
2205 error = VNOP_IOCTL(vp, F_GETPROTECTIONLEVEL, (caddr_t)retval, 0, &context);
2206
2207 vnode_put(vp);
2208 break;
2209 }
2210
2211 case F_GETDEFAULTPROTLEVEL: {
2212 if (fp->f_type != DTYPE_VNODE) {
2213 error = EBADF;
2214 goto out;
2215 }
2216
2217 vp = (struct vnode*) fp->f_data;
2218 proc_fdunlock(p);
2219
2220 if (vnode_getwithref(vp)) {
2221 error = ENOENT;
2222 goto outdrop;
2223 }
2224
2225 /*
2226 * if cp_get_major_vers fails, error will be set to proper errno
2227 * and cp_version will still be 0.
2228 */
2229
2230 error = VNOP_IOCTL(vp, F_GETDEFAULTPROTLEVEL, (caddr_t)retval, 0, &context);
2231
2232 vnode_put(vp);
2233 break;
2234 }
2235
2236 #endif /* CONFIG_PROTECT */
2237
2238 case F_MOVEDATAEXTENTS: {
2239 struct fileproc *fp2 = NULL;
2240 struct vnode *src_vp = NULLVP;
2241 struct vnode *dst_vp = NULLVP;
2242 /* We need to grab the 2nd FD out of the argments before moving on. */
2243 int fd2 = CAST_DOWN_EXPLICIT(int32_t, uap->arg);
2244
2245 error = priv_check_cred(kauth_cred_get(), PRIV_VFS_MOVE_DATA_EXTENTS, 0);
2246 if (error) {
2247 goto out;
2248 }
2249
2250 if (fp->f_type != DTYPE_VNODE) {
2251 error = EBADF;
2252 goto out;
2253 }
2254
2255 /*
2256 * For now, special case HFS+ and APFS only, since this
2257 * is SPI.
2258 */
2259 src_vp = (struct vnode *)fp->f_data;
2260 if (src_vp->v_tag != VT_HFS && src_vp->v_tag != VT_APFS) {
2261 error = ENOTSUP;
2262 goto out;
2263 }
2264
2265 /*
2266 * Get the references before we start acquiring iocounts on the vnodes,
2267 * while we still hold the proc fd lock
2268 */
2269 if ((error = fp_lookup(p, fd2, &fp2, 1))) {
2270 error = EBADF;
2271 goto out;
2272 }
2273 if (fp2->f_type != DTYPE_VNODE) {
2274 fp_drop(p, fd2, fp2, 1);
2275 error = EBADF;
2276 goto out;
2277 }
2278 dst_vp = (struct vnode *)fp2->f_data;
2279 if (dst_vp->v_tag != VT_HFS && dst_vp->v_tag != VT_APFS) {
2280 fp_drop(p, fd2, fp2, 1);
2281 error = ENOTSUP;
2282 goto out;
2283 }
2284
2285 #if CONFIG_MACF
2286 /* Re-do MAC checks against the new FD, pass in a fake argument */
2287 error = mac_file_check_fcntl(proc_ucred(p), fp2->f_fglob, uap->cmd, 0);
2288 if (error) {
2289 fp_drop(p, fd2, fp2, 1);
2290 goto out;
2291 }
2292 #endif
2293 /* Audit the 2nd FD */
2294 AUDIT_ARG(fd, fd2);
2295
2296 proc_fdunlock(p);
2297
2298 if (vnode_getwithref(src_vp)) {
2299 fp_drop(p, fd2, fp2, 0);
2300 error = ENOENT;
2301 goto outdrop;
2302 }
2303 if (vnode_getwithref(dst_vp)) {
2304 vnode_put(src_vp);
2305 fp_drop(p, fd2, fp2, 0);
2306 error = ENOENT;
2307 goto outdrop;
2308 }
2309
2310 /*
2311 * Basic asserts; validate they are not the same and that
2312 * both live on the same filesystem.
2313 */
2314 if (dst_vp == src_vp) {
2315 vnode_put(src_vp);
2316 vnode_put(dst_vp);
2317 fp_drop(p, fd2, fp2, 0);
2318 error = EINVAL;
2319 goto outdrop;
2320 }
2321
2322 if (dst_vp->v_mount != src_vp->v_mount) {
2323 vnode_put(src_vp);
2324 vnode_put(dst_vp);
2325 fp_drop(p, fd2, fp2, 0);
2326 error = EXDEV;
2327 goto outdrop;
2328 }
2329
2330 /* Now we have a legit pair of FDs. Go to work */
2331
2332 /* Now check for write access to the target files */
2333 if (vnode_authorize(src_vp, NULLVP,
2334 (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) {
2335 vnode_put(src_vp);
2336 vnode_put(dst_vp);
2337 fp_drop(p, fd2, fp2, 0);
2338 error = EBADF;
2339 goto outdrop;
2340 }
2341
2342 if (vnode_authorize(dst_vp, NULLVP,
2343 (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) {
2344 vnode_put(src_vp);
2345 vnode_put(dst_vp);
2346 fp_drop(p, fd2, fp2, 0);
2347 error = EBADF;
2348 goto outdrop;
2349 }
2350
2351 /* Verify that both vps point to files and not directories */
2352 if (!vnode_isreg(src_vp) || !vnode_isreg(dst_vp)) {
2353 error = EINVAL;
2354 vnode_put(src_vp);
2355 vnode_put(dst_vp);
2356 fp_drop(p, fd2, fp2, 0);
2357 goto outdrop;
2358 }
2359
2360 /*
2361 * The exchangedata syscall handler passes in 0 for the flags to VNOP_EXCHANGE.
2362 * We'll pass in our special bit indicating that the new behavior is expected
2363 */
2364
2365 error = VNOP_EXCHANGE(src_vp, dst_vp, FSOPT_EXCHANGE_DATA_ONLY, &context);
2366
2367 vnode_put(src_vp);
2368 vnode_put(dst_vp);
2369 fp_drop(p, fd2, fp2, 0);
2370 break;
2371 }
2372
2373 /*
2374 * SPI for making a file compressed.
2375 */
2376 case F_MAKECOMPRESSED: {
2377 uint32_t gcounter = CAST_DOWN_EXPLICIT(uint32_t, uap->arg);
2378
2379 if (fp->f_type != DTYPE_VNODE) {
2380 error = EBADF;
2381 goto out;
2382 }
2383
2384 vp = (struct vnode*) fp->f_data;
2385 proc_fdunlock(p);
2386
2387 /* get the vnode */
2388 if (vnode_getwithref(vp)) {
2389 error = ENOENT;
2390 goto outdrop;
2391 }
2392
2393 /* Is it a file? */
2394 if ((vnode_isreg(vp) == 0) && (vnode_islnk(vp) == 0)) {
2395 vnode_put(vp);
2396 error = EBADF;
2397 goto outdrop;
2398 }
2399
2400 /* invoke ioctl to pass off to FS */
2401 /* Only go forward if you have write access */
2402 vfs_context_t ctx = vfs_context_current();
2403 if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2404 vnode_put(vp);
2405 error = EBADF;
2406 goto outdrop;
2407 }
2408
2409 error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)&gcounter, 0, &context);
2410
2411 vnode_put(vp);
2412 break;
2413 }
2414
2415 /*
2416 * SPI (private) for indicating to a filesystem that subsequent writes to
2417 * the open FD will written to the Fastflow.
2418 */
2419 case F_SET_GREEDY_MODE:
2420 /* intentionally drop through to the same handler as F_SETSTATIC.
2421 * both fcntls should pass the argument and their selector into VNOP_IOCTL.
2422 */
2423
2424 /*
2425 * SPI (private) for indicating to a filesystem that subsequent writes to
2426 * the open FD will represent static content.
2427 */
2428 case F_SETSTATICCONTENT: {
2429 caddr_t ioctl_arg = NULL;
2430
2431 if (uap->arg) {
2432 ioctl_arg = (caddr_t) 1;
2433 }
2434
2435 if (fp->f_type != DTYPE_VNODE) {
2436 error = EBADF;
2437 goto out;
2438 }
2439 vp = (struct vnode *)fp->f_data;
2440 proc_fdunlock(p);
2441
2442 error = vnode_getwithref(vp);
2443 if (error) {
2444 error = ENOENT;
2445 goto outdrop;
2446 }
2447
2448 /* Only go forward if you have write access */
2449 vfs_context_t ctx = vfs_context_current();
2450 if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2451 vnode_put(vp);
2452 error = EBADF;
2453 goto outdrop;
2454 }
2455
2456 error = VNOP_IOCTL(vp, uap->cmd, ioctl_arg, 0, &context);
2457 (void)vnode_put(vp);
2458
2459 break;
2460 }
2461
2462 /*
2463 * SPI (private) for indicating to the lower level storage driver that the
2464 * subsequent writes should be of a particular IO type (burst, greedy, static),
2465 * or other flavors that may be necessary.
2466 */
2467 case F_SETIOTYPE: {
2468 caddr_t param_ptr;
2469 uint32_t param;
2470
2471 if (uap->arg) {
2472 /* extract 32 bits of flags from userland */
2473 param_ptr = (caddr_t) uap->arg;
2474 param = (uint32_t) param_ptr;
2475 } else {
2476 /* If no argument is specified, error out */
2477 error = EINVAL;
2478 goto out;
2479 }
2480
2481 /*
2482 * Validate the different types of flags that can be specified:
2483 * all of them are mutually exclusive for now.
2484 */
2485 switch (param) {
2486 case F_IOTYPE_ISOCHRONOUS:
2487 break;
2488
2489 default:
2490 error = EINVAL;
2491 goto out;
2492 }
2493
2494
2495 if (fp->f_type != DTYPE_VNODE) {
2496 error = EBADF;
2497 goto out;
2498 }
2499 vp = (struct vnode *)fp->f_data;
2500 proc_fdunlock(p);
2501
2502 error = vnode_getwithref(vp);
2503 if (error) {
2504 error = ENOENT;
2505 goto outdrop;
2506 }
2507
2508 /* Only go forward if you have write access */
2509 vfs_context_t ctx = vfs_context_current();
2510 if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2511 vnode_put(vp);
2512 error = EBADF;
2513 goto outdrop;
2514 }
2515
2516 error = VNOP_IOCTL(vp, uap->cmd, param_ptr, 0, &context);
2517 (void)vnode_put(vp);
2518
2519 break;
2520 }
2521
2522 /*
2523 * Set the vnode pointed to by 'fd'
2524 * and tag it as the (potentially future) backing store
2525 * for another filesystem
2526 */
2527 case F_SETBACKINGSTORE: {
2528 if (fp->f_type != DTYPE_VNODE) {
2529 error = EBADF;
2530 goto out;
2531 }
2532
2533 vp = (struct vnode *)fp->f_data;
2534
2535 if (vp->v_tag != VT_HFS) {
2536 error = EINVAL;
2537 goto out;
2538 }
2539 proc_fdunlock(p);
2540
2541 if (vnode_getwithref(vp)) {
2542 error = ENOENT;
2543 goto outdrop;
2544 }
2545
2546 /* only proceed if you have write access */
2547 vfs_context_t ctx = vfs_context_current();
2548 if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
2549 vnode_put(vp);
2550 error = EBADF;
2551 goto outdrop;
2552 }
2553
2554
2555 /* If arg != 0, set, otherwise unset */
2556 if (uap->arg) {
2557 error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)1, 0, &context);
2558 } else {
2559 error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)NULL, 0, &context);
2560 }
2561
2562 vnode_put(vp);
2563 break;
2564 }
2565
2566 /*
2567 * like F_GETPATH, but special semantics for
2568 * the mobile time machine handler.
2569 */
2570 case F_GETPATH_MTMINFO: {
2571 char *pathbufp;
2572 int pathlen;
2573
2574 if (fp->f_type != DTYPE_VNODE) {
2575 error = EBADF;
2576 goto out;
2577 }
2578 vp = (struct vnode *)fp->f_data;
2579 proc_fdunlock(p);
2580
2581 pathlen = MAXPATHLEN;
2582 MALLOC(pathbufp, char *, pathlen, M_TEMP, M_WAITOK);
2583 if (pathbufp == NULL) {
2584 error = ENOMEM;
2585 goto outdrop;
2586 }
2587 if ((error = vnode_getwithref(vp)) == 0) {
2588 int backingstore = 0;
2589
2590 /* Check for error from vn_getpath before moving on */
2591 if ((error = vn_getpath(vp, pathbufp, &pathlen)) == 0) {
2592 if (vp->v_tag == VT_HFS) {
2593 error = VNOP_IOCTL(vp, uap->cmd, (caddr_t) &backingstore, 0, &context);
2594 }
2595 (void)vnode_put(vp);
2596
2597 if (error == 0) {
2598 error = copyout((caddr_t)pathbufp, argp, pathlen);
2599 }
2600 if (error == 0) {
2601 /*
2602 * If the copyout was successful, now check to ensure
2603 * that this vnode is not a BACKINGSTORE vnode. mtmd
2604 * wants the path regardless.
2605 */
2606 if (backingstore) {
2607 error = EBUSY;
2608 }
2609 }
2610 } else {
2611 (void)vnode_put(vp);
2612 }
2613 }
2614 FREE(pathbufp, M_TEMP);
2615 goto outdrop;
2616 }
2617
2618 #if DEBUG || DEVELOPMENT
2619 case F_RECYCLE:
2620 if (fp->f_type != DTYPE_VNODE) {
2621 error = EBADF;
2622 goto out;
2623 }
2624 vp = (struct vnode *)fp->f_data;
2625 proc_fdunlock(p);
2626
2627 vnode_recycle(vp);
2628 break;
2629 #endif
2630
2631 default:
2632 /*
2633 * This is an fcntl() that we d not recognize at this level;
2634 * if this is a vnode, we send it down into the VNOP_IOCTL
2635 * for this vnode; this can include special devices, and will
2636 * effectively overload fcntl() to send ioctl()'s.
2637 */
2638 if ((uap->cmd & IOC_VOID) && (uap->cmd & IOC_INOUT)) {
2639 error = EINVAL;
2640 goto out;
2641 }
2642
2643 /* Catch any now-invalid fcntl() selectors */
2644 switch (uap->cmd) {
2645 case (int)APFSIOC_REVERT_TO_SNAPSHOT:
2646 case (int)FSIOC_FIOSEEKHOLE:
2647 case (int)FSIOC_FIOSEEKDATA:
2648 case (int)FSIOC_CAS_BSDFLAGS:
2649 case HFS_GET_BOOT_INFO:
2650 case HFS_SET_BOOT_INFO:
2651 case FIOPINSWAP:
2652 case F_MARKDEPENDENCY:
2653 case TIOCREVOKE:
2654 error = EINVAL;
2655 goto out;
2656 default:
2657 break;
2658 }
2659
2660 if (fp->f_type != DTYPE_VNODE) {
2661 error = EBADF;
2662 goto out;
2663 }
2664 vp = (struct vnode *)fp->f_data;
2665 proc_fdunlock(p);
2666
2667 if ((error = vnode_getwithref(vp)) == 0) {
2668 #define STK_PARAMS 128
2669 char stkbuf[STK_PARAMS] = {0};
2670 unsigned int size;
2671 caddr_t data, memp;
2672 /*
2673 * For this to work properly, we have to copy in the
2674 * ioctl() cmd argument if there is one; we must also
2675 * check that a command parameter, if present, does
2676 * not exceed the maximum command length dictated by
2677 * the number of bits we have available in the command
2678 * to represent a structure length. Finally, we have
2679 * to copy the results back out, if it is that type of
2680 * ioctl().
2681 */
2682 size = IOCPARM_LEN(uap->cmd);
2683 if (size > IOCPARM_MAX) {
2684 (void)vnode_put(vp);
2685 error = EINVAL;
2686 break;
2687 }
2688
2689 memp = NULL;
2690 if (size > sizeof(stkbuf)) {
2691 if ((memp = (caddr_t)kalloc(size)) == 0) {
2692 (void)vnode_put(vp);
2693 error = ENOMEM;
2694 goto outdrop;
2695 }
2696 data = memp;
2697 } else {
2698 data = &stkbuf[0];
2699 }
2700
2701 if (uap->cmd & IOC_IN) {
2702 if (size) {
2703 /* structure */
2704 error = copyin(argp, data, size);
2705 if (error) {
2706 (void)vnode_put(vp);
2707 if (memp) {
2708 kfree(memp, size);
2709 }
2710 goto outdrop;
2711 }
2712
2713 /* Bzero the section beyond that which was needed */
2714 if (size <= sizeof(stkbuf)) {
2715 bzero((((uint8_t*)data) + size), (sizeof(stkbuf) - size));
2716 }
2717 } else {
2718 /* int */
2719 if (is64bit) {
2720 *(user_addr_t *)data = argp;
2721 } else {
2722 *(uint32_t *)data = (uint32_t)argp;
2723 }
2724 };
2725 } else if ((uap->cmd & IOC_OUT) && size) {
2726 /*
2727 * Zero the buffer so the user always
2728 * gets back something deterministic.
2729 */
2730 bzero(data, size);
2731 } else if (uap->cmd & IOC_VOID) {
2732 if (is64bit) {
2733 *(user_addr_t *)data = argp;
2734 } else {
2735 *(uint32_t *)data = (uint32_t)argp;
2736 }
2737 }
2738
2739 error = VNOP_IOCTL(vp, uap->cmd, CAST_DOWN(caddr_t, data), 0, &context);
2740
2741 (void)vnode_put(vp);
2742
2743 /* Copy any output data to user */
2744 if (error == 0 && (uap->cmd & IOC_OUT) && size) {
2745 error = copyout(data, argp, size);
2746 }
2747 if (memp) {
2748 kfree(memp, size);
2749 }
2750 }
2751 break;
2752 }
2753
2754 outdrop:
2755 AUDIT_ARG(vnpath_withref, vp, ARG_VNODE1);
2756 fp_drop(p, fd, fp, 0);
2757 return error;
2758 out:
2759 fp_drop(p, fd, fp, 1);
2760 proc_fdunlock(p);
2761 return error;
2762 }
2763
2764
2765 /*
2766 * finishdup
2767 *
2768 * Description: Common code for dup, dup2, and fcntl(F_DUPFD).
2769 *
2770 * Parameters: p Process performing the dup
2771 * old The fd to dup
2772 * new The fd to dup it to
2773 * fd_flags Flags to augment the new fd
2774 * retval Pointer to the call return area
2775 *
2776 * Returns: 0 Success
2777 * EBADF
2778 * ENOMEM
2779 *
2780 * Implicit returns:
2781 * *retval (modified) The new descriptor
2782 *
2783 * Locks: Assumes proc_fdlock for process pointing to fdp is held by
2784 * the caller
2785 *
2786 * Notes: This function may drop and reacquire this lock; it is unsafe
2787 * for a caller to assume that other state protected by the lock
2788 * has not been subsequently changed out from under it.
2789 */
2790 int
2791 finishdup(proc_t p,
2792 struct filedesc *fdp, int old, int new, int fd_flags, int32_t *retval)
2793 {
2794 struct fileproc *nfp;
2795 struct fileproc *ofp;
2796 #if CONFIG_MACF
2797 int error;
2798 #endif
2799
2800 #if DIAGNOSTIC
2801 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
2802 #endif
2803 if ((ofp = fdp->fd_ofiles[old]) == NULL ||
2804 (fdp->fd_ofileflags[old] & UF_RESERVED)) {
2805 fdrelse(p, new);
2806 return EBADF;
2807 }
2808 fg_ref(ofp);
2809
2810 #if CONFIG_MACF
2811 error = mac_file_check_dup(proc_ucred(p), ofp->f_fglob, new);
2812 if (error) {
2813 fg_drop(ofp);
2814 fdrelse(p, new);
2815 return error;
2816 }
2817 #endif
2818
2819 proc_fdunlock(p);
2820
2821 nfp = fileproc_alloc_init(NULL);
2822
2823 proc_fdlock(p);
2824
2825 if (nfp == NULL) {
2826 fg_drop(ofp);
2827 fdrelse(p, new);
2828 return ENOMEM;
2829 }
2830
2831 nfp->f_fglob = ofp->f_fglob;
2832
2833 #if DIAGNOSTIC
2834 if (fdp->fd_ofiles[new] != 0) {
2835 panic("finishdup: overwriting fd_ofiles with new %d", new);
2836 }
2837 if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0) {
2838 panic("finishdup: unreserved fileflags with new %d", new);
2839 }
2840 #endif
2841
2842 if (new > fdp->fd_lastfile) {
2843 fdp->fd_lastfile = new;
2844 }
2845 *fdflags(p, new) |= fd_flags;
2846 procfdtbl_releasefd(p, new, nfp);
2847 *retval = new;
2848 return 0;
2849 }
2850
2851
2852 /*
2853 * close
2854 *
2855 * Description: The implementation of the close(2) system call
2856 *
2857 * Parameters: p Process in whose per process file table
2858 * the close is to occur
2859 * uap->fd fd to be closed
2860 * retval <unused>
2861 *
2862 * Returns: 0 Success
2863 * fp_lookup:EBADF Bad file descriptor
2864 * fp_guard_exception:??? Guarded file descriptor
2865 * close_internal:EBADF
2866 * close_internal:??? Anything returnable by a per-fileops
2867 * close function
2868 */
2869 int
2870 close(proc_t p, struct close_args *uap, int32_t *retval)
2871 {
2872 __pthread_testcancel(1);
2873 return close_nocancel(p, (struct close_nocancel_args *)uap, retval);
2874 }
2875
2876
2877 int
2878 close_nocancel(proc_t p, struct close_nocancel_args *uap, __unused int32_t *retval)
2879 {
2880 struct fileproc *fp;
2881 int fd = uap->fd;
2882 int error;
2883
2884 AUDIT_SYSCLOSE(p, fd);
2885
2886 proc_fdlock(p);
2887
2888 if ((error = fp_lookup(p, fd, &fp, 1))) {
2889 proc_fdunlock(p);
2890 return error;
2891 }
2892
2893 if (FP_ISGUARDED(fp, GUARD_CLOSE)) {
2894 error = fp_guard_exception(p, fd, fp, kGUARD_EXC_CLOSE);
2895 (void) fp_drop(p, fd, fp, 1);
2896 proc_fdunlock(p);
2897 return error;
2898 }
2899
2900 error = close_internal_locked(p, fd, fp, 0);
2901
2902 proc_fdunlock(p);
2903
2904 return error;
2905 }
2906
2907
2908 /*
2909 * close_internal_locked
2910 *
2911 * Close a file descriptor.
2912 *
2913 * Parameters: p Process in whose per process file table
2914 * the close is to occur
2915 * fd fd to be closed
2916 * fp fileproc associated with the fd
2917 *
2918 * Returns: 0 Success
2919 * EBADF fd already in close wait state
2920 * closef_locked:??? Anything returnable by a per-fileops
2921 * close function
2922 *
2923 * Locks: Assumes proc_fdlock for process is held by the caller and returns
2924 * with lock held
2925 *
2926 * Notes: This function may drop and reacquire this lock; it is unsafe
2927 * for a caller to assume that other state protected by the lock
2928 * has not been subsequently changed out from under it.
2929 */
2930 int
2931 close_internal_locked(proc_t p, int fd, struct fileproc *fp, int flags)
2932 {
2933 struct filedesc *fdp = p->p_fd;
2934 int error = 0;
2935 int resvfd = flags & FD_DUP2RESV;
2936
2937
2938 #if DIAGNOSTIC
2939 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
2940 #endif
2941
2942 /* Keep people from using the filedesc while we are closing it */
2943 procfdtbl_markclosefd(p, fd);
2944
2945
2946 if ((fp->f_flags & FP_CLOSING) == FP_CLOSING) {
2947 panic("close_internal_locked: being called on already closing fd");
2948 }
2949
2950
2951 #if DIAGNOSTIC
2952 if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0) {
2953 panic("close_internal: unreserved fileflags with fd %d", fd);
2954 }
2955 #endif
2956
2957 fp->f_flags |= FP_CLOSING;
2958
2959 if ((fp->f_flags & FP_AIOISSUED) || kauth_authorize_fileop_has_listeners()) {
2960 proc_fdunlock(p);
2961
2962 if ((fp->f_type == DTYPE_VNODE) && kauth_authorize_fileop_has_listeners()) {
2963 /*
2964 * call out to allow 3rd party notification of close.
2965 * Ignore result of kauth_authorize_fileop call.
2966 */
2967 if (vnode_getwithref((vnode_t)fp->f_data) == 0) {
2968 u_int fileop_flags = 0;
2969 if ((fp->f_flags & FP_WRITTEN) != 0) {
2970 fileop_flags |= KAUTH_FILEOP_CLOSE_MODIFIED;
2971 }
2972 kauth_authorize_fileop(fp->f_fglob->fg_cred, KAUTH_FILEOP_CLOSE,
2973 (uintptr_t)fp->f_data, (uintptr_t)fileop_flags);
2974 vnode_put((vnode_t)fp->f_data);
2975 }
2976 }
2977 if (fp->f_flags & FP_AIOISSUED) {
2978 /*
2979 * cancel all async IO requests that can be cancelled.
2980 */
2981 _aio_close( p, fd );
2982 }
2983
2984 proc_fdlock(p);
2985 }
2986
2987 if (fd < fdp->fd_knlistsize) {
2988 knote_fdclose(p, fd);
2989 }
2990
2991 /* release the ref returned from fp_lookup before calling drain */
2992 (void) os_ref_release_locked(&fp->f_iocount);
2993 fileproc_drain(p, fp);
2994
2995 if (fp->f_flags & FP_WAITEVENT) {
2996 (void)waitevent_close(p, fp);
2997 }
2998
2999 if (resvfd == 0) {
3000 _fdrelse(p, fd);
3001 } else {
3002 procfdtbl_reservefd(p, fd);
3003 }
3004
3005 if (ENTR_SHOULDTRACE && fp->f_type == DTYPE_SOCKET) {
3006 KERNEL_ENERGYTRACE(kEnTrActKernSocket, DBG_FUNC_END,
3007 fd, 0, (int64_t)VM_KERNEL_ADDRPERM(fp->f_data));
3008 }
3009
3010 error = closef_locked(fp, fp->f_fglob, p);
3011 if ((fp->f_flags & FP_WAITCLOSE) == FP_WAITCLOSE) {
3012 wakeup(&fp->f_flags);
3013 }
3014 fp->f_flags &= ~(FP_WAITCLOSE | FP_CLOSING);
3015
3016 proc_fdunlock(p);
3017
3018 fileproc_free(fp);
3019
3020 proc_fdlock(p);
3021
3022 #if DIAGNOSTIC
3023 if (resvfd != 0) {
3024 if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0) {
3025 panic("close with reserved fd returns with freed fd:%d: proc: %p", fd, p);
3026 }
3027 }
3028 #endif
3029
3030 return error;
3031 }
3032
3033
3034 /*
3035 * fstat1
3036 *
3037 * Description: Return status information about a file descriptor.
3038 *
3039 * Parameters: p The process doing the fstat
3040 * fd The fd to stat
3041 * ub The user stat buffer
3042 * xsecurity The user extended security
3043 * buffer, or 0 if none
3044 * xsecurity_size The size of xsecurity, or 0
3045 * if no xsecurity
3046 * isstat64 Flag to indicate 64 bit version
3047 * for inode size, etc.
3048 *
3049 * Returns: 0 Success
3050 * EBADF
3051 * EFAULT
3052 * fp_lookup:EBADF Bad file descriptor
3053 * vnode_getwithref:???
3054 * copyout:EFAULT
3055 * vnode_getwithref:???
3056 * vn_stat:???
3057 * soo_stat:???
3058 * pipe_stat:???
3059 * pshm_stat:???
3060 * kqueue_stat:???
3061 *
3062 * Notes: Internal implementation for all other fstat() related
3063 * functions
3064 *
3065 * XXX switch on node type is bogus; need a stat in struct
3066 * XXX fileops instead.
3067 */
3068 static int
3069 fstat1(proc_t p, int fd, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsecurity_size, int isstat64)
3070 {
3071 struct fileproc *fp;
3072 union {
3073 struct stat sb;
3074 struct stat64 sb64;
3075 } source;
3076 union {
3077 struct user64_stat user64_sb;
3078 struct user32_stat user32_sb;
3079 struct user64_stat64 user64_sb64;
3080 struct user32_stat64 user32_sb64;
3081 } dest;
3082 int error, my_size;
3083 file_type_t type;
3084 caddr_t data;
3085 kauth_filesec_t fsec;
3086 user_size_t xsecurity_bufsize;
3087 vfs_context_t ctx = vfs_context_current();
3088 void * sbptr;
3089
3090
3091 AUDIT_ARG(fd, fd);
3092
3093 if ((error = fp_lookup(p, fd, &fp, 0)) != 0) {
3094 return error;
3095 }
3096 type = fp->f_type;
3097 data = fp->f_data;
3098 fsec = KAUTH_FILESEC_NONE;
3099
3100 sbptr = (void *)&source;
3101
3102 switch (type) {
3103 case DTYPE_VNODE:
3104 if ((error = vnode_getwithref((vnode_t)data)) == 0) {
3105 /*
3106 * If the caller has the file open, and is not
3107 * requesting extended security information, we are
3108 * going to let them get the basic stat information.
3109 */
3110 if (xsecurity == USER_ADDR_NULL) {
3111 error = vn_stat_noauth((vnode_t)data, sbptr, NULL, isstat64, 0, ctx,
3112 fp->f_fglob->fg_cred);
3113 } else {
3114 error = vn_stat((vnode_t)data, sbptr, &fsec, isstat64, 0, ctx);
3115 }
3116
3117 AUDIT_ARG(vnpath, (struct vnode *)data, ARG_VNODE1);
3118 (void)vnode_put((vnode_t)data);
3119 }
3120 break;
3121
3122 #if SOCKETS
3123 case DTYPE_SOCKET:
3124 error = soo_stat((struct socket *)data, sbptr, isstat64);
3125 break;
3126 #endif /* SOCKETS */
3127
3128 case DTYPE_PIPE:
3129 error = pipe_stat((void *)data, sbptr, isstat64);
3130 break;
3131
3132 case DTYPE_PSXSHM:
3133 error = pshm_stat((void *)data, sbptr, isstat64);
3134 break;
3135
3136 case DTYPE_KQUEUE:
3137 error = kqueue_stat((void *)data, sbptr, isstat64, p);
3138 break;
3139
3140 default:
3141 error = EBADF;
3142 goto out;
3143 }
3144 if (error == 0) {
3145 caddr_t sbp;
3146
3147 if (isstat64 != 0) {
3148 source.sb64.st_lspare = 0;
3149 source.sb64.st_qspare[0] = 0LL;
3150 source.sb64.st_qspare[1] = 0LL;
3151
3152 if (IS_64BIT_PROCESS(current_proc())) {
3153 munge_user64_stat64(&source.sb64, &dest.user64_sb64);
3154 my_size = sizeof(dest.user64_sb64);
3155 sbp = (caddr_t)&dest.user64_sb64;
3156 } else {
3157 munge_user32_stat64(&source.sb64, &dest.user32_sb64);
3158 my_size = sizeof(dest.user32_sb64);
3159 sbp = (caddr_t)&dest.user32_sb64;
3160 }
3161 } else {
3162 source.sb.st_lspare = 0;
3163 source.sb.st_qspare[0] = 0LL;
3164 source.sb.st_qspare[1] = 0LL;
3165 if (IS_64BIT_PROCESS(current_proc())) {
3166 munge_user64_stat(&source.sb, &dest.user64_sb);
3167 my_size = sizeof(dest.user64_sb);
3168 sbp = (caddr_t)&dest.user64_sb;
3169 } else {
3170 munge_user32_stat(&source.sb, &dest.user32_sb);
3171 my_size = sizeof(dest.user32_sb);
3172 sbp = (caddr_t)&dest.user32_sb;
3173 }
3174 }
3175
3176 error = copyout(sbp, ub, my_size);
3177 }
3178
3179 /* caller wants extended security information? */
3180 if (xsecurity != USER_ADDR_NULL) {
3181 /* did we get any? */
3182 if (fsec == KAUTH_FILESEC_NONE) {
3183 if (susize(xsecurity_size, 0) != 0) {
3184 error = EFAULT;
3185 goto out;
3186 }
3187 } else {
3188 /* find the user buffer size */
3189 xsecurity_bufsize = fusize(xsecurity_size);
3190
3191 /* copy out the actual data size */
3192 if (susize(xsecurity_size, KAUTH_FILESEC_COPYSIZE(fsec)) != 0) {
3193 error = EFAULT;
3194 goto out;
3195 }
3196
3197 /* if the caller supplied enough room, copy out to it */
3198 if (xsecurity_bufsize >= KAUTH_FILESEC_COPYSIZE(fsec)) {
3199 error = copyout(fsec, xsecurity, KAUTH_FILESEC_COPYSIZE(fsec));
3200 }
3201 }
3202 }
3203 out:
3204 fp_drop(p, fd, fp, 0);
3205 if (fsec != NULL) {
3206 kauth_filesec_free(fsec);
3207 }
3208 return error;
3209 }
3210
3211
3212 /*
3213 * fstat_extended
3214 *
3215 * Description: Extended version of fstat supporting returning extended
3216 * security information
3217 *
3218 * Parameters: p The process doing the fstat
3219 * uap->fd The fd to stat
3220 * uap->ub The user stat buffer
3221 * uap->xsecurity The user extended security
3222 * buffer, or 0 if none
3223 * uap->xsecurity_size The size of xsecurity, or 0
3224 *
3225 * Returns: 0 Success
3226 * !0 Errno (see fstat1)
3227 */
3228 int
3229 fstat_extended(proc_t p, struct fstat_extended_args *uap, __unused int32_t *retval)
3230 {
3231 return fstat1(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 0);
3232 }
3233
3234
3235 /*
3236 * fstat
3237 *
3238 * Description: Get file status for the file associated with fd
3239 *
3240 * Parameters: p The process doing the fstat
3241 * uap->fd The fd to stat
3242 * uap->ub The user stat buffer
3243 *
3244 * Returns: 0 Success
3245 * !0 Errno (see fstat1)
3246 */
3247 int
3248 fstat(proc_t p, struct fstat_args *uap, __unused int32_t *retval)
3249 {
3250 return fstat1(p, uap->fd, uap->ub, 0, 0, 0);
3251 }
3252
3253
3254 /*
3255 * fstat64_extended
3256 *
3257 * Description: Extended version of fstat64 supporting returning extended
3258 * security information
3259 *
3260 * Parameters: p The process doing the fstat
3261 * uap->fd The fd to stat
3262 * uap->ub The user stat buffer
3263 * uap->xsecurity The user extended security
3264 * buffer, or 0 if none
3265 * uap->xsecurity_size The size of xsecurity, or 0
3266 *
3267 * Returns: 0 Success
3268 * !0 Errno (see fstat1)
3269 */
3270 int
3271 fstat64_extended(proc_t p, struct fstat64_extended_args *uap, __unused int32_t *retval)
3272 {
3273 return fstat1(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 1);
3274 }
3275
3276
3277 /*
3278 * fstat64
3279 *
3280 * Description: Get 64 bit version of the file status for the file associated
3281 * with fd
3282 *
3283 * Parameters: p The process doing the fstat
3284 * uap->fd The fd to stat
3285 * uap->ub The user stat buffer
3286 *
3287 * Returns: 0 Success
3288 * !0 Errno (see fstat1)
3289 */
3290 int
3291 fstat64(proc_t p, struct fstat64_args *uap, __unused int32_t *retval)
3292 {
3293 return fstat1(p, uap->fd, uap->ub, 0, 0, 1);
3294 }
3295
3296
3297 /*
3298 * fpathconf
3299 *
3300 * Description: Return pathconf information about a file descriptor.
3301 *
3302 * Parameters: p Process making the request
3303 * uap->fd fd to get information about
3304 * uap->name Name of information desired
3305 * retval Pointer to the call return area
3306 *
3307 * Returns: 0 Success
3308 * EINVAL
3309 * fp_lookup:EBADF Bad file descriptor
3310 * vnode_getwithref:???
3311 * vn_pathconf:???
3312 *
3313 * Implicit returns:
3314 * *retval (modified) Returned information (numeric)
3315 */
3316 int
3317 fpathconf(proc_t p, struct fpathconf_args *uap, int32_t *retval)
3318 {
3319 int fd = uap->fd;
3320 struct fileproc *fp;
3321 struct vnode *vp;
3322 int error = 0;
3323 file_type_t type;
3324 caddr_t data;
3325
3326
3327 AUDIT_ARG(fd, uap->fd);
3328 if ((error = fp_lookup(p, fd, &fp, 0))) {
3329 return error;
3330 }
3331 type = fp->f_type;
3332 data = fp->f_data;
3333
3334 switch (type) {
3335 case DTYPE_SOCKET:
3336 if (uap->name != _PC_PIPE_BUF) {
3337 error = EINVAL;
3338 goto out;
3339 }
3340 *retval = PIPE_BUF;
3341 error = 0;
3342 goto out;
3343
3344 case DTYPE_PIPE:
3345 if (uap->name != _PC_PIPE_BUF) {
3346 error = EINVAL;
3347 goto out;
3348 }
3349 *retval = PIPE_BUF;
3350 error = 0;
3351 goto out;
3352
3353 case DTYPE_VNODE:
3354 vp = (struct vnode *)data;
3355
3356 if ((error = vnode_getwithref(vp)) == 0) {
3357 AUDIT_ARG(vnpath, vp, ARG_VNODE1);
3358
3359 error = vn_pathconf(vp, uap->name, retval, vfs_context_current());
3360
3361 (void)vnode_put(vp);
3362 }
3363 goto out;
3364
3365 default:
3366 error = EINVAL;
3367 goto out;
3368 }
3369 /*NOTREACHED*/
3370 out:
3371 fp_drop(p, fd, fp, 0);
3372 return error;
3373 }
3374
3375 /*
3376 * Statistics counter for the number of times a process calling fdalloc()
3377 * has resulted in an expansion of the per process open file table.
3378 *
3379 * XXX This would likely be of more use if it were per process
3380 */
3381 int fdexpand;
3382
3383
3384 /*
3385 * fdalloc
3386 *
3387 * Description: Allocate a file descriptor for the process.
3388 *
3389 * Parameters: p Process to allocate the fd in
3390 * want The fd we would prefer to get
3391 * result Pointer to fd we got
3392 *
3393 * Returns: 0 Success
3394 * EMFILE
3395 * ENOMEM
3396 *
3397 * Implicit returns:
3398 * *result (modified) The fd which was allocated
3399 */
3400 int
3401 fdalloc(proc_t p, int want, int *result)
3402 {
3403 struct filedesc *fdp = p->p_fd;
3404 int i;
3405 int lim, last, numfiles, oldnfiles;
3406 struct fileproc **newofiles, **ofiles;
3407 char *newofileflags;
3408
3409 /*
3410 * Search for a free descriptor starting at the higher
3411 * of want or fd_freefile. If that fails, consider
3412 * expanding the ofile array.
3413 */
3414 #if DIAGNOSTIC
3415 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
3416 #endif
3417
3418 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
3419 for (;;) {
3420 last = min(fdp->fd_nfiles, lim);
3421 if ((i = want) < fdp->fd_freefile) {
3422 i = fdp->fd_freefile;
3423 }
3424 for (; i < last; i++) {
3425 if (fdp->fd_ofiles[i] == NULL && !(fdp->fd_ofileflags[i] & UF_RESERVED)) {
3426 procfdtbl_reservefd(p, i);
3427 if (i > fdp->fd_lastfile) {
3428 fdp->fd_lastfile = i;
3429 }
3430 if (want <= fdp->fd_freefile) {
3431 fdp->fd_freefile = i;
3432 }
3433 *result = i;
3434 return 0;
3435 }
3436 }
3437
3438 /*
3439 * No space in current array. Expand?
3440 */
3441 if (fdp->fd_nfiles >= lim) {
3442 return EMFILE;
3443 }
3444 if (fdp->fd_nfiles < NDEXTENT) {
3445 numfiles = NDEXTENT;
3446 } else {
3447 numfiles = 2 * fdp->fd_nfiles;
3448 }
3449 /* Enforce lim */
3450 if (numfiles > lim) {
3451 numfiles = lim;
3452 }
3453 proc_fdunlock(p);
3454 MALLOC_ZONE(newofiles, struct fileproc **,
3455 numfiles * OFILESIZE, M_OFILETABL, M_WAITOK);
3456 proc_fdlock(p);
3457 if (newofiles == NULL) {
3458 return ENOMEM;
3459 }
3460 if (fdp->fd_nfiles >= numfiles) {
3461 FREE_ZONE(newofiles, numfiles * OFILESIZE, M_OFILETABL);
3462 continue;
3463 }
3464 newofileflags = (char *) &newofiles[numfiles];
3465 /*
3466 * Copy the existing ofile and ofileflags arrays
3467 * and zero the new portion of each array.
3468 */
3469 oldnfiles = fdp->fd_nfiles;
3470 (void) memcpy(newofiles, fdp->fd_ofiles,
3471 oldnfiles * sizeof(*fdp->fd_ofiles));
3472 (void) memset(&newofiles[oldnfiles], 0,
3473 (numfiles - oldnfiles) * sizeof(*fdp->fd_ofiles));
3474
3475 (void) memcpy(newofileflags, fdp->fd_ofileflags,
3476 oldnfiles * sizeof(*fdp->fd_ofileflags));
3477 (void) memset(&newofileflags[oldnfiles], 0,
3478 (numfiles - oldnfiles) *
3479 sizeof(*fdp->fd_ofileflags));
3480 ofiles = fdp->fd_ofiles;
3481 fdp->fd_ofiles = newofiles;
3482 fdp->fd_ofileflags = newofileflags;
3483 fdp->fd_nfiles = numfiles;
3484 FREE_ZONE(ofiles, oldnfiles * OFILESIZE, M_OFILETABL);
3485 fdexpand++;
3486 }
3487 }
3488
3489
3490 /*
3491 * fdavail
3492 *
3493 * Description: Check to see whether n user file descriptors are available
3494 * to the process p.
3495 *
3496 * Parameters: p Process to check in
3497 * n The number of fd's desired
3498 *
3499 * Returns: 0 No
3500 * 1 Yes
3501 *
3502 * Locks: Assumes proc_fdlock for process is held by the caller
3503 *
3504 * Notes: The answer only remains valid so long as the proc_fdlock is
3505 * held by the caller.
3506 */
3507 int
3508 fdavail(proc_t p, int n)
3509 {
3510 struct filedesc *fdp = p->p_fd;
3511 struct fileproc **fpp;
3512 char *flags;
3513 int i, lim;
3514
3515 lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
3516 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
3517 return 1;
3518 }
3519 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
3520 flags = &fdp->fd_ofileflags[fdp->fd_freefile];
3521 for (i = fdp->fd_nfiles - fdp->fd_freefile; --i >= 0; fpp++, flags++) {
3522 if (*fpp == NULL && !(*flags & UF_RESERVED) && --n <= 0) {
3523 return 1;
3524 }
3525 }
3526 return 0;
3527 }
3528
3529
3530 /*
3531 * fdrelse
3532 *
3533 * Description: Legacy KPI wrapper function for _fdrelse
3534 *
3535 * Parameters: p Process in which fd lives
3536 * fd fd to free
3537 *
3538 * Returns: void
3539 *
3540 * Locks: Assumes proc_fdlock for process is held by the caller
3541 */
3542 void
3543 fdrelse(proc_t p, int fd)
3544 {
3545 _fdrelse(p, fd);
3546 }
3547
3548
3549 /*
3550 * fdgetf_noref
3551 *
3552 * Description: Get the fileproc pointer for the given fd from the per process
3553 * open file table without taking an explicit reference on it.
3554 *
3555 * Parameters: p Process containing fd
3556 * fd fd to obtain fileproc for
3557 * resultfp Pointer to pointer return area
3558 *
3559 * Returns: 0 Success
3560 * EBADF
3561 *
3562 * Implicit returns:
3563 * *resultfp (modified) Pointer to fileproc pointer
3564 *
3565 * Locks: Assumes proc_fdlock for process is held by the caller
3566 *
3567 * Notes: Because there is no reference explicitly taken, the returned
3568 * fileproc pointer is only valid so long as the proc_fdlock
3569 * remains held by the caller.
3570 */
3571 int
3572 fdgetf_noref(proc_t p, int fd, struct fileproc **resultfp)
3573 {
3574 struct filedesc *fdp = p->p_fd;
3575 struct fileproc *fp;
3576
3577 if (fd < 0 || fd >= fdp->fd_nfiles ||
3578 (fp = fdp->fd_ofiles[fd]) == NULL ||
3579 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3580 return EBADF;
3581 }
3582 if (resultfp) {
3583 *resultfp = fp;
3584 }
3585 return 0;
3586 }
3587
3588
3589 /*
3590 * fp_getfvp
3591 *
3592 * Description: Get fileproc and vnode pointer for a given fd from the per
3593 * process open file table of the specified process, and if
3594 * successful, increment the f_iocount
3595 *
3596 * Parameters: p Process in which fd lives
3597 * fd fd to get information for
3598 * resultfp Pointer to result fileproc
3599 * pointer area, or 0 if none
3600 * resultvp Pointer to result vnode pointer
3601 * area, or 0 if none
3602 *
3603 * Returns: 0 Success
3604 * EBADF Bad file descriptor
3605 * ENOTSUP fd does not refer to a vnode
3606 *
3607 * Implicit returns:
3608 * *resultfp (modified) Fileproc pointer
3609 * *resultvp (modified) vnode pointer
3610 *
3611 * Notes: The resultfp and resultvp fields are optional, and may be
3612 * independently specified as NULL to skip returning information
3613 *
3614 * Locks: Internally takes and releases proc_fdlock
3615 */
3616 int
3617 fp_getfvp(proc_t p, int fd, struct fileproc **resultfp, struct vnode **resultvp)
3618 {
3619 struct filedesc *fdp = p->p_fd;
3620 struct fileproc *fp;
3621
3622 proc_fdlock_spin(p);
3623 if (fd < 0 || fd >= fdp->fd_nfiles ||
3624 (fp = fdp->fd_ofiles[fd]) == NULL ||
3625 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3626 proc_fdunlock(p);
3627 return EBADF;
3628 }
3629 if (fp->f_type != DTYPE_VNODE) {
3630 proc_fdunlock(p);
3631 return ENOTSUP;
3632 }
3633 os_ref_retain_locked(&fp->f_iocount);
3634
3635 if (resultfp) {
3636 *resultfp = fp;
3637 }
3638 if (resultvp) {
3639 *resultvp = (struct vnode *)fp->f_data;
3640 }
3641 proc_fdunlock(p);
3642
3643 return 0;
3644 }
3645
3646
3647 /*
3648 * fp_getfvpandvid
3649 *
3650 * Description: Get fileproc, vnode pointer, and vid for a given fd from the
3651 * per process open file table of the specified process, and if
3652 * successful, increment the f_iocount
3653 *
3654 * Parameters: p Process in which fd lives
3655 * fd fd to get information for
3656 * resultfp Pointer to result fileproc
3657 * pointer area, or 0 if none
3658 * resultvp Pointer to result vnode pointer
3659 * area, or 0 if none
3660 * vidp Pointer to resuld vid area
3661 *
3662 * Returns: 0 Success
3663 * EBADF Bad file descriptor
3664 * ENOTSUP fd does not refer to a vnode
3665 *
3666 * Implicit returns:
3667 * *resultfp (modified) Fileproc pointer
3668 * *resultvp (modified) vnode pointer
3669 * *vidp vid value
3670 *
3671 * Notes: The resultfp and resultvp fields are optional, and may be
3672 * independently specified as NULL to skip returning information
3673 *
3674 * Locks: Internally takes and releases proc_fdlock
3675 */
3676 int
3677 fp_getfvpandvid(proc_t p, int fd, struct fileproc **resultfp,
3678 struct vnode **resultvp, uint32_t *vidp)
3679 {
3680 struct filedesc *fdp = p->p_fd;
3681 struct fileproc *fp;
3682
3683 proc_fdlock_spin(p);
3684 if (fd < 0 || fd >= fdp->fd_nfiles ||
3685 (fp = fdp->fd_ofiles[fd]) == NULL ||
3686 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3687 proc_fdunlock(p);
3688 return EBADF;
3689 }
3690 if (fp->f_type != DTYPE_VNODE) {
3691 proc_fdunlock(p);
3692 return ENOTSUP;
3693 }
3694 os_ref_retain_locked(&fp->f_iocount);
3695
3696 if (resultfp) {
3697 *resultfp = fp;
3698 }
3699 if (resultvp) {
3700 *resultvp = (struct vnode *)fp->f_data;
3701 }
3702 if (vidp) {
3703 *vidp = (uint32_t)vnode_vid((struct vnode *)fp->f_data);
3704 }
3705 proc_fdunlock(p);
3706
3707 return 0;
3708 }
3709
3710
3711 /*
3712 * fp_getfsock
3713 *
3714 * Description: Get fileproc and socket pointer for a given fd from the
3715 * per process open file table of the specified process, and if
3716 * successful, increment the f_iocount
3717 *
3718 * Parameters: p Process in which fd lives
3719 * fd fd to get information for
3720 * resultfp Pointer to result fileproc
3721 * pointer area, or 0 if none
3722 * results Pointer to result socket
3723 * pointer area, or 0 if none
3724 *
3725 * Returns: EBADF The file descriptor is invalid
3726 * EOPNOTSUPP The file descriptor is not a socket
3727 * 0 Success
3728 *
3729 * Implicit returns:
3730 * *resultfp (modified) Fileproc pointer
3731 * *results (modified) socket pointer
3732 *
3733 * Notes: EOPNOTSUPP should probably be ENOTSOCK; this function is only
3734 * ever called from accept1().
3735 */
3736 int
3737 fp_getfsock(proc_t p, int fd, struct fileproc **resultfp,
3738 struct socket **results)
3739 {
3740 struct filedesc *fdp = p->p_fd;
3741 struct fileproc *fp;
3742
3743 proc_fdlock_spin(p);
3744 if (fd < 0 || fd >= fdp->fd_nfiles ||
3745 (fp = fdp->fd_ofiles[fd]) == NULL ||
3746 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3747 proc_fdunlock(p);
3748 return EBADF;
3749 }
3750 if (fp->f_type != DTYPE_SOCKET) {
3751 proc_fdunlock(p);
3752 return EOPNOTSUPP;
3753 }
3754 os_ref_retain_locked(&fp->f_iocount);
3755
3756 if (resultfp) {
3757 *resultfp = fp;
3758 }
3759 if (results) {
3760 *results = (struct socket *)fp->f_data;
3761 }
3762 proc_fdunlock(p);
3763
3764 return 0;
3765 }
3766
3767
3768 /*
3769 * fp_getfkq
3770 *
3771 * Description: Get fileproc and kqueue pointer for a given fd from the
3772 * per process open file table of the specified process, and if
3773 * successful, increment the f_iocount
3774 *
3775 * Parameters: p Process in which fd lives
3776 * fd fd to get information for
3777 * resultfp Pointer to result fileproc
3778 * pointer area, or 0 if none
3779 * resultkq Pointer to result kqueue
3780 * pointer area, or 0 if none
3781 *
3782 * Returns: EBADF The file descriptor is invalid
3783 * EBADF The file descriptor is not a socket
3784 * 0 Success
3785 *
3786 * Implicit returns:
3787 * *resultfp (modified) Fileproc pointer
3788 * *resultkq (modified) kqueue pointer
3789 *
3790 * Notes: The second EBADF should probably be something else to make
3791 * the error condition distinct.
3792 */
3793 int
3794 fp_getfkq(proc_t p, int fd, struct fileproc **resultfp,
3795 struct kqueue **resultkq)
3796 {
3797 struct filedesc *fdp = p->p_fd;
3798 struct fileproc *fp;
3799
3800 proc_fdlock_spin(p);
3801 if (fd < 0 || fd >= fdp->fd_nfiles ||
3802 (fp = fdp->fd_ofiles[fd]) == NULL ||
3803 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3804 proc_fdunlock(p);
3805 return EBADF;
3806 }
3807 if (fp->f_type != DTYPE_KQUEUE) {
3808 proc_fdunlock(p);
3809 return EBADF;
3810 }
3811 os_ref_retain_locked(&fp->f_iocount);
3812
3813 if (resultfp) {
3814 *resultfp = fp;
3815 }
3816 if (resultkq) {
3817 *resultkq = (struct kqueue *)fp->f_data;
3818 }
3819 proc_fdunlock(p);
3820
3821 return 0;
3822 }
3823
3824
3825 /*
3826 * fp_getfpshm
3827 *
3828 * Description: Get fileproc and POSIX shared memory pointer for a given fd
3829 * from the per process open file table of the specified process
3830 * and if successful, increment the f_iocount
3831 *
3832 * Parameters: p Process in which fd lives
3833 * fd fd to get information for
3834 * resultfp Pointer to result fileproc
3835 * pointer area, or 0 if none
3836 * resultpshm Pointer to result POSIX
3837 * shared memory pointer
3838 * pointer area, or 0 if none
3839 *
3840 * Returns: EBADF The file descriptor is invalid
3841 * EBADF The file descriptor is not a POSIX
3842 * shared memory area
3843 * 0 Success
3844 *
3845 * Implicit returns:
3846 * *resultfp (modified) Fileproc pointer
3847 * *resultpshm (modified) POSIX shared memory pointer
3848 *
3849 * Notes: The second EBADF should probably be something else to make
3850 * the error condition distinct.
3851 */
3852 int
3853 fp_getfpshm(proc_t p, int fd, struct fileproc **resultfp,
3854 struct pshmnode **resultpshm)
3855 {
3856 struct filedesc *fdp = p->p_fd;
3857 struct fileproc *fp;
3858
3859 proc_fdlock_spin(p);
3860 if (fd < 0 || fd >= fdp->fd_nfiles ||
3861 (fp = fdp->fd_ofiles[fd]) == NULL ||
3862 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3863 proc_fdunlock(p);
3864 return EBADF;
3865 }
3866 if (fp->f_type != DTYPE_PSXSHM) {
3867 proc_fdunlock(p);
3868 return EBADF;
3869 }
3870 os_ref_retain_locked(&fp->f_iocount);
3871
3872 if (resultfp) {
3873 *resultfp = fp;
3874 }
3875 if (resultpshm) {
3876 *resultpshm = (struct pshmnode *)fp->f_data;
3877 }
3878 proc_fdunlock(p);
3879
3880 return 0;
3881 }
3882
3883
3884 /*
3885 * fp_getfsem
3886 *
3887 * Description: Get fileproc and POSIX semaphore pointer for a given fd from
3888 * the per process open file table of the specified process
3889 * and if successful, increment the f_iocount
3890 *
3891 * Parameters: p Process in which fd lives
3892 * fd fd to get information for
3893 * resultfp Pointer to result fileproc
3894 * pointer area, or 0 if none
3895 * resultpsem Pointer to result POSIX
3896 * semaphore pointer area, or
3897 * 0 if none
3898 *
3899 * Returns: EBADF The file descriptor is invalid
3900 * EBADF The file descriptor is not a POSIX
3901 * semaphore
3902 * 0 Success
3903 *
3904 * Implicit returns:
3905 * *resultfp (modified) Fileproc pointer
3906 * *resultpsem (modified) POSIX semaphore pointer
3907 *
3908 * Notes: The second EBADF should probably be something else to make
3909 * the error condition distinct.
3910 *
3911 * In order to support unnamed POSIX semaphores, the named
3912 * POSIX semaphores will have to move out of the per-process
3913 * open filetable, and into a global table that is shared with
3914 * unnamed POSIX semaphores, since unnamed POSIX semaphores
3915 * are typically used by declaring instances in shared memory,
3916 * and there's no other way to do this without changing the
3917 * underlying type, which would introduce binary compatibility
3918 * issues.
3919 */
3920 int
3921 fp_getfpsem(proc_t p, int fd, struct fileproc **resultfp,
3922 struct psemnode **resultpsem)
3923 {
3924 struct filedesc *fdp = p->p_fd;
3925 struct fileproc *fp;
3926
3927 proc_fdlock_spin(p);
3928 if (fd < 0 || fd >= fdp->fd_nfiles ||
3929 (fp = fdp->fd_ofiles[fd]) == NULL ||
3930 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3931 proc_fdunlock(p);
3932 return EBADF;
3933 }
3934 if (fp->f_type != DTYPE_PSXSEM) {
3935 proc_fdunlock(p);
3936 return EBADF;
3937 }
3938 os_ref_retain_locked(&fp->f_iocount);
3939
3940 if (resultfp) {
3941 *resultfp = fp;
3942 }
3943 if (resultpsem) {
3944 *resultpsem = (struct psemnode *)fp->f_data;
3945 }
3946 proc_fdunlock(p);
3947
3948 return 0;
3949 }
3950
3951
3952 /*
3953 * fp_getfpipe
3954 *
3955 * Description: Get fileproc and pipe pointer for a given fd from the
3956 * per process open file table of the specified process
3957 * and if successful, increment the f_iocount
3958 *
3959 * Parameters: p Process in which fd lives
3960 * fd fd to get information for
3961 * resultfp Pointer to result fileproc
3962 * pointer area, or 0 if none
3963 * resultpipe Pointer to result pipe
3964 * pointer area, or 0 if none
3965 *
3966 * Returns: EBADF The file descriptor is invalid
3967 * EBADF The file descriptor is not a socket
3968 * 0 Success
3969 *
3970 * Implicit returns:
3971 * *resultfp (modified) Fileproc pointer
3972 * *resultpipe (modified) pipe pointer
3973 *
3974 * Notes: The second EBADF should probably be something else to make
3975 * the error condition distinct.
3976 */
3977 int
3978 fp_getfpipe(proc_t p, int fd, struct fileproc **resultfp,
3979 struct pipe **resultpipe)
3980 {
3981 struct filedesc *fdp = p->p_fd;
3982 struct fileproc *fp;
3983
3984 proc_fdlock_spin(p);
3985 if (fd < 0 || fd >= fdp->fd_nfiles ||
3986 (fp = fdp->fd_ofiles[fd]) == NULL ||
3987 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
3988 proc_fdunlock(p);
3989 return EBADF;
3990 }
3991 if (fp->f_type != DTYPE_PIPE) {
3992 proc_fdunlock(p);
3993 return EBADF;
3994 }
3995 os_ref_retain_locked(&fp->f_iocount);
3996
3997 if (resultfp) {
3998 *resultfp = fp;
3999 }
4000 if (resultpipe) {
4001 *resultpipe = (struct pipe *)fp->f_data;
4002 }
4003 proc_fdunlock(p);
4004
4005 return 0;
4006 }
4007
4008
4009 /*
4010 * fp_lookup
4011 *
4012 * Description: Get fileproc pointer for a given fd from the per process
4013 * open file table of the specified process and if successful,
4014 * increment the f_iocount
4015 *
4016 * Parameters: p Process in which fd lives
4017 * fd fd to get information for
4018 * resultfp Pointer to result fileproc
4019 * pointer area, or 0 if none
4020 * locked !0 if the caller holds the
4021 * proc_fdlock, 0 otherwise
4022 *
4023 * Returns: 0 Success
4024 * EBADF Bad file descriptor
4025 *
4026 * Implicit returns:
4027 * *resultfp (modified) Fileproc pointer
4028 *
4029 * Locks: If the argument 'locked' is non-zero, then the caller is
4030 * expected to have taken and held the proc_fdlock; if it is
4031 * zero, than this routine internally takes and drops this lock.
4032 */
4033 int
4034 fp_lookup(proc_t p, int fd, struct fileproc **resultfp, int locked)
4035 {
4036 struct filedesc *fdp = p->p_fd;
4037 struct fileproc *fp;
4038
4039 if (!locked) {
4040 proc_fdlock_spin(p);
4041 }
4042 if (fd < 0 || fdp == NULL || fd >= fdp->fd_nfiles ||
4043 (fp = fdp->fd_ofiles[fd]) == NULL ||
4044 (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
4045 if (!locked) {
4046 proc_fdunlock(p);
4047 }
4048 return EBADF;
4049 }
4050 os_ref_retain_locked(&fp->f_iocount);
4051
4052 if (resultfp) {
4053 *resultfp = fp;
4054 }
4055 if (!locked) {
4056 proc_fdunlock(p);
4057 }
4058
4059 return 0;
4060 }
4061
4062
4063 /*
4064 * fp_tryswap
4065 *
4066 * Description: Swap the fileproc pointer for a given fd with a new
4067 * fileproc pointer in the per-process open file table of
4068 * the specified process. The fdlock must be held at entry.
4069 * Iff the swap is successful, the old fileproc pointer is freed.
4070 *
4071 * Parameters: p Process containing the fd
4072 * fd The fd of interest
4073 * nfp Pointer to the newfp
4074 *
4075 * Returns: 0 Success
4076 * EBADF Bad file descriptor
4077 * EINTR Interrupted
4078 * EKEEPLOOKING Other references were active, try again.
4079 */
4080 int
4081 fp_tryswap(proc_t p, int fd, struct fileproc *nfp)
4082 {
4083 struct fileproc *fp;
4084 int error;
4085
4086 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
4087
4088 if (0 != (error = fp_lookup(p, fd, &fp, 1))) {
4089 return error;
4090 }
4091 /*
4092 * At this point, our caller (change_guardedfd_np) has
4093 * one f_iocount reference, and we just took another
4094 * one to begin the replacement.
4095 * fp and nfp have a +1 reference from allocation.
4096 * Thus if no-one else is looking, f_iocount should be 3.
4097 */
4098 if (os_ref_get_count(&fp->f_iocount) < 3 ||
4099 1 != os_ref_get_count(&nfp->f_iocount)) {
4100 panic("%s: f_iocount", __func__);
4101 } else if (3 == os_ref_get_count(&fp->f_iocount)) {
4102 /* Copy the contents of *fp, preserving the "type" of *nfp */
4103
4104 nfp->f_flags = (nfp->f_flags & FP_TYPEMASK) |
4105 (fp->f_flags & ~FP_TYPEMASK);
4106 os_ref_retain_locked(&nfp->f_iocount);
4107 os_ref_retain_locked(&nfp->f_iocount);
4108 nfp->f_fglob = fp->f_fglob;
4109 nfp->f_wset = fp->f_wset;
4110
4111 p->p_fd->fd_ofiles[fd] = nfp;
4112 fp_drop(p, fd, nfp, 1);
4113
4114 os_ref_release_live(&fp->f_iocount);
4115 os_ref_release_live(&fp->f_iocount);
4116 fileproc_free(fp);
4117 } else {
4118 /*
4119 * Wait for all other active references to evaporate.
4120 */
4121 p->p_fpdrainwait = 1;
4122 error = msleep(&p->p_fpdrainwait, &p->p_fdmlock,
4123 PRIBIO | PCATCH, "tryswap fpdrain", NULL);
4124 if (0 == error) {
4125 /*
4126 * Return an "internal" errno to trigger a full
4127 * reevaluation of the change-guard attempt.
4128 */
4129 error = EKEEPLOOKING;
4130 }
4131 (void) fp_drop(p, fd, fp, 1);
4132 }
4133 return error;
4134 }
4135
4136
4137 /*
4138 * fp_drop_written
4139 *
4140 * Description: Set the FP_WRITTEN flag on the fileproc and drop the I/O
4141 * reference previously taken by calling fp_lookup et. al.
4142 *
4143 * Parameters: p Process in which the fd lives
4144 * fd fd associated with the fileproc
4145 * fp fileproc on which to set the
4146 * flag and drop the reference
4147 *
4148 * Returns: 0 Success
4149 * fp_drop:EBADF Bad file descriptor
4150 *
4151 * Locks: This function internally takes and drops the proc_fdlock for
4152 * the supplied process
4153 *
4154 * Notes: The fileproc must correspond to the fd in the supplied proc
4155 */
4156 int
4157 fp_drop_written(proc_t p, int fd, struct fileproc *fp)
4158 {
4159 int error;
4160
4161 proc_fdlock_spin(p);
4162
4163 fp->f_flags |= FP_WRITTEN;
4164
4165 error = fp_drop(p, fd, fp, 1);
4166
4167 proc_fdunlock(p);
4168
4169 return error;
4170 }
4171
4172
4173 /*
4174 * fp_drop_event
4175 *
4176 * Description: Set the FP_WAITEVENT flag on the fileproc and drop the I/O
4177 * reference previously taken by calling fp_lookup et. al.
4178 *
4179 * Parameters: p Process in which the fd lives
4180 * fd fd associated with the fileproc
4181 * fp fileproc on which to set the
4182 * flag and drop the reference
4183 *
4184 * Returns: 0 Success
4185 * fp_drop:EBADF Bad file descriptor
4186 *
4187 * Locks: This function internally takes and drops the proc_fdlock for
4188 * the supplied process
4189 *
4190 * Notes: The fileproc must correspond to the fd in the supplied proc
4191 */
4192 int
4193 fp_drop_event(proc_t p, int fd, struct fileproc *fp)
4194 {
4195 int error;
4196
4197 proc_fdlock_spin(p);
4198
4199 fp->f_flags |= FP_WAITEVENT;
4200
4201 error = fp_drop(p, fd, fp, 1);
4202
4203 proc_fdunlock(p);
4204
4205 return error;
4206 }
4207
4208
4209 /*
4210 * fp_drop
4211 *
4212 * Description: Drop the I/O reference previously taken by calling fp_lookup
4213 * et. al.
4214 *
4215 * Parameters: p Process in which the fd lives
4216 * fd fd associated with the fileproc
4217 * fp fileproc on which to set the
4218 * flag and drop the reference
4219 * locked flag to internally take and
4220 * drop proc_fdlock if it is not
4221 * already held by the caller
4222 *
4223 * Returns: 0 Success
4224 * EBADF Bad file descriptor
4225 *
4226 * Locks: This function internally takes and drops the proc_fdlock for
4227 * the supplied process if 'locked' is non-zero, and assumes that
4228 * the caller already holds this lock if 'locked' is non-zero.
4229 *
4230 * Notes: The fileproc must correspond to the fd in the supplied proc
4231 */
4232 int
4233 fp_drop(proc_t p, int fd, struct fileproc *fp, int locked)
4234 {
4235 struct filedesc *fdp = p->p_fd;
4236 int needwakeup = 0;
4237
4238 if (!locked) {
4239 proc_fdlock_spin(p);
4240 }
4241 if ((fp == FILEPROC_NULL) && (fd < 0 || fd >= fdp->fd_nfiles ||
4242 (fp = fdp->fd_ofiles[fd]) == NULL ||
4243 ((fdp->fd_ofileflags[fd] & UF_RESERVED) &&
4244 !(fdp->fd_ofileflags[fd] & UF_CLOSING)))) {
4245 if (!locked) {
4246 proc_fdunlock(p);
4247 }
4248 return EBADF;
4249 }
4250
4251 if (1 == os_ref_release_locked(&fp->f_iocount)) {
4252 if (fp->f_flags & FP_SELCONFLICT) {
4253 fp->f_flags &= ~FP_SELCONFLICT;
4254 }
4255
4256 if (p->p_fpdrainwait) {
4257 p->p_fpdrainwait = 0;
4258 needwakeup = 1;
4259 }
4260 }
4261 if (!locked) {
4262 proc_fdunlock(p);
4263 }
4264 if (needwakeup) {
4265 wakeup(&p->p_fpdrainwait);
4266 }
4267
4268 return 0;
4269 }
4270
4271
4272 /*
4273 * file_vnode
4274 *
4275 * Description: Given an fd, look it up in the current process's per process
4276 * open file table, and return its internal vnode pointer.
4277 *
4278 * Parameters: fd fd to obtain vnode from
4279 * vpp pointer to vnode return area
4280 *
4281 * Returns: 0 Success
4282 * EINVAL The fd does not refer to a
4283 * vnode fileproc entry
4284 * fp_lookup:EBADF Bad file descriptor
4285 *
4286 * Implicit returns:
4287 * *vpp (modified) Returned vnode pointer
4288 *
4289 * Locks: This function internally takes and drops the proc_fdlock for
4290 * the current process
4291 *
4292 * Notes: If successful, this function increments the f_iocount on the
4293 * fd's corresponding fileproc.
4294 *
4295 * The fileproc referenced is not returned; because of this, care
4296 * must be taken to not drop the last reference (e.g. by closing
4297 * the file). This is inherently unsafe, since the reference may
4298 * not be recoverable from the vnode, if there is a subsequent
4299 * close that destroys the associate fileproc. The caller should
4300 * therefore retain their own reference on the fileproc so that
4301 * the f_iocount can be dropped subsequently. Failure to do this
4302 * can result in the returned pointer immediately becoming invalid
4303 * following the call.
4304 *
4305 * Use of this function is discouraged.
4306 */
4307 int
4308 file_vnode(int fd, struct vnode **vpp)
4309 {
4310 proc_t p = current_proc();
4311 struct fileproc *fp;
4312 int error;
4313
4314 proc_fdlock_spin(p);
4315 if ((error = fp_lookup(p, fd, &fp, 1))) {
4316 proc_fdunlock(p);
4317 return error;
4318 }
4319 if (fp->f_type != DTYPE_VNODE) {
4320 fp_drop(p, fd, fp, 1);
4321 proc_fdunlock(p);
4322 return EINVAL;
4323 }
4324 if (vpp != NULL) {
4325 *vpp = (struct vnode *)fp->f_data;
4326 }
4327 proc_fdunlock(p);
4328
4329 return 0;
4330 }
4331
4332
4333 /*
4334 * file_vnode_withvid
4335 *
4336 * Description: Given an fd, look it up in the current process's per process
4337 * open file table, and return its internal vnode pointer.
4338 *
4339 * Parameters: fd fd to obtain vnode from
4340 * vpp pointer to vnode return area
4341 * vidp pointer to vid of the returned vnode
4342 *
4343 * Returns: 0 Success
4344 * EINVAL The fd does not refer to a
4345 * vnode fileproc entry
4346 * fp_lookup:EBADF Bad file descriptor
4347 *
4348 * Implicit returns:
4349 * *vpp (modified) Returned vnode pointer
4350 *
4351 * Locks: This function internally takes and drops the proc_fdlock for
4352 * the current process
4353 *
4354 * Notes: If successful, this function increments the f_iocount on the
4355 * fd's corresponding fileproc.
4356 *
4357 * The fileproc referenced is not returned; because of this, care
4358 * must be taken to not drop the last reference (e.g. by closing
4359 * the file). This is inherently unsafe, since the reference may
4360 * not be recoverable from the vnode, if there is a subsequent
4361 * close that destroys the associate fileproc. The caller should
4362 * therefore retain their own reference on the fileproc so that
4363 * the f_iocount can be dropped subsequently. Failure to do this
4364 * can result in the returned pointer immediately becoming invalid
4365 * following the call.
4366 *
4367 * Use of this function is discouraged.
4368 */
4369 int
4370 file_vnode_withvid(int fd, struct vnode **vpp, uint32_t * vidp)
4371 {
4372 proc_t p = current_proc();
4373 struct fileproc *fp;
4374 vnode_t vp;
4375 int error;
4376
4377 proc_fdlock_spin(p);
4378 if ((error = fp_lookup(p, fd, &fp, 1))) {
4379 proc_fdunlock(p);
4380 return error;
4381 }
4382 if (fp->f_type != DTYPE_VNODE) {
4383 fp_drop(p, fd, fp, 1);
4384 proc_fdunlock(p);
4385 return EINVAL;
4386 }
4387 vp = (struct vnode *)fp->f_data;
4388 if (vpp != NULL) {
4389 *vpp = vp;
4390 }
4391
4392 if ((vidp != NULL) && (vp != NULLVP)) {
4393 *vidp = (uint32_t)vp->v_id;
4394 }
4395
4396 proc_fdunlock(p);
4397
4398 return 0;
4399 }
4400
4401
4402 /*
4403 * file_socket
4404 *
4405 * Description: Given an fd, look it up in the current process's per process
4406 * open file table, and return its internal socket pointer.
4407 *
4408 * Parameters: fd fd to obtain vnode from
4409 * sp pointer to socket return area
4410 *
4411 * Returns: 0 Success
4412 * ENOTSOCK Not a socket
4413 * fp_lookup:EBADF Bad file descriptor
4414 *
4415 * Implicit returns:
4416 * *sp (modified) Returned socket pointer
4417 *
4418 * Locks: This function internally takes and drops the proc_fdlock for
4419 * the current process
4420 *
4421 * Notes: If successful, this function increments the f_iocount on the
4422 * fd's corresponding fileproc.
4423 *
4424 * The fileproc referenced is not returned; because of this, care
4425 * must be taken to not drop the last reference (e.g. by closing
4426 * the file). This is inherently unsafe, since the reference may
4427 * not be recoverable from the socket, if there is a subsequent
4428 * close that destroys the associate fileproc. The caller should
4429 * therefore retain their own reference on the fileproc so that
4430 * the f_iocount can be dropped subsequently. Failure to do this
4431 * can result in the returned pointer immediately becoming invalid
4432 * following the call.
4433 *
4434 * Use of this function is discouraged.
4435 */
4436 int
4437 file_socket(int fd, struct socket **sp)
4438 {
4439 proc_t p = current_proc();
4440 struct fileproc *fp;
4441 int error;
4442
4443 proc_fdlock_spin(p);
4444 if ((error = fp_lookup(p, fd, &fp, 1))) {
4445 proc_fdunlock(p);
4446 return error;
4447 }
4448 if (fp->f_type != DTYPE_SOCKET) {
4449 fp_drop(p, fd, fp, 1);
4450 proc_fdunlock(p);
4451 return ENOTSOCK;
4452 }
4453 *sp = (struct socket *)fp->f_data;
4454 proc_fdunlock(p);
4455
4456 return 0;
4457 }
4458
4459
4460 /*
4461 * file_flags
4462 *
4463 * Description: Given an fd, look it up in the current process's per process
4464 * open file table, and return its fileproc's flags field.
4465 *
4466 * Parameters: fd fd whose flags are to be
4467 * retrieved
4468 * flags pointer to flags data area
4469 *
4470 * Returns: 0 Success
4471 * ENOTSOCK Not a socket
4472 * fp_lookup:EBADF Bad file descriptor
4473 *
4474 * Implicit returns:
4475 * *flags (modified) Returned flags field
4476 *
4477 * Locks: This function internally takes and drops the proc_fdlock for
4478 * the current process
4479 *
4480 * Notes: This function will internally increment and decrement the
4481 * f_iocount of the fileproc as part of its operation.
4482 */
4483 int
4484 file_flags(int fd, int *flags)
4485 {
4486 proc_t p = current_proc();
4487 struct fileproc *fp;
4488 int error;
4489
4490 proc_fdlock_spin(p);
4491 if ((error = fp_lookup(p, fd, &fp, 1))) {
4492 proc_fdunlock(p);
4493 return error;
4494 }
4495 *flags = (int)fp->f_flag;
4496 fp_drop(p, fd, fp, 1);
4497 proc_fdunlock(p);
4498
4499 return 0;
4500 }
4501
4502
4503 /*
4504 * file_drop
4505 *
4506 * Description: Drop an iocount reference on an fd, and wake up any waiters
4507 * for draining (i.e. blocked in fileproc_drain() called during
4508 * the last attempt to close a file).
4509 *
4510 * Parameters: fd fd on which an ioreference is
4511 * to be dropped
4512 *
4513 * Returns: 0 Success
4514 * EBADF Bad file descriptor
4515 *
4516 * Description: Given an fd, look it up in the current process's per process
4517 * open file table, and drop it's fileproc's f_iocount by one
4518 *
4519 * Notes: This is intended as a corresponding operation to the functions
4520 * file_vnode() and file_socket() operations.
4521 *
4522 * Technically, the close reference is supposed to be protected
4523 * by a fileproc_drain(), however, a drain will only block if
4524 * the fd refers to a character device, and that device has had
4525 * preparefileread() called on it. If it refers to something
4526 * other than a character device, then the drain will occur and
4527 * block each close attempt, rather than merely the last close.
4528 *
4529 * Since it's possible for an fd that refers to a character
4530 * device to have an intermediate close followed by an open to
4531 * cause a different file to correspond to that descriptor,
4532 * unless there was a cautionary reference taken on the fileproc,
4533 * this is an inherently unsafe function. This happens in the
4534 * case where multiple fd's in a process refer to the same
4535 * character device (e.g. stdin/out/err pointing to a tty, etc.).
4536 *
4537 * Use of this function is discouraged.
4538 */
4539 int
4540 file_drop(int fd)
4541 {
4542 struct fileproc *fp;
4543 proc_t p = current_proc();
4544 int needwakeup = 0;
4545
4546 proc_fdlock_spin(p);
4547 if (fd < 0 || fd >= p->p_fd->fd_nfiles ||
4548 (fp = p->p_fd->fd_ofiles[fd]) == NULL ||
4549 ((p->p_fd->fd_ofileflags[fd] & UF_RESERVED) &&
4550 !(p->p_fd->fd_ofileflags[fd] & UF_CLOSING))) {
4551 proc_fdunlock(p);
4552 return EBADF;
4553 }
4554
4555 if (1 == os_ref_release_locked(&fp->f_iocount)) {
4556 if (fp->f_flags & FP_SELCONFLICT) {
4557 fp->f_flags &= ~FP_SELCONFLICT;
4558 }
4559
4560 if (p->p_fpdrainwait) {
4561 p->p_fpdrainwait = 0;
4562 needwakeup = 1;
4563 }
4564 }
4565 proc_fdunlock(p);
4566
4567 if (needwakeup) {
4568 wakeup(&p->p_fpdrainwait);
4569 }
4570 return 0;
4571 }
4572
4573
4574 static int falloc_withalloc_locked(proc_t, struct fileproc **, int *,
4575 vfs_context_t, struct fileproc * (*)(void *), void *, int);
4576
4577 /*
4578 * falloc
4579 *
4580 * Description: Allocate an entry in the per process open file table and
4581 * return the corresponding fileproc and fd.
4582 *
4583 * Parameters: p The process in whose open file
4584 * table the fd is to be allocated
4585 * resultfp Pointer to fileproc pointer
4586 * return area
4587 * resultfd Pointer to fd return area
4588 * ctx VFS context
4589 *
4590 * Returns: 0 Success
4591 * falloc:ENFILE Too many open files in system
4592 * falloc:EMFILE Too many open files in process
4593 * falloc:ENOMEM M_FILEPROC or M_FILEGLOB zone
4594 * exhausted
4595 *
4596 * Implicit returns:
4597 * *resultfd (modified) Returned fileproc pointer
4598 * *resultfd (modified) Returned fd
4599 *
4600 * Locks: This function takes and drops the proc_fdlock; if this lock
4601 * is already held, use falloc_locked() instead.
4602 *
4603 * Notes: This function takes separate process and context arguments
4604 * solely to support kern_exec.c; otherwise, it would take
4605 * neither, and expect falloc_locked() to use the
4606 * vfs_context_current() routine internally.
4607 */
4608 int
4609 falloc(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx)
4610 {
4611 return falloc_withalloc(p, resultfp, resultfd, ctx,
4612 fileproc_alloc_init, NULL);
4613 }
4614
4615 /*
4616 * Like falloc, but including the fileproc allocator and create-args
4617 */
4618 int
4619 falloc_withalloc(proc_t p, struct fileproc **resultfp, int *resultfd,
4620 vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *arg)
4621 {
4622 int error;
4623
4624 proc_fdlock(p);
4625 error = falloc_withalloc_locked(p,
4626 resultfp, resultfd, ctx, fp_zalloc, arg, 1);
4627 proc_fdunlock(p);
4628
4629 return error;
4630 }
4631
4632 /*
4633 * "uninitialized" ops -- ensure fg->fg_ops->fo_type always exists
4634 */
4635 static const struct fileops uninitops;
4636
4637 /*
4638 * falloc_locked
4639 *
4640 * Create a new open file structure and allocate
4641 * a file descriptor for the process that refers to it.
4642 *
4643 * Returns: 0 Success
4644 *
4645 * Description: Allocate an entry in the per process open file table and
4646 * return the corresponding fileproc and fd.
4647 *
4648 * Parameters: p The process in whose open file
4649 * table the fd is to be allocated
4650 * resultfp Pointer to fileproc pointer
4651 * return area
4652 * resultfd Pointer to fd return area
4653 * ctx VFS context
4654 * locked Flag to indicate whether the
4655 * caller holds proc_fdlock
4656 *
4657 * Returns: 0 Success
4658 * ENFILE Too many open files in system
4659 * fdalloc:EMFILE Too many open files in process
4660 * ENOMEM M_FILEPROC or M_FILEGLOB zone
4661 * exhausted
4662 * fdalloc:ENOMEM
4663 *
4664 * Implicit returns:
4665 * *resultfd (modified) Returned fileproc pointer
4666 * *resultfd (modified) Returned fd
4667 *
4668 * Locks: If the parameter 'locked' is zero, this function takes and
4669 * drops the proc_fdlock; if non-zero, the caller must hold the
4670 * lock.
4671 *
4672 * Notes: If you intend to use a non-zero 'locked' parameter, use the
4673 * utility function falloc() instead.
4674 *
4675 * This function takes separate process and context arguments
4676 * solely to support kern_exec.c; otherwise, it would take
4677 * neither, and use the vfs_context_current() routine internally.
4678 */
4679 int
4680 falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd,
4681 vfs_context_t ctx, int locked)
4682 {
4683 return falloc_withalloc_locked(p, resultfp, resultfd, ctx,
4684 fileproc_alloc_init, NULL, locked);
4685 }
4686
4687 static int
4688 falloc_withalloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd,
4689 vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *crarg,
4690 int locked)
4691 {
4692 struct fileproc *fp;
4693 struct fileglob *fg;
4694 int error, nfd;
4695
4696 if (nfiles >= maxfiles) {
4697 tablefull("file");
4698 return ENFILE;
4699 }
4700
4701 if (!locked) {
4702 proc_fdlock(p);
4703 }
4704
4705 if ((error = fdalloc(p, 0, &nfd))) {
4706 if (!locked) {
4707 proc_fdunlock(p);
4708 }
4709 return error;
4710 }
4711
4712 #if CONFIG_MACF
4713 error = mac_file_check_create(proc_ucred(p));
4714 if (error) {
4715 if (!locked) {
4716 proc_fdunlock(p);
4717 }
4718 return error;
4719 }
4720 #endif
4721
4722 /*
4723 * Allocate a new file descriptor.
4724 * If the process has file descriptor zero open, add to the list
4725 * of open files at that point, otherwise put it at the front of
4726 * the list of open files.
4727 */
4728 proc_fdunlock(p);
4729
4730 fp = (*fp_zalloc)(crarg);
4731 if (fp == NULL) {
4732 if (locked) {
4733 proc_fdlock(p);
4734 }
4735 return ENOMEM;
4736 }
4737 MALLOC_ZONE(fg, struct fileglob *, sizeof(struct fileglob), M_FILEGLOB, M_WAITOK);
4738 if (fg == NULL) {
4739 fileproc_free(fp);
4740 if (locked) {
4741 proc_fdlock(p);
4742 }
4743 return ENOMEM;
4744 }
4745 bzero(fg, sizeof(struct fileglob));
4746 lck_mtx_init(&fg->fg_lock, file_lck_grp, file_lck_attr);
4747
4748 os_ref_retain_locked(&fp->f_iocount);
4749 fg->fg_count = 1;
4750 fg->fg_ops = &uninitops;
4751 fp->f_fglob = fg;
4752 #if CONFIG_MACF
4753 mac_file_label_init(fg);
4754 #endif
4755
4756 kauth_cred_ref(ctx->vc_ucred);
4757
4758 proc_fdlock(p);
4759
4760 fp->f_cred = ctx->vc_ucred;
4761
4762 #if CONFIG_MACF
4763 mac_file_label_associate(fp->f_cred, fg);
4764 #endif
4765
4766 OSAddAtomic(1, &nfiles);
4767
4768 p->p_fd->fd_ofiles[nfd] = fp;
4769
4770 if (!locked) {
4771 proc_fdunlock(p);
4772 }
4773
4774 if (resultfp) {
4775 *resultfp = fp;
4776 }
4777 if (resultfd) {
4778 *resultfd = nfd;
4779 }
4780
4781 return 0;
4782 }
4783
4784
4785 /*
4786 * fg_free
4787 *
4788 * Description: Free a file structure; drop the global open file count, and
4789 * drop the credential reference, if the fileglob has one, and
4790 * destroy the instance mutex before freeing
4791 *
4792 * Parameters: fg Pointer to fileglob to be
4793 * freed
4794 *
4795 * Returns: void
4796 */
4797 void
4798 fg_free(struct fileglob *fg)
4799 {
4800 OSAddAtomic(-1, &nfiles);
4801
4802 if (fg->fg_vn_data) {
4803 fg_vn_data_free(fg->fg_vn_data);
4804 fg->fg_vn_data = NULL;
4805 }
4806
4807 if (IS_VALID_CRED(fg->fg_cred)) {
4808 kauth_cred_unref(&fg->fg_cred);
4809 }
4810 lck_mtx_destroy(&fg->fg_lock, file_lck_grp);
4811
4812 #if CONFIG_MACF
4813 mac_file_label_destroy(fg);
4814 #endif
4815 FREE_ZONE(fg, sizeof *fg, M_FILEGLOB);
4816 }
4817
4818
4819 /*
4820 * fg_get_vnode
4821 *
4822 * Description: Return vnode associated with the file structure, if
4823 * any. The lifetime of the returned vnode is bound to
4824 * the lifetime of the file structure.
4825 *
4826 * Parameters: fg Pointer to fileglob to
4827 * inspect
4828 *
4829 * Returns: vnode_t
4830 */
4831 vnode_t
4832 fg_get_vnode(struct fileglob *fg)
4833 {
4834 if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE) {
4835 return (vnode_t)fg->fg_data;
4836 } else {
4837 return NULL;
4838 }
4839 }
4840
4841 /*
4842 * fdexec
4843 *
4844 * Description: Perform close-on-exec processing for all files in a process
4845 * that are either marked as close-on-exec, or which were in the
4846 * process of being opened at the time of the execve
4847 *
4848 * Also handles the case (via posix_spawn()) where -all-
4849 * files except those marked with "inherit" as treated as
4850 * close-on-exec.
4851 *
4852 * Parameters: p Pointer to process calling
4853 * execve
4854 *
4855 * Returns: void
4856 *
4857 * Locks: This function internally takes and drops proc_fdlock()
4858 * But assumes tables don't grow/change while unlocked.
4859 *
4860 */
4861 void
4862 fdexec(proc_t p, short flags, int self_exec)
4863 {
4864 struct filedesc *fdp = p->p_fd;
4865 int i;
4866 boolean_t cloexec_default = (flags & POSIX_SPAWN_CLOEXEC_DEFAULT) != 0;
4867 thread_t self = current_thread();
4868 struct uthread *ut = get_bsdthread_info(self);
4869 struct kqworkq *dealloc_kqwq = NULL;
4870
4871 /*
4872 * If the current thread is bound as a workq/workloop
4873 * servicing thread, we need to unbind it first.
4874 */
4875 if (ut->uu_kqr_bound && self_exec) {
4876 kqueue_threadreq_unbind(p, ut->uu_kqr_bound);
4877 }
4878
4879 proc_fdlock(p);
4880
4881 /*
4882 * Deallocate the knotes for this process
4883 * and mark the tables non-existent so
4884 * subsequent kqueue closes go faster.
4885 */
4886 knotes_dealloc(p);
4887 assert(fdp->fd_knlistsize == 0);
4888 assert(fdp->fd_knhashmask == 0);
4889
4890 for (i = fdp->fd_lastfile; i >= 0; i--) {
4891 struct fileproc *fp = fdp->fd_ofiles[i];
4892 char *flagp = &fdp->fd_ofileflags[i];
4893
4894 if (fp && cloexec_default) {
4895 /*
4896 * Reverse the usual semantics of file descriptor
4897 * inheritance - all of them should be closed
4898 * except files marked explicitly as "inherit" and
4899 * not marked close-on-exec.
4900 */
4901 if ((*flagp & (UF_EXCLOSE | UF_INHERIT)) != UF_INHERIT) {
4902 *flagp |= UF_EXCLOSE;
4903 }
4904 *flagp &= ~UF_INHERIT;
4905 }
4906
4907 if (
4908 ((*flagp & (UF_RESERVED | UF_EXCLOSE)) == UF_EXCLOSE)
4909 #if CONFIG_MACF
4910 || (fp && mac_file_check_inherit(proc_ucred(p), fp->f_fglob))
4911 #endif
4912 ) {
4913 procfdtbl_clearfd(p, i);
4914 if (i == fdp->fd_lastfile && i > 0) {
4915 fdp->fd_lastfile--;
4916 }
4917 if (i < fdp->fd_freefile) {
4918 fdp->fd_freefile = i;
4919 }
4920
4921 /*
4922 * Wait for any third party viewers (e.g., lsof)
4923 * to release their references to this fileproc.
4924 */
4925 while (os_ref_get_count(&fp->f_iocount) > 1) {
4926 p->p_fpdrainwait = 1;
4927 msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO,
4928 "fpdrain", NULL);
4929 }
4930 if (fp->f_flags & FP_WAITEVENT) {
4931 (void)waitevent_close(p, fp);
4932 }
4933 closef_locked(fp, fp->f_fglob, p);
4934
4935 fileproc_free(fp);
4936 }
4937 }
4938
4939 /* release the per-process workq kq */
4940 if (fdp->fd_wqkqueue) {
4941 dealloc_kqwq = fdp->fd_wqkqueue;
4942 fdp->fd_wqkqueue = NULL;
4943 }
4944
4945 proc_fdunlock(p);
4946
4947 /* Anything to free? */
4948 if (dealloc_kqwq) {
4949 kqworkq_dealloc(dealloc_kqwq);
4950 }
4951 }
4952
4953
4954 /*
4955 * fdcopy
4956 *
4957 * Description: Copy a filedesc structure. This is normally used as part of
4958 * forkproc() when forking a new process, to copy the per process
4959 * open file table over to the new process.
4960 *
4961 * Parameters: p Process whose open file table
4962 * is to be copied (parent)
4963 * uth_cdir Per thread current working
4964 * cirectory, or NULL
4965 *
4966 * Returns: NULL Copy failed
4967 * !NULL Pointer to new struct filedesc
4968 *
4969 * Locks: This function internally takes and drops proc_fdlock()
4970 *
4971 * Notes: Files are copied directly, ignoring the new resource limits
4972 * for the process that's being copied into. Since the descriptor
4973 * references are just additional references, this does not count
4974 * against the number of open files on the system.
4975 *
4976 * The struct filedesc includes the current working directory,
4977 * and the current root directory, if the process is chroot'ed.
4978 *
4979 * If the exec was called by a thread using a per thread current
4980 * working directory, we inherit the working directory from the
4981 * thread making the call, rather than from the process.
4982 *
4983 * In the case of a failure to obtain a reference, for most cases,
4984 * the file entry will be silently dropped. There's an exception
4985 * for the case of a chroot dir, since a failure to to obtain a
4986 * reference there would constitute an "escape" from the chroot
4987 * environment, which must not be allowed. In that case, we will
4988 * deny the execve() operation, rather than allowing the escape.
4989 */
4990 struct filedesc *
4991 fdcopy(proc_t p, vnode_t uth_cdir)
4992 {
4993 struct filedesc *newfdp, *fdp = p->p_fd;
4994 int i;
4995 struct fileproc *ofp, *fp;
4996 vnode_t v_dir;
4997
4998 MALLOC_ZONE(newfdp, struct filedesc *,
4999 sizeof(*newfdp), M_FILEDESC, M_WAITOK);
5000 if (newfdp == NULL) {
5001 return NULL;
5002 }
5003
5004 proc_fdlock(p);
5005
5006 /*
5007 * the FD_CHROOT flag will be inherited via this copy
5008 */
5009 (void) memcpy(newfdp, fdp, sizeof(*newfdp));
5010
5011 /*
5012 * If we are running with per-thread current working directories,
5013 * inherit the new current working directory from the current thread
5014 * instead, before we take our references.
5015 */
5016 if (uth_cdir != NULLVP) {
5017 newfdp->fd_cdir = uth_cdir;
5018 }
5019
5020 /*
5021 * For both fd_cdir and fd_rdir make sure we get
5022 * a valid reference... if we can't, than set
5023 * set the pointer(s) to NULL in the child... this
5024 * will keep us from using a non-referenced vp
5025 * and allows us to do the vnode_rele only on
5026 * a properly referenced vp
5027 */
5028 if ((v_dir = newfdp->fd_cdir)) {
5029 if (vnode_getwithref(v_dir) == 0) {
5030 if ((vnode_ref(v_dir))) {
5031 newfdp->fd_cdir = NULL;
5032 }
5033 vnode_put(v_dir);
5034 } else {
5035 newfdp->fd_cdir = NULL;
5036 }
5037 }
5038 if (newfdp->fd_cdir == NULL && fdp->fd_cdir) {
5039 /*
5040 * we couldn't get a new reference on
5041 * the current working directory being
5042 * inherited... we might as well drop
5043 * our reference from the parent also
5044 * since the vnode has gone DEAD making
5045 * it useless... by dropping it we'll
5046 * be that much closer to recycling it
5047 */
5048 vnode_rele(fdp->fd_cdir);
5049 fdp->fd_cdir = NULL;
5050 }
5051
5052 if ((v_dir = newfdp->fd_rdir)) {
5053 if (vnode_getwithref(v_dir) == 0) {
5054 if ((vnode_ref(v_dir))) {
5055 newfdp->fd_rdir = NULL;
5056 }
5057 vnode_put(v_dir);
5058 } else {
5059 newfdp->fd_rdir = NULL;
5060 }
5061 }
5062 /* Coming from a chroot environment and unable to get a reference... */
5063 if (newfdp->fd_rdir == NULL && fdp->fd_rdir) {
5064 /*
5065 * We couldn't get a new reference on
5066 * the chroot directory being
5067 * inherited... this is fatal, since
5068 * otherwise it would constitute an
5069 * escape from a chroot environment by
5070 * the new process.
5071 */
5072 if (newfdp->fd_cdir) {
5073 vnode_rele(newfdp->fd_cdir);
5074 }
5075 FREE_ZONE(newfdp, sizeof *newfdp, M_FILEDESC);
5076 return NULL;
5077 }
5078
5079 /*
5080 * If the number of open files fits in the internal arrays
5081 * of the open file structure, use them, otherwise allocate
5082 * additional memory for the number of descriptors currently
5083 * in use.
5084 */
5085 if (newfdp->fd_lastfile < NDFILE) {
5086 i = NDFILE;
5087 } else {
5088 /*
5089 * Compute the smallest multiple of NDEXTENT needed
5090 * for the file descriptors currently in use,
5091 * allowing the table to shrink.
5092 */
5093 i = newfdp->fd_nfiles;
5094 while (i > 1 + 2 * NDEXTENT && i > 1 + newfdp->fd_lastfile * 2) {
5095 i /= 2;
5096 }
5097 }
5098 proc_fdunlock(p);
5099
5100 MALLOC_ZONE(newfdp->fd_ofiles, struct fileproc **,
5101 i * OFILESIZE, M_OFILETABL, M_WAITOK);
5102 if (newfdp->fd_ofiles == NULL) {
5103 if (newfdp->fd_cdir) {
5104 vnode_rele(newfdp->fd_cdir);
5105 }
5106 if (newfdp->fd_rdir) {
5107 vnode_rele(newfdp->fd_rdir);
5108 }
5109
5110 FREE_ZONE(newfdp, sizeof(*newfdp), M_FILEDESC);
5111 return NULL;
5112 }
5113 (void) memset(newfdp->fd_ofiles, 0, i * OFILESIZE);
5114 proc_fdlock(p);
5115
5116 newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
5117 newfdp->fd_nfiles = i;
5118
5119 if (fdp->fd_nfiles > 0) {
5120 struct fileproc **fpp;
5121 char *flags;
5122
5123 (void) memcpy(newfdp->fd_ofiles, fdp->fd_ofiles,
5124 (newfdp->fd_lastfile + 1) * sizeof(*fdp->fd_ofiles));
5125 (void) memcpy(newfdp->fd_ofileflags, fdp->fd_ofileflags,
5126 (newfdp->fd_lastfile + 1) * sizeof(*fdp->fd_ofileflags));
5127
5128 fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile];
5129 flags = &newfdp->fd_ofileflags[newfdp->fd_lastfile];
5130 for (i = newfdp->fd_lastfile; i >= 0; i--, fpp--, flags--) {
5131 if ((ofp = *fpp) != NULL &&
5132 0 == (ofp->f_fglob->fg_lflags & FG_CONFINED) &&
5133 0 == (*flags & (UF_FORKCLOSE | UF_RESERVED))) {
5134 #if DEBUG
5135 if (FILEPROC_TYPE(ofp) != FTYPE_SIMPLE) {
5136 panic("complex fileproc");
5137 }
5138 #endif
5139 fp = fileproc_alloc_init(NULL);
5140 if (fp == NULL) {
5141 /*
5142 * XXX no room to copy, unable to
5143 * XXX safely unwind state at present
5144 */
5145 *fpp = NULL;
5146 } else {
5147 fp->f_flags |=
5148 (ofp->f_flags & ~FP_TYPEMASK);
5149 fp->f_fglob = ofp->f_fglob;
5150 (void)fg_ref(fp);
5151 *fpp = fp;
5152 }
5153 } else {
5154 *fpp = NULL;
5155 *flags = 0;
5156 }
5157 if (*fpp == NULL) {
5158 if (i == newfdp->fd_lastfile && i > 0) {
5159 newfdp->fd_lastfile--;
5160 }
5161 if (i < newfdp->fd_freefile) {
5162 newfdp->fd_freefile = i;
5163 }
5164 }
5165 }
5166 }
5167
5168 proc_fdunlock(p);
5169
5170 /*
5171 * Initialize knote and kqueue tracking structs
5172 */
5173 newfdp->fd_knlist = NULL;
5174 newfdp->fd_knlistsize = 0;
5175 newfdp->fd_knhash = NULL;
5176 newfdp->fd_knhashmask = 0;
5177 newfdp->fd_kqhash = NULL;
5178 newfdp->fd_kqhashmask = 0;
5179 newfdp->fd_wqkqueue = NULL;
5180 lck_mtx_init(&newfdp->fd_kqhashlock, proc_kqhashlock_grp, proc_lck_attr);
5181 lck_mtx_init(&newfdp->fd_knhashlock, proc_knhashlock_grp, proc_lck_attr);
5182
5183 return newfdp;
5184 }
5185
5186
5187 /*
5188 * fdfree
5189 *
5190 * Description: Release a filedesc (per process open file table) structure;
5191 * this is done on process exit(), or from forkproc_free() if
5192 * the fork fails for some reason subsequent to a successful
5193 * call to fdcopy()
5194 *
5195 * Parameters: p Pointer to process going away
5196 *
5197 * Returns: void
5198 *
5199 * Locks: This function internally takes and drops proc_fdlock()
5200 */
5201 void
5202 fdfree(proc_t p)
5203 {
5204 struct filedesc *fdp;
5205 struct fileproc *fp;
5206 struct kqworkq *dealloc_kqwq = NULL;
5207 int i;
5208
5209 proc_fdlock(p);
5210
5211 if (p == kernproc || NULL == (fdp = p->p_fd)) {
5212 proc_fdunlock(p);
5213 return;
5214 }
5215
5216 extern struct filedesc filedesc0;
5217
5218 if (&filedesc0 == fdp) {
5219 panic("filedesc0");
5220 }
5221
5222 /*
5223 * deallocate all the knotes up front and claim empty
5224 * tables to make any subsequent kqueue closes faster.
5225 */
5226 knotes_dealloc(p);
5227 assert(fdp->fd_knlistsize == 0);
5228 assert(fdp->fd_knhashmask == 0);
5229
5230 /*
5231 * dealloc all workloops that have outstanding retains
5232 * when created with scheduling parameters.
5233 */
5234 kqworkloops_dealloc(p);
5235
5236 /* close file descriptors */
5237 if (fdp->fd_nfiles > 0 && fdp->fd_ofiles) {
5238 for (i = fdp->fd_lastfile; i >= 0; i--) {
5239 if ((fp = fdp->fd_ofiles[i]) != NULL) {
5240 if (fdp->fd_ofileflags[i] & UF_RESERVED) {
5241 panic("fdfree: found fp with UF_RESERVED");
5242 }
5243
5244 fileproc_drain(p, fp);
5245 procfdtbl_reservefd(p, i);
5246
5247 if (fp->f_flags & FP_WAITEVENT) {
5248 (void)waitevent_close(p, fp);
5249 }
5250 (void) closef_locked(fp, fp->f_fglob, p);
5251 fileproc_free(fp);
5252 }
5253 }
5254 FREE_ZONE(fdp->fd_ofiles, fdp->fd_nfiles * OFILESIZE, M_OFILETABL);
5255 fdp->fd_ofiles = NULL;
5256 fdp->fd_nfiles = 0;
5257 }
5258
5259 if (fdp->fd_wqkqueue) {
5260 dealloc_kqwq = fdp->fd_wqkqueue;
5261 fdp->fd_wqkqueue = NULL;
5262 }
5263
5264 proc_fdunlock(p);
5265
5266 if (dealloc_kqwq) {
5267 kqworkq_dealloc(dealloc_kqwq);
5268 }
5269 if (fdp->fd_cdir) {
5270 vnode_rele(fdp->fd_cdir);
5271 }
5272 if (fdp->fd_rdir) {
5273 vnode_rele(fdp->fd_rdir);
5274 }
5275
5276 proc_fdlock_spin(p);
5277 p->p_fd = NULL;
5278 proc_fdunlock(p);
5279
5280 if (fdp->fd_kqhash) {
5281 for (uint32_t j = 0; j <= fdp->fd_kqhashmask; j++) {
5282 assert(LIST_EMPTY(&fdp->fd_kqhash[j]));
5283 }
5284 FREE(fdp->fd_kqhash, M_KQUEUE);
5285 }
5286
5287 lck_mtx_destroy(&fdp->fd_kqhashlock, proc_kqhashlock_grp);
5288 lck_mtx_destroy(&fdp->fd_knhashlock, proc_knhashlock_grp);
5289
5290 FREE_ZONE(fdp, sizeof(*fdp), M_FILEDESC);
5291 }
5292
5293 /*
5294 * closef_locked
5295 *
5296 * Description: Internal form of closef; called with proc_fdlock held
5297 *
5298 * Parameters: fp Pointer to fileproc for fd
5299 * fg Pointer to fileglob for fd
5300 * p Pointer to proc structure
5301 *
5302 * Returns: 0 Success
5303 * closef_finish:??? Anything returnable by a per-fileops
5304 * close function
5305 *
5306 * Note: Decrements reference count on file structure; if this was the
5307 * last reference, then closef_finish() is called
5308 *
5309 * p and fp are allowed to be NULL when closing a file that was
5310 * being passed in a message (but only if we are called when this
5311 * is NOT the last reference).
5312 */
5313 int
5314 closef_locked(struct fileproc *fp, struct fileglob *fg, proc_t p)
5315 {
5316 struct vnode *vp;
5317 struct flock lf;
5318 struct vfs_context context;
5319 int error;
5320
5321 if (fg == NULL) {
5322 return 0;
5323 }
5324
5325 /* Set up context with cred stashed in fg */
5326 if (p == current_proc()) {
5327 context.vc_thread = current_thread();
5328 } else {
5329 context.vc_thread = NULL;
5330 }
5331 context.vc_ucred = fg->fg_cred;
5332
5333 /*
5334 * POSIX record locking dictates that any close releases ALL
5335 * locks owned by this process. This is handled by setting
5336 * a flag in the unlock to free ONLY locks obeying POSIX
5337 * semantics, and not to free BSD-style file locks.
5338 * If the descriptor was in a message, POSIX-style locks
5339 * aren't passed with the descriptor.
5340 */
5341 if (p && (p->p_ladvflag & P_LADVLOCK) &&
5342 DTYPE_VNODE == FILEGLOB_DTYPE(fg)) {
5343 proc_fdunlock(p);
5344
5345 lf.l_whence = SEEK_SET;
5346 lf.l_start = 0;
5347 lf.l_len = 0;
5348 lf.l_type = F_UNLCK;
5349 vp = (struct vnode *)fg->fg_data;
5350
5351 if ((error = vnode_getwithref(vp)) == 0) {
5352 (void) VNOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX, &context, NULL);
5353 (void)vnode_put(vp);
5354 }
5355 proc_fdlock(p);
5356 }
5357 lck_mtx_lock_spin(&fg->fg_lock);
5358 fg->fg_count--;
5359
5360 if (fg->fg_count > 0) {
5361 lck_mtx_unlock(&fg->fg_lock);
5362 return 0;
5363 }
5364 #if DIAGNOSTIC
5365 if (fg->fg_count != 0) {
5366 panic("fg %p: being freed with bad fg_count (%d)", fg, fg->fg_count);
5367 }
5368 #endif
5369
5370 if (fp && (fp->f_flags & FP_WRITTEN)) {
5371 fg->fg_flag |= FWASWRITTEN;
5372 }
5373
5374 fg->fg_lflags |= FG_TERM;
5375 lck_mtx_unlock(&fg->fg_lock);
5376
5377 if (p) {
5378 proc_fdunlock(p);
5379 }
5380
5381 /* Since we ensure that fg->fg_ops is always initialized,
5382 * it is safe to invoke fo_close on the fg */
5383 error = fo_close(fg, &context);
5384
5385 fg_free(fg);
5386
5387 if (p) {
5388 proc_fdlock(p);
5389 }
5390
5391 return error;
5392 }
5393
5394
5395 /*
5396 * fileproc_drain
5397 *
5398 * Description: Drain out pending I/O operations
5399 *
5400 * Parameters: p Process closing this file
5401 * fp fileproc struct for the open
5402 * instance on the file
5403 *
5404 * Returns: void
5405 *
5406 * Locks: Assumes the caller holds the proc_fdlock
5407 *
5408 * Notes: For character devices, this occurs on the last close of the
5409 * device; for all other file descriptors, this occurs on each
5410 * close to prevent fd's from being closed out from under
5411 * operations currently in progress and blocked
5412 *
5413 * See Also: file_vnode(), file_socket(), file_drop(), and the cautions
5414 * regarding their use and interaction with this function.
5415 */
5416 void
5417 fileproc_drain(proc_t p, struct fileproc * fp)
5418 {
5419 struct vfs_context context;
5420
5421 context.vc_thread = proc_thread(p); /* XXX */
5422 context.vc_ucred = fp->f_fglob->fg_cred;
5423
5424 /* Set the vflag for drain */
5425 fileproc_modify_vflags(fp, FPV_DRAIN, FALSE);
5426
5427 while (os_ref_get_count(&fp->f_iocount) > 1) {
5428 lck_mtx_convert_spin(&p->p_fdmlock);
5429
5430 fo_drain(fp, &context);
5431 if ((fp->f_flags & FP_INSELECT) == FP_INSELECT) {
5432 if (waitq_wakeup64_all((struct waitq *)fp->f_wset, NO_EVENT64,
5433 THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT) {
5434 panic("bad wait queue for waitq_wakeup64_all %p (fp:%p)", fp->f_wset, fp);
5435 }
5436 }
5437 if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
5438 if (waitq_wakeup64_all(&select_conflict_queue, NO_EVENT64,
5439 THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT) {
5440 panic("bad select_conflict_queue");
5441 }
5442 }
5443 p->p_fpdrainwait = 1;
5444
5445 msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO, "fpdrain", NULL);
5446 }
5447 #if DIAGNOSTIC
5448 if ((fp->f_flags & FP_INSELECT) != 0) {
5449 panic("FP_INSELECT set on drained fp");
5450 }
5451 #endif
5452 if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
5453 fp->f_flags &= ~FP_SELCONFLICT;
5454 }
5455 }
5456
5457
5458 /*
5459 * fp_free
5460 *
5461 * Description: Release the fd and free the fileproc associated with the fd
5462 * in the per process open file table of the specified process;
5463 * these values must correspond.
5464 *
5465 * Parameters: p Process containing fd
5466 * fd fd to be released
5467 * fp fileproc to be freed
5468 */
5469 void
5470 fp_free(proc_t p, int fd, struct fileproc * fp)
5471 {
5472 proc_fdlock_spin(p);
5473 fdrelse(p, fd);
5474 proc_fdunlock(p);
5475
5476 fg_free(fp->f_fglob);
5477 os_ref_release_live(&fp->f_iocount);
5478 fileproc_free(fp);
5479 }
5480
5481
5482 /*
5483 * flock
5484 *
5485 * Description: Apply an advisory lock on a file descriptor.
5486 *
5487 * Parameters: p Process making request
5488 * uap->fd fd on which the lock is to be
5489 * attempted
5490 * uap->how (Un)Lock bits, including type
5491 * retval Pointer to the call return area
5492 *
5493 * Returns: 0 Success
5494 * fp_getfvp:EBADF Bad file descriptor
5495 * fp_getfvp:ENOTSUP fd does not refer to a vnode
5496 * vnode_getwithref:???
5497 * VNOP_ADVLOCK:???
5498 *
5499 * Implicit returns:
5500 * *retval (modified) Size of dtable
5501 *
5502 * Notes: Just attempt to get a record lock of the requested type on
5503 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
5504 */
5505 int
5506 flock(proc_t p, struct flock_args *uap, __unused int32_t *retval)
5507 {
5508 int fd = uap->fd;
5509 int how = uap->how;
5510 struct fileproc *fp;
5511 struct vnode *vp;
5512 struct flock lf;
5513 vfs_context_t ctx = vfs_context_current();
5514 int error = 0;
5515
5516 AUDIT_ARG(fd, uap->fd);
5517 if ((error = fp_getfvp(p, fd, &fp, &vp))) {
5518 return error;
5519 }
5520 if ((error = vnode_getwithref(vp))) {
5521 goto out1;
5522 }
5523 AUDIT_ARG(vnpath, vp, ARG_VNODE1);
5524
5525 lf.l_whence = SEEK_SET;
5526 lf.l_start = 0;
5527 lf.l_len = 0;
5528 if (how & LOCK_UN) {
5529 lf.l_type = F_UNLCK;
5530 fp->f_flag &= ~FHASLOCK;
5531 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_UNLCK, &lf, F_FLOCK, ctx, NULL);
5532 goto out;
5533 }
5534 if (how & LOCK_EX) {
5535 lf.l_type = F_WRLCK;
5536 } else if (how & LOCK_SH) {
5537 lf.l_type = F_RDLCK;
5538 } else {
5539 error = EBADF;
5540 goto out;
5541 }
5542 #if CONFIG_MACF
5543 error = mac_file_check_lock(proc_ucred(p), fp->f_fglob, F_SETLK, &lf);
5544 if (error) {
5545 goto out;
5546 }
5547 #endif
5548 error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf,
5549 (how & LOCK_NB ? F_FLOCK : F_FLOCK | F_WAIT),
5550 ctx, NULL);
5551 if (!error) {
5552 fp->f_flag |= FHASLOCK;
5553 }
5554 out:
5555 (void)vnode_put(vp);
5556 out1:
5557 fp_drop(p, fd, fp, 0);
5558 return error;
5559 }
5560
5561 /*
5562 * fileport_makeport
5563 *
5564 * Description: Obtain a Mach send right for a given file descriptor.
5565 *
5566 * Parameters: p Process calling fileport
5567 * uap->fd The fd to reference
5568 * uap->portnamep User address at which to place port name.
5569 *
5570 * Returns: 0 Success.
5571 * EBADF Bad file descriptor.
5572 * EINVAL File descriptor had type that cannot be sent, misc. other errors.
5573 * EFAULT Address at which to store port name is not valid.
5574 * EAGAIN Resource shortage.
5575 *
5576 * Implicit returns:
5577 * On success, name of send right is stored at user-specified address.
5578 */
5579 int
5580 fileport_makeport(proc_t p, struct fileport_makeport_args *uap,
5581 __unused int *retval)
5582 {
5583 int err;
5584 int fd = uap->fd;
5585 user_addr_t user_portaddr = uap->portnamep;
5586 struct fileproc *fp = FILEPROC_NULL;
5587 struct fileglob *fg = NULL;
5588 ipc_port_t fileport;
5589 mach_port_name_t name = MACH_PORT_NULL;
5590
5591 proc_fdlock(p);
5592 err = fp_lookup(p, fd, &fp, 1);
5593 if (err != 0) {
5594 goto out_unlock;
5595 }
5596
5597 if (!file_issendable(p, fp)) {
5598 err = EINVAL;
5599 goto out_unlock;
5600 }
5601
5602 if (FP_ISGUARDED(fp, GUARD_FILEPORT)) {
5603 err = fp_guard_exception(p, fd, fp, kGUARD_EXC_FILEPORT);
5604 goto out_unlock;
5605 }
5606
5607 /* Dropped when port is deallocated */
5608 fg = fp->f_fglob;
5609 fg_ref(fp);
5610
5611 proc_fdunlock(p);
5612
5613 /* Allocate and initialize a port */
5614 fileport = fileport_alloc(fg);
5615 if (fileport == IPC_PORT_NULL) {
5616 err = EAGAIN;
5617 fg_drop(fp);
5618 goto out;
5619 }
5620
5621 /* Add an entry. Deallocates port on failure. */
5622 name = ipc_port_copyout_send(fileport, get_task_ipcspace(p->task));
5623 if (!MACH_PORT_VALID(name)) {
5624 err = EINVAL;
5625 goto out;
5626 }
5627
5628 err = copyout(&name, user_portaddr, sizeof(mach_port_name_t));
5629 if (err != 0) {
5630 goto out;
5631 }
5632
5633 /* Tag the fileglob for debugging purposes */
5634 lck_mtx_lock_spin(&fg->fg_lock);
5635 fg->fg_lflags |= FG_PORTMADE;
5636 lck_mtx_unlock(&fg->fg_lock);
5637
5638 fp_drop(p, fd, fp, 0);
5639
5640 return 0;
5641
5642 out_unlock:
5643 proc_fdunlock(p);
5644 out:
5645 if (MACH_PORT_VALID(name)) {
5646 /* Don't care if another thread races us to deallocate the entry */
5647 (void) mach_port_deallocate(get_task_ipcspace(p->task), name);
5648 }
5649
5650 if (fp != FILEPROC_NULL) {
5651 fp_drop(p, fd, fp, 0);
5652 }
5653
5654 return err;
5655 }
5656
5657 void
5658 fileport_releasefg(struct fileglob *fg)
5659 {
5660 (void)closef_locked(NULL, fg, PROC_NULL);
5661
5662 return;
5663 }
5664
5665 /*
5666 * fileport_makefd_internal
5667 *
5668 * Description: Obtain the file descriptor for a given Mach send right.
5669 *
5670 * Returns: 0 Success
5671 * EINVAL Invalid Mach port name, or port is not for a file.
5672 * fdalloc:EMFILE
5673 * fdalloc:ENOMEM Unable to allocate fileproc or extend file table.
5674 *
5675 * Implicit returns:
5676 * *retval (modified) The new descriptor
5677 */
5678 int
5679 fileport_makefd_internal(proc_t p, ipc_port_t port, int uf_flags, int *retval)
5680 {
5681 struct fileglob *fg;
5682 struct fileproc *fp = FILEPROC_NULL;
5683 int fd;
5684 int err;
5685
5686 fg = fileport_port_to_fileglob(port);
5687 if (fg == NULL) {
5688 err = EINVAL;
5689 goto out;
5690 }
5691
5692 fp = fileproc_alloc_init(NULL);
5693 if (fp == FILEPROC_NULL) {
5694 err = ENOMEM;
5695 goto out;
5696 }
5697
5698 fp->f_fglob = fg;
5699 fg_ref(fp);
5700
5701 proc_fdlock(p);
5702 err = fdalloc(p, 0, &fd);
5703 if (err != 0) {
5704 proc_fdunlock(p);
5705 fg_drop(fp);
5706 goto out;
5707 }
5708 if (uf_flags) {
5709 *fdflags(p, fd) |= uf_flags;
5710 }
5711
5712 procfdtbl_releasefd(p, fd, fp);
5713 proc_fdunlock(p);
5714
5715 *retval = fd;
5716 err = 0;
5717 out:
5718 if ((fp != NULL) && (0 != err)) {
5719 fileproc_free(fp);
5720 }
5721
5722 return err;
5723 }
5724
5725 /*
5726 * fileport_makefd
5727 *
5728 * Description: Obtain the file descriptor for a given Mach send right.
5729 *
5730 * Parameters: p Process calling fileport
5731 * uap->port Name of send right to file port.
5732 *
5733 * Returns: 0 Success
5734 * EINVAL Invalid Mach port name, or port is not for a file.
5735 * fdalloc:EMFILE
5736 * fdalloc:ENOMEM Unable to allocate fileproc or extend file table.
5737 *
5738 * Implicit returns:
5739 * *retval (modified) The new descriptor
5740 */
5741 int
5742 fileport_makefd(proc_t p, struct fileport_makefd_args *uap, int32_t *retval)
5743 {
5744 ipc_port_t port = IPC_PORT_NULL;
5745 mach_port_name_t send = uap->port;
5746 kern_return_t res;
5747 int err;
5748
5749 res = ipc_object_copyin(get_task_ipcspace(p->task),
5750 send, MACH_MSG_TYPE_COPY_SEND, &port, 0, NULL, IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND);
5751
5752 if (res == KERN_SUCCESS) {
5753 err = fileport_makefd_internal(p, port, UF_EXCLOSE, retval);
5754 } else {
5755 err = EINVAL;
5756 }
5757
5758 if (IPC_PORT_NULL != port) {
5759 ipc_port_release_send(port);
5760 }
5761
5762 return err;
5763 }
5764
5765
5766 /*
5767 * dupfdopen
5768 *
5769 * Description: Duplicate the specified descriptor to a free descriptor;
5770 * this is the second half of fdopen(), above.
5771 *
5772 * Parameters: fdp filedesc pointer to fill in
5773 * indx fd to dup to
5774 * dfd fd to dup from
5775 * mode mode to set on new fd
5776 * error command code
5777 *
5778 * Returns: 0 Success
5779 * EBADF Source fd is bad
5780 * EACCES Requested mode not allowed
5781 * !0 'error', if not ENODEV or
5782 * ENXIO
5783 *
5784 * Notes: XXX This is not thread safe; see fdopen() above
5785 */
5786 int
5787 dupfdopen(struct filedesc *fdp, int indx, int dfd, int flags, int error)
5788 {
5789 struct fileproc *wfp;
5790 struct fileproc *fp;
5791 #if CONFIG_MACF
5792 int myerror;
5793 #endif
5794 proc_t p = current_proc();
5795
5796 /*
5797 * If the to-be-dup'd fd number is greater than the allowed number
5798 * of file descriptors, or the fd to be dup'd has already been
5799 * closed, reject. Note, check for new == old is necessary as
5800 * falloc could allocate an already closed to-be-dup'd descriptor
5801 * as the new descriptor.
5802 */
5803 proc_fdlock(p);
5804
5805 fp = fdp->fd_ofiles[indx];
5806 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
5807 (wfp = fdp->fd_ofiles[dfd]) == NULL || wfp == fp ||
5808 (fdp->fd_ofileflags[dfd] & UF_RESERVED)) {
5809 proc_fdunlock(p);
5810 return EBADF;
5811 }
5812 #if CONFIG_MACF
5813 myerror = mac_file_check_dup(proc_ucred(p), wfp->f_fglob, dfd);
5814 if (myerror) {
5815 proc_fdunlock(p);
5816 return myerror;
5817 }
5818 #endif
5819 /*
5820 * There are two cases of interest here.
5821 *
5822 * For ENODEV simply dup (dfd) to file descriptor
5823 * (indx) and return.
5824 *
5825 * For ENXIO steal away the file structure from (dfd) and
5826 * store it in (indx). (dfd) is effectively closed by
5827 * this operation.
5828 *
5829 * Any other error code is just returned.
5830 */
5831 switch (error) {
5832 case ENODEV:
5833 if (FP_ISGUARDED(wfp, GUARD_DUP)) {
5834 proc_fdunlock(p);
5835 return EPERM;
5836 }
5837
5838 /*
5839 * Check that the mode the file is being opened for is a
5840 * subset of the mode of the existing descriptor.
5841 */
5842 if (((flags & (FREAD | FWRITE)) | wfp->f_flag) != wfp->f_flag) {
5843 proc_fdunlock(p);
5844 return EACCES;
5845 }
5846 if (indx > fdp->fd_lastfile) {
5847 fdp->fd_lastfile = indx;
5848 }
5849 (void)fg_ref(wfp);
5850
5851 if (fp->f_fglob) {
5852 fg_free(fp->f_fglob);
5853 }
5854 fp->f_fglob = wfp->f_fglob;
5855
5856 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd] |
5857 (flags & O_CLOEXEC) ? UF_EXCLOSE : 0;
5858
5859 proc_fdunlock(p);
5860 return 0;
5861
5862 default:
5863 proc_fdunlock(p);
5864 return error;
5865 }
5866 /* NOTREACHED */
5867 }
5868
5869
5870 /*
5871 * fg_ref
5872 *
5873 * Description: Add a reference to a fileglob by fileproc
5874 *
5875 * Parameters: fp fileproc containing fileglob
5876 * pointer
5877 *
5878 * Returns: void
5879 *
5880 * Notes: XXX Should use OSAddAtomic?
5881 */
5882 void
5883 fg_ref(struct fileproc * fp)
5884 {
5885 struct fileglob *fg;
5886
5887 fg = fp->f_fglob;
5888
5889 lck_mtx_lock_spin(&fg->fg_lock);
5890
5891 #if DIAGNOSTIC
5892 if ((fp->f_flags & ~((unsigned int)FP_VALID_FLAGS)) != 0) {
5893 panic("fg_ref: invalid bits on fp %p", fp);
5894 }
5895
5896 if (fg->fg_count == 0) {
5897 panic("fg_ref: adding fgcount to zeroed fg: fp %p fg %p",
5898 fp, fg);
5899 }
5900 #endif
5901 fg->fg_count++;
5902 lck_mtx_unlock(&fg->fg_lock);
5903 }
5904
5905
5906 /*
5907 * fg_drop
5908 *
5909 * Description: Remove a reference to a fileglob by fileproc
5910 *
5911 * Parameters: fp fileproc containing fileglob
5912 * pointer
5913 *
5914 * Returns: void
5915 *
5916 * Notes: XXX Should use OSAddAtomic?
5917 */
5918 void
5919 fg_drop(struct fileproc * fp)
5920 {
5921 struct fileglob *fg;
5922
5923 fg = fp->f_fglob;
5924 lck_mtx_lock_spin(&fg->fg_lock);
5925 fg->fg_count--;
5926 lck_mtx_unlock(&fg->fg_lock);
5927 }
5928
5929 #if SOCKETS
5930 /*
5931 * fg_insertuipc_mark
5932 *
5933 * Description: Mark fileglob for insertion onto message queue if needed
5934 * Also takes fileglob reference
5935 *
5936 * Parameters: fg Fileglob pointer to insert
5937 *
5938 * Returns: true, if the fileglob needs to be inserted onto msg queue
5939 *
5940 * Locks: Takes and drops fg_lock, potentially many times
5941 */
5942 boolean_t
5943 fg_insertuipc_mark(struct fileglob * fg)
5944 {
5945 boolean_t insert = FALSE;
5946
5947 lck_mtx_lock_spin(&fg->fg_lock);
5948 while (fg->fg_lflags & FG_RMMSGQ) {
5949 lck_mtx_convert_spin(&fg->fg_lock);
5950
5951 fg->fg_lflags |= FG_WRMMSGQ;
5952 msleep(&fg->fg_lflags, &fg->fg_lock, 0, "fg_insertuipc", NULL);
5953 }
5954
5955 fg->fg_count++;
5956 fg->fg_msgcount++;
5957 if (fg->fg_msgcount == 1) {
5958 fg->fg_lflags |= FG_INSMSGQ;
5959 insert = TRUE;
5960 }
5961 lck_mtx_unlock(&fg->fg_lock);
5962 return insert;
5963 }
5964
5965 /*
5966 * fg_insertuipc
5967 *
5968 * Description: Insert marked fileglob onto message queue
5969 *
5970 * Parameters: fg Fileglob pointer to insert
5971 *
5972 * Returns: void
5973 *
5974 * Locks: Takes and drops fg_lock & uipc_lock
5975 * DO NOT call this function with proc_fdlock held as unp_gc()
5976 * can potentially try to acquire proc_fdlock, which can result
5977 * in a deadlock if this function is in unp_gc_wait().
5978 */
5979 void
5980 fg_insertuipc(struct fileglob * fg)
5981 {
5982 if (fg->fg_lflags & FG_INSMSGQ) {
5983 lck_mtx_lock_spin(uipc_lock);
5984 unp_gc_wait();
5985 LIST_INSERT_HEAD(&fmsghead, fg, f_msglist);
5986 lck_mtx_unlock(uipc_lock);
5987 lck_mtx_lock(&fg->fg_lock);
5988 fg->fg_lflags &= ~FG_INSMSGQ;
5989 if (fg->fg_lflags & FG_WINSMSGQ) {
5990 fg->fg_lflags &= ~FG_WINSMSGQ;
5991 wakeup(&fg->fg_lflags);
5992 }
5993 lck_mtx_unlock(&fg->fg_lock);
5994 }
5995 }
5996
5997 /*
5998 * fg_removeuipc_mark
5999 *
6000 * Description: Mark the fileglob for removal from message queue if needed
6001 * Also releases fileglob message queue reference
6002 *
6003 * Parameters: fg Fileglob pointer to remove
6004 *
6005 * Returns: true, if the fileglob needs to be removed from msg queue
6006 *
6007 * Locks: Takes and drops fg_lock, potentially many times
6008 */
6009 boolean_t
6010 fg_removeuipc_mark(struct fileglob * fg)
6011 {
6012 boolean_t remove = FALSE;
6013
6014 lck_mtx_lock_spin(&fg->fg_lock);
6015 while (fg->fg_lflags & FG_INSMSGQ) {
6016 lck_mtx_convert_spin(&fg->fg_lock);
6017
6018 fg->fg_lflags |= FG_WINSMSGQ;
6019 msleep(&fg->fg_lflags, &fg->fg_lock, 0, "fg_removeuipc", NULL);
6020 }
6021 fg->fg_msgcount--;
6022 if (fg->fg_msgcount == 0) {
6023 fg->fg_lflags |= FG_RMMSGQ;
6024 remove = TRUE;
6025 }
6026 lck_mtx_unlock(&fg->fg_lock);
6027 return remove;
6028 }
6029
6030 /*
6031 * fg_removeuipc
6032 *
6033 * Description: Remove marked fileglob from message queue
6034 *
6035 * Parameters: fg Fileglob pointer to remove
6036 *
6037 * Returns: void
6038 *
6039 * Locks: Takes and drops fg_lock & uipc_lock
6040 * DO NOT call this function with proc_fdlock held as unp_gc()
6041 * can potentially try to acquire proc_fdlock, which can result
6042 * in a deadlock if this function is in unp_gc_wait().
6043 */
6044 void
6045 fg_removeuipc(struct fileglob * fg)
6046 {
6047 if (fg->fg_lflags & FG_RMMSGQ) {
6048 lck_mtx_lock_spin(uipc_lock);
6049 unp_gc_wait();
6050 LIST_REMOVE(fg, f_msglist);
6051 lck_mtx_unlock(uipc_lock);
6052 lck_mtx_lock(&fg->fg_lock);
6053 fg->fg_lflags &= ~FG_RMMSGQ;
6054 if (fg->fg_lflags & FG_WRMMSGQ) {
6055 fg->fg_lflags &= ~FG_WRMMSGQ;
6056 wakeup(&fg->fg_lflags);
6057 }
6058 lck_mtx_unlock(&fg->fg_lock);
6059 }
6060 }
6061 #endif /* SOCKETS */
6062
6063 /*
6064 * fo_read
6065 *
6066 * Description: Generic fileops read indirected through the fileops pointer
6067 * in the fileproc structure
6068 *
6069 * Parameters: fp fileproc structure pointer
6070 * uio user I/O structure pointer
6071 * flags FOF_ flags
6072 * ctx VFS context for operation
6073 *
6074 * Returns: 0 Success
6075 * !0 Errno from read
6076 */
6077 int
6078 fo_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
6079 {
6080 return (*fp->f_ops->fo_read)(fp, uio, flags, ctx);
6081 }
6082
6083 int
6084 fo_no_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
6085 {
6086 #pragma unused(fp, uio, flags, ctx)
6087 return ENXIO;
6088 }
6089
6090
6091 /*
6092 * fo_write
6093 *
6094 * Description: Generic fileops write indirected through the fileops pointer
6095 * in the fileproc structure
6096 *
6097 * Parameters: fp fileproc structure pointer
6098 * uio user I/O structure pointer
6099 * flags FOF_ flags
6100 * ctx VFS context for operation
6101 *
6102 * Returns: 0 Success
6103 * !0 Errno from write
6104 */
6105 int
6106 fo_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
6107 {
6108 return (*fp->f_ops->fo_write)(fp, uio, flags, ctx);
6109 }
6110
6111 int
6112 fo_no_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
6113 {
6114 #pragma unused(fp, uio, flags, ctx)
6115 return ENXIO;
6116 }
6117
6118
6119 /*
6120 * fo_ioctl
6121 *
6122 * Description: Generic fileops ioctl indirected through the fileops pointer
6123 * in the fileproc structure
6124 *
6125 * Parameters: fp fileproc structure pointer
6126 * com ioctl command
6127 * data pointer to internalized copy
6128 * of user space ioctl command
6129 * parameter data in kernel space
6130 * ctx VFS context for operation
6131 *
6132 * Returns: 0 Success
6133 * !0 Errno from ioctl
6134 *
6135 * Locks: The caller is assumed to have held the proc_fdlock; this
6136 * function releases and reacquires this lock. If the caller
6137 * accesses data protected by this lock prior to calling this
6138 * function, it will need to revalidate/reacquire any cached
6139 * protected data obtained prior to the call.
6140 */
6141 int
6142 fo_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
6143 {
6144 int error;
6145
6146 proc_fdunlock(vfs_context_proc(ctx));
6147 error = (*fp->f_ops->fo_ioctl)(fp, com, data, ctx);
6148 proc_fdlock(vfs_context_proc(ctx));
6149 return error;
6150 }
6151
6152 int
6153 fo_no_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
6154 {
6155 #pragma unused(fp, com, data, ctx)
6156 return ENOTTY;
6157 }
6158
6159
6160 /*
6161 * fo_select
6162 *
6163 * Description: Generic fileops select indirected through the fileops pointer
6164 * in the fileproc structure
6165 *
6166 * Parameters: fp fileproc structure pointer
6167 * which select which
6168 * wql pointer to wait queue list
6169 * ctx VFS context for operation
6170 *
6171 * Returns: 0 Success
6172 * !0 Errno from select
6173 */
6174 int
6175 fo_select(struct fileproc *fp, int which, void *wql, vfs_context_t ctx)
6176 {
6177 return (*fp->f_ops->fo_select)(fp, which, wql, ctx);
6178 }
6179
6180 int
6181 fo_no_select(struct fileproc *fp, int which, void *wql, vfs_context_t ctx)
6182 {
6183 #pragma unused(fp, which, wql, ctx)
6184 return ENOTSUP;
6185 }
6186
6187
6188 /*
6189 * fo_close
6190 *
6191 * Description: Generic fileops close indirected through the fileops pointer
6192 * in the fileproc structure
6193 *
6194 * Parameters: fp fileproc structure pointer for
6195 * file to close
6196 * ctx VFS context for operation
6197 *
6198 * Returns: 0 Success
6199 * !0 Errno from close
6200 */
6201 int
6202 fo_close(struct fileglob *fg, vfs_context_t ctx)
6203 {
6204 return (*fg->fg_ops->fo_close)(fg, ctx);
6205 }
6206
6207
6208 /*
6209 * fo_drain
6210 *
6211 * Description: Generic fileops kqueue filter indirected through the fileops
6212 * pointer in the fileproc structure
6213 *
6214 * Parameters: fp fileproc structure pointer
6215 * ctx VFS context for operation
6216 *
6217 * Returns: 0 Success
6218 * !0 errno from drain
6219 */
6220 int
6221 fo_drain(struct fileproc *fp, vfs_context_t ctx)
6222 {
6223 return (*fp->f_ops->fo_drain)(fp, ctx);
6224 }
6225
6226 int
6227 fo_no_drain(struct fileproc *fp, vfs_context_t ctx)
6228 {
6229 #pragma unused(fp, ctx)
6230 return ENOTSUP;
6231 }
6232
6233
6234 /*
6235 * fo_kqfilter
6236 *
6237 * Description: Generic fileops kqueue filter indirected through the fileops
6238 * pointer in the fileproc structure
6239 *
6240 * Parameters: fp fileproc structure pointer
6241 * kn pointer to knote to filter on
6242 *
6243 * Returns: (kn->kn_flags & EV_ERROR) error in kn->kn_data
6244 * 0 Filter is not active
6245 * !0 Filter is active
6246 */
6247 int
6248 fo_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
6249 {
6250 return (*fp->f_ops->fo_kqfilter)(fp, kn, kev);
6251 }
6252
6253 int
6254 fo_no_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
6255 {
6256 #pragma unused(fp, kev)
6257 knote_set_error(kn, ENOTSUP);
6258 return 0;
6259 }
6260
6261
6262 /*
6263 * The ability to send a file descriptor to another
6264 * process is opt-in by file type.
6265 */
6266 boolean_t
6267 file_issendable(proc_t p, struct fileproc *fp)
6268 {
6269 proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
6270
6271 switch (fp->f_type) {
6272 case DTYPE_VNODE:
6273 case DTYPE_SOCKET:
6274 case DTYPE_PIPE:
6275 case DTYPE_PSXSHM:
6276 case DTYPE_NETPOLICY:
6277 return 0 == (fp->f_fglob->fg_lflags & FG_CONFINED);
6278 default:
6279 /* DTYPE_KQUEUE, DTYPE_FSEVENTS, DTYPE_PSXSEM */
6280 return FALSE;
6281 }
6282 }
6283
6284 os_refgrp_decl(, f_iocount_refgrp, "f_iocount", NULL);
6285
6286 struct fileproc *
6287 fileproc_alloc_init(__unused void *arg)
6288 {
6289 struct fileproc *fp;
6290
6291 MALLOC_ZONE(fp, struct fileproc *, sizeof(*fp), M_FILEPROC, M_WAITOK);
6292 if (fp) {
6293 bzero(fp, sizeof(*fp));
6294 os_ref_init(&fp->f_iocount, &f_iocount_refgrp);
6295 }
6296
6297 return fp;
6298 }
6299
6300
6301 void
6302 fileproc_free(struct fileproc *fp)
6303 {
6304 os_ref_count_t __unused refc = os_ref_release(&fp->f_iocount);
6305 #if DEVELOPMENT || DEBUG
6306 if (0 != refc) {
6307 panic("%s: pid %d refc: %u != 0",
6308 __func__, proc_pid(current_proc()), refc);
6309 }
6310 #endif
6311 switch (FILEPROC_TYPE(fp)) {
6312 case FTYPE_SIMPLE:
6313 FREE_ZONE(fp, sizeof(*fp), M_FILEPROC);
6314 break;
6315 case FTYPE_GUARDED:
6316 guarded_fileproc_free(fp);
6317 break;
6318 default:
6319 panic("%s: corrupt fp %p flags %x", __func__, fp, fp->f_flags);
6320 }
6321 }
6322
6323 void
6324 fileproc_modify_vflags(struct fileproc *fp, fileproc_vflags_t vflags, boolean_t clearflags)
6325 {
6326 if (clearflags) {
6327 os_atomic_andnot(&fp->f_vflags, vflags, relaxed);
6328 } else {
6329 os_atomic_or(&fp->f_vflags, vflags, relaxed);
6330 }
6331 }
6332
6333 fileproc_vflags_t
6334 fileproc_get_vflags(struct fileproc *fp)
6335 {
6336 return os_atomic_load(&fp->f_vflags, relaxed);
6337 }
mirror of XNU