1 /*
2 * Copyright (c) 2000-2021 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 NeXT Computer, Inc. All Rights Reserved */
29 /*
30 * Copyright (c) 1982, 1986, 1989, 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 * @(#)sys_generic.c 8.9 (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/ioctl.h>
79 #include <sys/file_internal.h>
80 #include <sys/proc_internal.h>
81 #include <sys/socketvar.h>
82 #include <sys/uio_internal.h>
83 #include <sys/kernel.h>
84 #include <sys/guarded.h>
85 #include <sys/stat.h>
86 #include <sys/malloc.h>
87 #include <sys/sysproto.h>
88
89 #include <sys/mount_internal.h>
90 #include <sys/protosw.h>
91 #include <sys/ev.h>
92 #include <sys/user.h>
93 #include <sys/kdebug.h>
94 #include <sys/poll.h>
95 #include <sys/event.h>
96 #include <sys/eventvar.h>
97 #include <sys/proc.h>
98 #include <sys/kauth.h>
99
100 #include <machine/smp.h>
101 #include <mach/mach_types.h>
102 #include <kern/kern_types.h>
103 #include <kern/assert.h>
104 #include <kern/kalloc.h>
105 #include <kern/thread.h>
106 #include <kern/clock.h>
107 #include <kern/ledger.h>
108 #include <kern/monotonic.h>
109 #include <kern/task.h>
110 #include <kern/telemetry.h>
111 #include <kern/waitq.h>
112 #include <kern/sched_hygiene.h>
113 #include <kern/sched_prim.h>
114 #include <kern/mpsc_queue.h>
115 #include <kern/debug.h>
116
117 #include <sys/mbuf.h>
118 #include <sys/domain.h>
119 #include <sys/socket.h>
120 #include <sys/socketvar.h>
121 #include <sys/errno.h>
122 #include <sys/syscall.h>
123 #include <sys/pipe.h>
124
125 #include <security/audit/audit.h>
126
127 #include <net/if.h>
128 #include <net/route.h>
129
130 #include <netinet/in.h>
131 #include <netinet/in_systm.h>
132 #include <netinet/ip.h>
133 #include <netinet/in_pcb.h>
134 #include <netinet/ip_var.h>
135 #include <netinet/ip6.h>
136 #include <netinet/tcp.h>
137 #include <netinet/tcp_fsm.h>
138 #include <netinet/tcp_seq.h>
139 #include <netinet/tcp_timer.h>
140 #include <netinet/tcp_var.h>
141 #include <netinet/tcpip.h>
142 #include <netinet/tcp_debug.h>
143 /* for wait queue based select */
144 #include <kern/waitq.h>
145 #include <sys/vnode_internal.h>
146 /* for remote time api*/
147 #include <kern/remote_time.h>
148 #include <os/log.h>
149 #include <sys/log_data.h>
150
151 #if CONFIG_MACF
152 #include <security/mac_framework.h>
153 #endif
154
155 #ifdef CONFIG_KDP_INTERACTIVE_DEBUGGING
156 #include <mach_debug/mach_debug_types.h>
157 #endif
158
159 #if MONOTONIC
160 #include <machine/monotonic.h>
161 #endif /* MONOTONIC */
162
163 /* for entitlement check */
164 #include <IOKit/IOBSD.h>
165
166 /* XXX should be in a header file somewhere */
167 extern kern_return_t IOBSDGetPlatformUUID(__darwin_uuid_t uuid, mach_timespec_t timeoutp);
168
169 int do_uiowrite(struct proc *p, struct fileproc *fp, uio_t uio, int flags, user_ssize_t *retval);
170 __private_extern__ int dofileread(vfs_context_t ctx, struct fileproc *fp,
171 user_addr_t bufp, user_size_t nbyte,
172 off_t offset, int flags, user_ssize_t *retval);
173 __private_extern__ int dofilewrite(vfs_context_t ctx, struct fileproc *fp,
174 user_addr_t bufp, user_size_t nbyte,
175 off_t offset, int flags, user_ssize_t *retval);
176 static int preparefileread(struct proc *p, struct fileproc **fp_ret, int fd, int check_for_vnode);
177
178 /* needed by guarded_writev, etc. */
179 int write_internal(struct proc *p, int fd, user_addr_t buf, user_size_t nbyte,
180 off_t offset, int flags, guardid_t *puguard, user_ssize_t *retval);
181 int writev_uio(struct proc *p, int fd, user_addr_t user_iovp, int iovcnt, off_t offset, int flags,
182 guardid_t *puguard, user_ssize_t *retval);
183
184 #define f_flag fp_glob->fg_flag
185 #define f_type fp_glob->fg_ops->fo_type
186 #define f_cred fp_glob->fg_cred
187 #define f_ops fp_glob->fg_ops
188
189 /*
190 * Validate if the file can be used for random access (pread, pwrite, etc).
191 *
192 * Conditions:
193 * proc_fdlock is held
194 *
195 * Returns: 0 Success
196 * ESPIPE
197 * ENXIO
198 */
199 static int
valid_for_random_access(struct fileproc * fp)200 valid_for_random_access(struct fileproc *fp)
201 {
202 if (__improbable(fp->f_type != DTYPE_VNODE)) {
203 return ESPIPE;
204 }
205
206 vnode_t vp = (struct vnode *)fp_get_data(fp);
207 if (__improbable(vnode_isfifo(vp))) {
208 return ESPIPE;
209 }
210
211 if (__improbable(vp->v_flag & VISTTY)) {
212 return ENXIO;
213 }
214
215 return 0;
216 }
217
218 /*
219 * Returns: 0 Success
220 * EBADF
221 * ESPIPE
222 * ENXIO
223 * fp_lookup:EBADF
224 * valid_for_random_access:ESPIPE
225 * valid_for_random_access:ENXIO
226 */
227 static int
preparefileread(struct proc * p,struct fileproc ** fp_ret,int fd,int check_for_pread)228 preparefileread(struct proc *p, struct fileproc **fp_ret, int fd, int check_for_pread)
229 {
230 int error;
231 struct fileproc *fp;
232
233 AUDIT_ARG(fd, fd);
234
235 proc_fdlock_spin(p);
236
237 error = fp_lookup(p, fd, &fp, 1);
238
239 if (error) {
240 proc_fdunlock(p);
241 return error;
242 }
243 if ((fp->f_flag & FREAD) == 0) {
244 error = EBADF;
245 goto out;
246 }
247 if (check_for_pread) {
248 if ((error = valid_for_random_access(fp))) {
249 goto out;
250 }
251 }
252
253 *fp_ret = fp;
254
255 proc_fdunlock(p);
256 return 0;
257
258 out:
259 fp_drop(p, fd, fp, 1);
260 proc_fdunlock(p);
261 return error;
262 }
263
264 static int
fp_readv(vfs_context_t ctx,struct fileproc * fp,uio_t uio,int flags,user_ssize_t * retval)265 fp_readv(vfs_context_t ctx, struct fileproc *fp, uio_t uio, int flags,
266 user_ssize_t *retval)
267 {
268 int error;
269 user_ssize_t count;
270
271 if ((error = uio_calculateresid(uio))) {
272 *retval = 0;
273 return error;
274 }
275
276 count = uio_resid(uio);
277 error = fo_read(fp, uio, flags, ctx);
278
279 switch (error) {
280 case ERESTART:
281 case EINTR:
282 case EWOULDBLOCK:
283 if (uio_resid(uio) != count) {
284 error = 0;
285 }
286 break;
287
288 default:
289 break;
290 }
291
292 *retval = count - uio_resid(uio);
293 return error;
294 }
295
296 /*
297 * Returns: 0 Success
298 * EINVAL
299 * fo_read:???
300 */
301 __private_extern__ int
dofileread(vfs_context_t ctx,struct fileproc * fp,user_addr_t bufp,user_size_t nbyte,off_t offset,int flags,user_ssize_t * retval)302 dofileread(vfs_context_t ctx, struct fileproc *fp,
303 user_addr_t bufp, user_size_t nbyte, off_t offset, int flags,
304 user_ssize_t *retval)
305 {
306 uio_stackbuf_t uio_buf[UIO_SIZEOF(1)];
307 uio_t uio;
308 int spacetype;
309
310 if (nbyte > INT_MAX) {
311 *retval = 0;
312 return EINVAL;
313 }
314
315 spacetype = vfs_context_is64bit(ctx) ? UIO_USERSPACE64 : UIO_USERSPACE32;
316 uio = uio_createwithbuffer(1, offset, spacetype, UIO_READ, &uio_buf[0],
317 sizeof(uio_buf));
318
319 if (uio_addiov(uio, bufp, nbyte) != 0) {
320 *retval = 0;
321 return EINVAL;
322 }
323
324 return fp_readv(ctx, fp, uio, flags, retval);
325 }
326
327 static int
readv_internal(struct proc * p,int fd,uio_t uio,int flags,user_ssize_t * retval)328 readv_internal(struct proc *p, int fd, uio_t uio, int flags,
329 user_ssize_t *retval)
330 {
331 struct fileproc *fp = NULL;
332 struct vfs_context context;
333 int error;
334
335 if ((error = preparefileread(p, &fp, fd, flags & FOF_OFFSET))) {
336 *retval = 0;
337 return error;
338 }
339
340 context = *(vfs_context_current());
341 context.vc_ucred = fp->fp_glob->fg_cred;
342
343 error = fp_readv(&context, fp, uio, flags, retval);
344
345 fp_drop(p, fd, fp, 0);
346 return error;
347 }
348
349 static int
read_internal(struct proc * p,int fd,user_addr_t buf,user_size_t nbyte,off_t offset,int flags,user_ssize_t * retval)350 read_internal(struct proc *p, int fd, user_addr_t buf, user_size_t nbyte,
351 off_t offset, int flags, user_ssize_t *retval)
352 {
353 uio_stackbuf_t uio_buf[UIO_SIZEOF(1)];
354 uio_t uio;
355 int spacetype = IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
356
357 if (nbyte > INT_MAX) {
358 *retval = 0;
359 return EINVAL;
360 }
361
362 uio = uio_createwithbuffer(1, offset, spacetype, UIO_READ,
363 &uio_buf[0], sizeof(uio_buf));
364
365 if (uio_addiov(uio, buf, nbyte) != 0) {
366 *retval = 0;
367 return EINVAL;
368 }
369
370 return readv_internal(p, fd, uio, flags, retval);
371 }
372
373 int
read_nocancel(struct proc * p,struct read_nocancel_args * uap,user_ssize_t * retval)374 read_nocancel(struct proc *p, struct read_nocancel_args *uap, user_ssize_t *retval)
375 {
376 return read_internal(p, uap->fd, uap->cbuf, uap->nbyte, (off_t)-1, 0,
377 retval);
378 }
379
380 /*
381 * Read system call.
382 *
383 * Returns: 0 Success
384 * preparefileread:EBADF
385 * preparefileread:ESPIPE
386 * preparefileread:ENXIO
387 * preparefileread:EBADF
388 * dofileread:???
389 */
390 int
read(struct proc * p,struct read_args * uap,user_ssize_t * retval)391 read(struct proc *p, struct read_args *uap, user_ssize_t *retval)
392 {
393 __pthread_testcancel(1);
394 return read_nocancel(p, (struct read_nocancel_args *)uap, retval);
395 }
396
397 int
pread_nocancel(struct proc * p,struct pread_nocancel_args * uap,user_ssize_t * retval)398 pread_nocancel(struct proc *p, struct pread_nocancel_args *uap, user_ssize_t *retval)
399 {
400 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_SC_EXTENDED_INFO, SYS_pread) | DBG_FUNC_NONE),
401 uap->fd, uap->nbyte, (unsigned int)((uap->offset >> 32)), (unsigned int)(uap->offset), 0);
402
403 return read_internal(p, uap->fd, uap->buf, uap->nbyte, uap->offset,
404 FOF_OFFSET, retval);
405 }
406
407 /*
408 * Pread system call
409 *
410 * Returns: 0 Success
411 * preparefileread:EBADF
412 * preparefileread:ESPIPE
413 * preparefileread:ENXIO
414 * preparefileread:EBADF
415 * dofileread:???
416 */
417 int
pread(struct proc * p,struct pread_args * uap,user_ssize_t * retval)418 pread(struct proc *p, struct pread_args *uap, user_ssize_t *retval)
419 {
420 __pthread_testcancel(1);
421 return pread_nocancel(p, (struct pread_nocancel_args *)uap, retval);
422 }
423
424 /*
425 * Vector read.
426 *
427 * Returns: 0 Success
428 * EINVAL
429 * ENOMEM
430 * preparefileread:EBADF
431 * preparefileread:ESPIPE
432 * preparefileread:ENXIO
433 * preparefileread:EBADF
434 * copyin:EFAULT
435 * rd_uio:???
436 */
437 static int
readv_uio(struct proc * p,int fd,user_addr_t user_iovp,int iovcnt,off_t offset,int flags,user_ssize_t * retval)438 readv_uio(struct proc *p, int fd,
439 user_addr_t user_iovp, int iovcnt, off_t offset, int flags,
440 user_ssize_t *retval)
441 {
442 uio_t uio = NULL;
443 int error;
444 struct user_iovec *iovp;
445
446 if (iovcnt <= 0 || iovcnt > UIO_MAXIOV) {
447 error = EINVAL;
448 goto out;
449 }
450
451 uio = uio_create(iovcnt, offset,
452 (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32),
453 UIO_READ);
454
455 iovp = uio_iovsaddr(uio);
456 if (iovp == NULL) {
457 error = ENOMEM;
458 goto out;
459 }
460
461 error = copyin_user_iovec_array(user_iovp,
462 IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32,
463 iovcnt, iovp);
464
465 if (error) {
466 goto out;
467 }
468
469 error = readv_internal(p, fd, uio, flags, retval);
470
471 out:
472 if (uio != NULL) {
473 uio_free(uio);
474 }
475
476 return error;
477 }
478
479 int
readv_nocancel(struct proc * p,struct readv_nocancel_args * uap,user_ssize_t * retval)480 readv_nocancel(struct proc *p, struct readv_nocancel_args *uap, user_ssize_t *retval)
481 {
482 return readv_uio(p, uap->fd, uap->iovp, uap->iovcnt, 0, 0, retval);
483 }
484
485 /*
486 * Scatter read system call.
487 */
488 int
readv(struct proc * p,struct readv_args * uap,user_ssize_t * retval)489 readv(struct proc *p, struct readv_args *uap, user_ssize_t *retval)
490 {
491 __pthread_testcancel(1);
492 return readv_nocancel(p, (struct readv_nocancel_args *)uap, retval);
493 }
494
495 int
sys_preadv_nocancel(struct proc * p,struct preadv_nocancel_args * uap,user_ssize_t * retval)496 sys_preadv_nocancel(struct proc *p, struct preadv_nocancel_args *uap, user_ssize_t *retval)
497 {
498 return readv_uio(p, uap->fd, uap->iovp, uap->iovcnt, uap->offset,
499 FOF_OFFSET, retval);
500 }
501
502 /*
503 * Preadv system call
504 */
505 int
sys_preadv(struct proc * p,struct preadv_args * uap,user_ssize_t * retval)506 sys_preadv(struct proc *p, struct preadv_args *uap, user_ssize_t *retval)
507 {
508 __pthread_testcancel(1);
509 return sys_preadv_nocancel(p, (struct preadv_nocancel_args *)uap, retval);
510 }
511
512 /*
513 * Returns: 0 Success
514 * EBADF
515 * ESPIPE
516 * ENXIO
517 * fp_lookup:EBADF
518 * fp_guard_exception:???
519 * valid_for_random_access:ESPIPE
520 * valid_for_random_access:ENXIO
521 */
522 static int
preparefilewrite(struct proc * p,struct fileproc ** fp_ret,int fd,int check_for_pwrite,guardid_t * puguard)523 preparefilewrite(struct proc *p, struct fileproc **fp_ret, int fd, int check_for_pwrite,
524 guardid_t *puguard)
525 {
526 int error;
527 struct fileproc *fp;
528
529 AUDIT_ARG(fd, fd);
530
531 proc_fdlock_spin(p);
532
533 if (puguard) {
534 error = fp_lookup_guarded(p, fd, *puguard, &fp, 1);
535 if (error) {
536 proc_fdunlock(p);
537 return error;
538 }
539
540 if ((fp->f_flag & FWRITE) == 0) {
541 error = EBADF;
542 goto out;
543 }
544 } else {
545 error = fp_lookup(p, fd, &fp, 1);
546 if (error) {
547 proc_fdunlock(p);
548 return error;
549 }
550
551 /* Allow EBADF first. */
552 if ((fp->f_flag & FWRITE) == 0) {
553 error = EBADF;
554 goto out;
555 }
556
557 if (fp_isguarded(fp, GUARD_WRITE)) {
558 error = fp_guard_exception(p, fd, fp, kGUARD_EXC_WRITE);
559 goto out;
560 }
561 }
562
563 if (check_for_pwrite) {
564 if ((error = valid_for_random_access(fp))) {
565 goto out;
566 }
567 }
568
569 *fp_ret = fp;
570
571 proc_fdunlock(p);
572 return 0;
573
574 out:
575 fp_drop(p, fd, fp, 1);
576 proc_fdunlock(p);
577 return error;
578 }
579
580 static int
fp_writev(vfs_context_t ctx,struct fileproc * fp,uio_t uio,int flags,user_ssize_t * retval)581 fp_writev(vfs_context_t ctx, struct fileproc *fp, uio_t uio, int flags,
582 user_ssize_t *retval)
583 {
584 int error;
585 user_ssize_t count;
586
587 if ((error = uio_calculateresid(uio))) {
588 *retval = 0;
589 return error;
590 }
591
592 count = uio_resid(uio);
593 error = fo_write(fp, uio, flags, ctx);
594
595 switch (error) {
596 case ERESTART:
597 case EINTR:
598 case EWOULDBLOCK:
599 if (uio_resid(uio) != count) {
600 error = 0;
601 }
602 break;
603
604 case EPIPE:
605 if (fp->f_type != DTYPE_SOCKET &&
606 (fp->fp_glob->fg_lflags & FG_NOSIGPIPE) == 0) {
607 /* XXX Raise the signal on the thread? */
608 psignal(vfs_context_proc(ctx), SIGPIPE);
609 }
610 break;
611
612 default:
613 break;
614 }
615
616 if ((*retval = count - uio_resid(uio))) {
617 os_atomic_or(&fp->fp_glob->fg_flag, FWASWRITTEN, relaxed);
618 }
619
620 return error;
621 }
622
623 /*
624 * Returns: 0 Success
625 * EINVAL
626 * <fo_write>:EPIPE
627 * <fo_write>:??? [indirect through struct fileops]
628 */
629 __private_extern__ int
dofilewrite(vfs_context_t ctx,struct fileproc * fp,user_addr_t bufp,user_size_t nbyte,off_t offset,int flags,user_ssize_t * retval)630 dofilewrite(vfs_context_t ctx, struct fileproc *fp,
631 user_addr_t bufp, user_size_t nbyte, off_t offset, int flags,
632 user_ssize_t *retval)
633 {
634 uio_stackbuf_t uio_buf[UIO_SIZEOF(1)];
635 uio_t uio;
636 int spacetype;
637
638 if (nbyte > INT_MAX) {
639 *retval = 0;
640 return EINVAL;
641 }
642
643 spacetype = vfs_context_is64bit(ctx) ? UIO_USERSPACE64 : UIO_USERSPACE32;
644 uio = uio_createwithbuffer(1, offset, spacetype, UIO_WRITE, &uio_buf[0],
645 sizeof(uio_buf));
646
647 if (uio_addiov(uio, bufp, nbyte) != 0) {
648 *retval = 0;
649 return EINVAL;
650 }
651
652 return fp_writev(ctx, fp, uio, flags, retval);
653 }
654
655 static int
writev_internal(struct proc * p,int fd,uio_t uio,int flags,guardid_t * puguard,user_ssize_t * retval)656 writev_internal(struct proc *p, int fd, uio_t uio, int flags,
657 guardid_t *puguard, user_ssize_t *retval)
658 {
659 struct fileproc *fp = NULL;
660 struct vfs_context context;
661 int error;
662
663 if ((error = preparefilewrite(p, &fp, fd, flags & FOF_OFFSET, puguard))) {
664 *retval = 0;
665 return error;
666 }
667
668 context = *(vfs_context_current());
669 context.vc_ucred = fp->fp_glob->fg_cred;
670
671 error = fp_writev(&context, fp, uio, flags, retval);
672
673 fp_drop(p, fd, fp, 0);
674 return error;
675 }
676
677 int
write_internal(struct proc * p,int fd,user_addr_t buf,user_size_t nbyte,off_t offset,int flags,guardid_t * puguard,user_ssize_t * retval)678 write_internal(struct proc *p, int fd, user_addr_t buf, user_size_t nbyte,
679 off_t offset, int flags, guardid_t *puguard, user_ssize_t *retval)
680 {
681 uio_stackbuf_t uio_buf[UIO_SIZEOF(1)];
682 uio_t uio;
683 int spacetype = IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
684
685 if (nbyte > INT_MAX) {
686 *retval = 0;
687 return EINVAL;
688 }
689
690 uio = uio_createwithbuffer(1, offset, spacetype, UIO_WRITE,
691 &uio_buf[0], sizeof(uio_buf));
692
693 if (uio_addiov(uio, buf, nbyte) != 0) {
694 *retval = 0;
695 return EINVAL;
696 }
697
698 return writev_internal(p, fd, uio, flags, puguard, retval);
699 }
700
701 int
write_nocancel(struct proc * p,struct write_nocancel_args * uap,user_ssize_t * retval)702 write_nocancel(struct proc *p, struct write_nocancel_args *uap, user_ssize_t *retval)
703 {
704 return write_internal(p, uap->fd, uap->cbuf, uap->nbyte, (off_t)-1, 0,
705 NULL, retval);
706 }
707
708 /*
709 * Write system call
710 *
711 * Returns: 0 Success
712 * EBADF
713 * fp_lookup:EBADF
714 * dofilewrite:???
715 */
716 int
write(struct proc * p,struct write_args * uap,user_ssize_t * retval)717 write(struct proc *p, struct write_args *uap, user_ssize_t *retval)
718 {
719 __pthread_testcancel(1);
720 return write_nocancel(p, (struct write_nocancel_args *)uap, retval);
721 }
722
723 int
pwrite_nocancel(struct proc * p,struct pwrite_nocancel_args * uap,user_ssize_t * retval)724 pwrite_nocancel(struct proc *p, struct pwrite_nocancel_args *uap, user_ssize_t *retval)
725 {
726 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_SC_EXTENDED_INFO, SYS_pwrite) | DBG_FUNC_NONE),
727 uap->fd, uap->nbyte, (unsigned int)((uap->offset >> 32)), (unsigned int)(uap->offset), 0);
728
729 /* XXX: Should be < 0 instead? (See man page + pwritev) */
730 if (uap->offset == (off_t)-1) {
731 return EINVAL;
732 }
733
734 return write_internal(p, uap->fd, uap->buf, uap->nbyte, uap->offset,
735 FOF_OFFSET, NULL, retval);
736 }
737
738 /*
739 * pwrite system call
740 *
741 * Returns: 0 Success
742 * EBADF
743 * ESPIPE
744 * ENXIO
745 * EINVAL
746 * fp_lookup:EBADF
747 * dofilewrite:???
748 */
749 int
pwrite(struct proc * p,struct pwrite_args * uap,user_ssize_t * retval)750 pwrite(struct proc *p, struct pwrite_args *uap, user_ssize_t *retval)
751 {
752 __pthread_testcancel(1);
753 return pwrite_nocancel(p, (struct pwrite_nocancel_args *)uap, retval);
754 }
755
756 int
writev_uio(struct proc * p,int fd,user_addr_t user_iovp,int iovcnt,off_t offset,int flags,guardid_t * puguard,user_ssize_t * retval)757 writev_uio(struct proc *p, int fd,
758 user_addr_t user_iovp, int iovcnt, off_t offset, int flags,
759 guardid_t *puguard, user_ssize_t *retval)
760 {
761 uio_t uio = NULL;
762 int error;
763 struct user_iovec *iovp;
764
765 if (iovcnt <= 0 || iovcnt > UIO_MAXIOV || offset < 0) {
766 error = EINVAL;
767 goto out;
768 }
769
770 uio = uio_create(iovcnt, offset,
771 (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32),
772 UIO_WRITE);
773
774 iovp = uio_iovsaddr(uio);
775 if (iovp == NULL) {
776 error = ENOMEM;
777 goto out;
778 }
779
780 error = copyin_user_iovec_array(user_iovp,
781 IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32,
782 iovcnt, iovp);
783
784 if (error) {
785 goto out;
786 }
787
788 error = writev_internal(p, fd, uio, flags, puguard, retval);
789
790 out:
791 if (uio != NULL) {
792 uio_free(uio);
793 }
794
795 return error;
796 }
797
798 int
writev_nocancel(struct proc * p,struct writev_nocancel_args * uap,user_ssize_t * retval)799 writev_nocancel(struct proc *p, struct writev_nocancel_args *uap, user_ssize_t *retval)
800 {
801 return writev_uio(p, uap->fd, uap->iovp, uap->iovcnt, 0, 0, NULL, retval);
802 }
803
804 /*
805 * Gather write system call
806 */
807 int
writev(struct proc * p,struct writev_args * uap,user_ssize_t * retval)808 writev(struct proc *p, struct writev_args *uap, user_ssize_t *retval)
809 {
810 __pthread_testcancel(1);
811 return writev_nocancel(p, (struct writev_nocancel_args *)uap, retval);
812 }
813
814 int
sys_pwritev_nocancel(struct proc * p,struct pwritev_nocancel_args * uap,user_ssize_t * retval)815 sys_pwritev_nocancel(struct proc *p, struct pwritev_nocancel_args *uap, user_ssize_t *retval)
816 {
817 return writev_uio(p, uap->fd, uap->iovp, uap->iovcnt, uap->offset,
818 FOF_OFFSET, NULL, retval);
819 }
820
821 /*
822 * Pwritev system call
823 */
824 int
sys_pwritev(struct proc * p,struct pwritev_args * uap,user_ssize_t * retval)825 sys_pwritev(struct proc *p, struct pwritev_args *uap, user_ssize_t *retval)
826 {
827 __pthread_testcancel(1);
828 return sys_pwritev_nocancel(p, (struct pwritev_nocancel_args *)uap, retval);
829 }
830
831 /*
832 * Ioctl system call
833 *
834 * Returns: 0 Success
835 * EBADF
836 * ENOTTY
837 * ENOMEM
838 * ESRCH
839 * copyin:EFAULT
840 * copyoutEFAULT
841 * fp_lookup:EBADF Bad file descriptor
842 * fo_ioctl:???
843 */
844 int
ioctl(struct proc * p,struct ioctl_args * uap,__unused int32_t * retval)845 ioctl(struct proc *p, struct ioctl_args *uap, __unused int32_t *retval)
846 {
847 struct fileproc *fp = NULL;
848 int error = 0;
849 u_int size = 0;
850 caddr_t datap = NULL, memp = NULL;
851 boolean_t is64bit = FALSE;
852 int tmp = 0;
853 #define STK_PARAMS 128
854 char stkbuf[STK_PARAMS] = {};
855 int fd = uap->fd;
856 u_long com = uap->com;
857 struct vfs_context context = *vfs_context_current();
858
859 AUDIT_ARG(fd, uap->fd);
860 AUDIT_ARG(addr, uap->data);
861
862 is64bit = proc_is64bit(p);
863 #if CONFIG_AUDIT
864 if (is64bit) {
865 AUDIT_ARG(value64, com);
866 } else {
867 AUDIT_ARG(cmd, CAST_DOWN_EXPLICIT(int, com));
868 }
869 #endif /* CONFIG_AUDIT */
870
871 /*
872 * Interpret high order word to find amount of data to be
873 * copied to/from the user's address space.
874 */
875 size = IOCPARM_LEN(com);
876 if (size > IOCPARM_MAX) {
877 return ENOTTY;
878 }
879 if (size > sizeof(stkbuf)) {
880 memp = (caddr_t)kalloc_data(size, Z_WAITOK);
881 if (memp == 0) {
882 return ENOMEM;
883 }
884 datap = memp;
885 } else {
886 datap = &stkbuf[0];
887 }
888 if (com & IOC_IN) {
889 if (size) {
890 error = copyin(uap->data, datap, size);
891 if (error) {
892 goto out_nofp;
893 }
894 } else {
895 /* XXX - IOC_IN and no size? we should proably return an error here!! */
896 if (is64bit) {
897 *(user_addr_t *)datap = uap->data;
898 } else {
899 *(uint32_t *)datap = (uint32_t)uap->data;
900 }
901 }
902 } else if ((com & IOC_OUT) && size) {
903 /*
904 * Zero the buffer so the user always
905 * gets back something deterministic.
906 */
907 bzero(datap, size);
908 } else if (com & IOC_VOID) {
909 /* XXX - this is odd since IOC_VOID means no parameters */
910 if (is64bit) {
911 *(user_addr_t *)datap = uap->data;
912 } else {
913 *(uint32_t *)datap = (uint32_t)uap->data;
914 }
915 }
916
917 proc_fdlock(p);
918 error = fp_lookup(p, fd, &fp, 1);
919 if (error) {
920 proc_fdunlock(p);
921 goto out_nofp;
922 }
923
924 AUDIT_ARG(file, p, fp);
925
926 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
927 error = EBADF;
928 goto out;
929 }
930
931 context.vc_ucred = fp->fp_glob->fg_cred;
932
933 #if CONFIG_MACF
934 error = mac_file_check_ioctl(context.vc_ucred, fp->fp_glob, com);
935 if (error) {
936 goto out;
937 }
938 #endif
939
940 switch (com) {
941 case FIONCLEX:
942 fp->fp_flags &= ~FP_CLOEXEC;
943 break;
944
945 case FIOCLEX:
946 fp->fp_flags |= FP_CLOEXEC;
947 break;
948
949 case FIONBIO:
950 // FIXME (rdar://54898652)
951 //
952 // this code is broken if fnctl(F_SETFL), ioctl() are
953 // called concurrently for the same fileglob.
954 if ((tmp = *(int *)datap)) {
955 os_atomic_or(&fp->f_flag, FNONBLOCK, relaxed);
956 } else {
957 os_atomic_andnot(&fp->f_flag, FNONBLOCK, relaxed);
958 }
959 error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
960 break;
961
962 case FIOASYNC:
963 // FIXME (rdar://54898652)
964 //
965 // this code is broken if fnctl(F_SETFL), ioctl() are
966 // called concurrently for the same fileglob.
967 if ((tmp = *(int *)datap)) {
968 os_atomic_or(&fp->f_flag, FASYNC, relaxed);
969 } else {
970 os_atomic_andnot(&fp->f_flag, FASYNC, relaxed);
971 }
972 error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, &context);
973 break;
974
975 case FIOSETOWN:
976 tmp = *(int *)datap;
977 if (fp->f_type == DTYPE_SOCKET) {
978 ((struct socket *)fp_get_data(fp))->so_pgid = tmp;
979 break;
980 }
981 if (fp->f_type == DTYPE_PIPE) {
982 error = fo_ioctl(fp, TIOCSPGRP, (caddr_t)&tmp, &context);
983 break;
984 }
985 if (tmp <= 0) {
986 tmp = -tmp;
987 } else {
988 struct proc *p1 = proc_find(tmp);
989 if (p1 == 0) {
990 error = ESRCH;
991 break;
992 }
993 tmp = p1->p_pgrpid;
994 proc_rele(p1);
995 }
996 error = fo_ioctl(fp, TIOCSPGRP, (caddr_t)&tmp, &context);
997 break;
998
999 case FIOGETOWN:
1000 if (fp->f_type == DTYPE_SOCKET) {
1001 *(int *)datap = ((struct socket *)fp_get_data(fp))->so_pgid;
1002 break;
1003 }
1004 error = fo_ioctl(fp, TIOCGPGRP, datap, &context);
1005 *(int *)datap = -*(int *)datap;
1006 break;
1007
1008 default:
1009 error = fo_ioctl(fp, com, datap, &context);
1010 /*
1011 * Copy any data to user, size was
1012 * already set and checked above.
1013 */
1014 if (error == 0 && (com & IOC_OUT) && size) {
1015 error = copyout(datap, uap->data, (u_int)size);
1016 }
1017 break;
1018 }
1019 out:
1020 fp_drop(p, fd, fp, 1);
1021 proc_fdunlock(p);
1022
1023 out_nofp:
1024 if (memp) {
1025 kfree_data(memp, size);
1026 }
1027 return error;
1028 }
1029
1030 int selwait;
1031 #define SEL_FIRSTPASS 1
1032 #define SEL_SECONDPASS 2
1033 static int selprocess(struct proc *p, int error, int sel_pass);
1034 static int selscan(struct proc *p, struct _select * sel, struct _select_data * seldata,
1035 int nfd, int32_t *retval, int sel_pass, struct select_set *selset);
1036 static int selcount(struct proc *p, u_int32_t *ibits, int nfd, int *count);
1037 static int seldrop_locked(struct proc *p, u_int32_t *ibits, int nfd, int lim, int *need_wakeup);
1038 static int seldrop(struct proc *p, u_int32_t *ibits, int nfd, int lim);
1039 static int select_internal(struct proc *p, struct select_nocancel_args *uap, uint64_t timeout, int32_t *retval);
1040
1041 /*
1042 * This is used for the special device nodes that do not implement
1043 * a proper kevent filter (see filt_specattach).
1044 *
1045 * In order to enable kevents on those, the spec_filtops will pretend
1046 * to call select, and try to sniff the selrecord(), if it observes one,
1047 * the knote is attached, which pairs with selwakeup() or selthreadclear().
1048 *
1049 * The last issue remaining, is that we need to serialize filt_specdetach()
1050 * with this, but it really can't know the "selinfo" or any locking domain.
1051 * To make up for this, We protect knote list operations with a global lock,
1052 * which give us a safe shared locking domain.
1053 *
1054 * Note: It is a little distasteful, but we really have very few of those.
1055 * The big problem here is that sharing a lock domain without
1056 * any kind of shared knowledge is a little complicated.
1057 *
1058 * 1. filters can really implement their own kqueue integration
1059 * to side step this,
1060 *
1061 * 2. There's an opportunity to pick a private lock in selspec_attach()
1062 * because both the selinfo and the knote are locked at that time.
1063 * The cleanup story is however a little complicated.
1064 */
1065 static LCK_GRP_DECLARE(selspec_grp, "spec_filtops");
1066 static LCK_SPIN_DECLARE(selspec_lock, &selspec_grp);
1067
1068 /*
1069 * The "primitive" lock is held.
1070 * The knote lock is held.
1071 */
1072 void
selspec_attach(struct knote * kn,struct selinfo * si)1073 selspec_attach(struct knote *kn, struct selinfo *si)
1074 {
1075 struct selinfo *cur = os_atomic_load(&kn->kn_hook, relaxed);
1076
1077 if (cur == NULL) {
1078 si->si_flags |= SI_SELSPEC;
1079 lck_spin_lock(&selspec_lock);
1080 kn->kn_hook = si;
1081 KNOTE_ATTACH(&si->si_note, kn);
1082 lck_spin_unlock(&selspec_lock);
1083 } else {
1084 /*
1085 * selspec_attach() can be called from e.g. filt_spectouch()
1086 * which might be called before any event was dequeued.
1087 *
1088 * It is hence not impossible for the knote already be hooked.
1089 *
1090 * Note that selwakeup_internal() could possibly
1091 * already have cleared this pointer. This is a race
1092 * that filt_specprocess will debounce.
1093 */
1094 assert(si->si_flags & SI_SELSPEC);
1095 assert(cur == si);
1096 }
1097 }
1098
1099 /*
1100 * The "primitive" lock is _not_ held.
1101 *
1102 * knote "lock" is held
1103 */
1104 void
selspec_detach(struct knote * kn)1105 selspec_detach(struct knote *kn)
1106 {
1107 lck_spin_lock(&selspec_lock);
1108
1109 if (!KNOTE_IS_AUTODETACHED(kn)) {
1110 if (kn->kn_hook) {
1111 struct selinfo *sip = kn->kn_hook;
1112
1113 KNOTE_DETACH(&sip->si_note, kn);
1114 }
1115 }
1116
1117 kn->kn_hook = NULL;
1118
1119 lck_spin_unlock(&selspec_lock);
1120 }
1121
1122 /*
1123 * Select system call.
1124 *
1125 * Returns: 0 Success
1126 * EINVAL Invalid argument
1127 * EAGAIN Nonconformant error if allocation fails
1128 */
1129 int
select(struct proc * p,struct select_args * uap,int32_t * retval)1130 select(struct proc *p, struct select_args *uap, int32_t *retval)
1131 {
1132 __pthread_testcancel(1);
1133 return select_nocancel(p, (struct select_nocancel_args *)uap, retval);
1134 }
1135
1136 int
select_nocancel(struct proc * p,struct select_nocancel_args * uap,int32_t * retval)1137 select_nocancel(struct proc *p, struct select_nocancel_args *uap, int32_t *retval)
1138 {
1139 uint64_t timeout = 0;
1140
1141 if (uap->tv) {
1142 int err;
1143 struct timeval atv;
1144 if (IS_64BIT_PROCESS(p)) {
1145 struct user64_timeval atv64;
1146 err = copyin(uap->tv, (caddr_t)&atv64, sizeof(atv64));
1147 /* Loses resolution - assume timeout < 68 years */
1148 atv.tv_sec = (__darwin_time_t)atv64.tv_sec;
1149 atv.tv_usec = atv64.tv_usec;
1150 } else {
1151 struct user32_timeval atv32;
1152 err = copyin(uap->tv, (caddr_t)&atv32, sizeof(atv32));
1153 atv.tv_sec = atv32.tv_sec;
1154 atv.tv_usec = atv32.tv_usec;
1155 }
1156 if (err) {
1157 return err;
1158 }
1159
1160 if (itimerfix(&atv)) {
1161 err = EINVAL;
1162 return err;
1163 }
1164
1165 clock_absolutetime_interval_to_deadline(tvtoabstime(&atv), &timeout);
1166 }
1167
1168 return select_internal(p, uap, timeout, retval);
1169 }
1170
1171 int
pselect(struct proc * p,struct pselect_args * uap,int32_t * retval)1172 pselect(struct proc *p, struct pselect_args *uap, int32_t *retval)
1173 {
1174 __pthread_testcancel(1);
1175 return pselect_nocancel(p, (struct pselect_nocancel_args *)uap, retval);
1176 }
1177
1178 int
pselect_nocancel(struct proc * p,struct pselect_nocancel_args * uap,int32_t * retval)1179 pselect_nocancel(struct proc *p, struct pselect_nocancel_args *uap, int32_t *retval)
1180 {
1181 int err;
1182 struct uthread *ut;
1183 uint64_t timeout = 0;
1184
1185 if (uap->ts) {
1186 struct timespec ts;
1187
1188 if (IS_64BIT_PROCESS(p)) {
1189 struct user64_timespec ts64;
1190 err = copyin(uap->ts, (caddr_t)&ts64, sizeof(ts64));
1191 ts.tv_sec = (__darwin_time_t)ts64.tv_sec;
1192 ts.tv_nsec = (long)ts64.tv_nsec;
1193 } else {
1194 struct user32_timespec ts32;
1195 err = copyin(uap->ts, (caddr_t)&ts32, sizeof(ts32));
1196 ts.tv_sec = ts32.tv_sec;
1197 ts.tv_nsec = ts32.tv_nsec;
1198 }
1199 if (err) {
1200 return err;
1201 }
1202
1203 if (!timespec_is_valid(&ts)) {
1204 return EINVAL;
1205 }
1206 clock_absolutetime_interval_to_deadline(tstoabstime(&ts), &timeout);
1207 }
1208
1209 ut = current_uthread();
1210
1211 if (uap->mask != USER_ADDR_NULL) {
1212 /* save current mask, then copyin and set new mask */
1213 sigset_t newset;
1214 err = copyin(uap->mask, &newset, sizeof(sigset_t));
1215 if (err) {
1216 return err;
1217 }
1218 ut->uu_oldmask = ut->uu_sigmask;
1219 ut->uu_flag |= UT_SAS_OLDMASK;
1220 ut->uu_sigmask = (newset & ~sigcantmask);
1221 }
1222
1223 err = select_internal(p, (struct select_nocancel_args *)uap, timeout, retval);
1224
1225 if (err != EINTR && ut->uu_flag & UT_SAS_OLDMASK) {
1226 /*
1227 * Restore old mask (direct return case). NOTE: EINTR can also be returned
1228 * if the thread is cancelled. In that case, we don't reset the signal
1229 * mask to its original value (which usually happens in the signal
1230 * delivery path). This behavior is permitted by POSIX.
1231 */
1232 ut->uu_sigmask = ut->uu_oldmask;
1233 ut->uu_oldmask = 0;
1234 ut->uu_flag &= ~UT_SAS_OLDMASK;
1235 }
1236
1237 return err;
1238 }
1239
1240 void
select_cleanup_uthread(struct _select * sel)1241 select_cleanup_uthread(struct _select *sel)
1242 {
1243 kfree_data(sel->ibits, 2 * sel->nbytes);
1244 sel->ibits = sel->obits = NULL;
1245 sel->nbytes = 0;
1246 }
1247
1248 static int
select_grow_uthread_cache(struct _select * sel,uint32_t nbytes)1249 select_grow_uthread_cache(struct _select *sel, uint32_t nbytes)
1250 {
1251 uint32_t *buf;
1252
1253 buf = kalloc_data(2 * nbytes, Z_WAITOK | Z_ZERO);
1254 if (buf) {
1255 select_cleanup_uthread(sel);
1256 sel->ibits = buf;
1257 sel->obits = buf + nbytes / sizeof(uint32_t);
1258 sel->nbytes = nbytes;
1259 return true;
1260 }
1261 return false;
1262 }
1263
1264 static void
select_bzero_uthread_cache(struct _select * sel)1265 select_bzero_uthread_cache(struct _select *sel)
1266 {
1267 bzero(sel->ibits, sel->nbytes * 2);
1268 }
1269
1270 /*
1271 * Generic implementation of {,p}select. Care: we type-pun uap across the two
1272 * syscalls, which differ slightly. The first 4 arguments (nfds and the fd sets)
1273 * are identical. The 5th (timeout) argument points to different types, so we
1274 * unpack in the syscall-specific code, but the generic code still does a null
1275 * check on this argument to determine if a timeout was specified.
1276 */
1277 static int
select_internal(struct proc * p,struct select_nocancel_args * uap,uint64_t timeout,int32_t * retval)1278 select_internal(struct proc *p, struct select_nocancel_args *uap, uint64_t timeout, int32_t *retval)
1279 {
1280 struct uthread *uth = current_uthread();
1281 struct _select *sel = &uth->uu_select;
1282 struct _select_data *seldata = &uth->uu_save.uus_select_data;
1283 int error = 0;
1284 u_int ni, nw;
1285
1286 *retval = 0;
1287
1288 seldata->abstime = timeout;
1289 seldata->args = uap;
1290 seldata->retval = retval;
1291 seldata->count = 0;
1292
1293 if (uap->nd < 0) {
1294 return EINVAL;
1295 }
1296
1297 if (uap->nd > p->p_fd.fd_nfiles) {
1298 uap->nd = p->p_fd.fd_nfiles; /* forgiving; slightly wrong */
1299 }
1300 nw = howmany(uap->nd, NFDBITS);
1301 ni = nw * sizeof(fd_mask);
1302
1303 /*
1304 * if the previously allocated space for the bits is smaller than
1305 * what is requested or no space has yet been allocated for this
1306 * thread, allocate enough space now.
1307 *
1308 * Note: If this process fails, select() will return EAGAIN; this
1309 * is the same thing pool() returns in a no-memory situation, but
1310 * it is not a POSIX compliant error code for select().
1311 */
1312 if (sel->nbytes >= (3 * ni)) {
1313 select_bzero_uthread_cache(sel);
1314 } else if (!select_grow_uthread_cache(sel, 3 * ni)) {
1315 return EAGAIN;
1316 }
1317
1318 /*
1319 * get the bits from the user address space
1320 */
1321 #define getbits(name, x) \
1322 (uap->name ? copyin(uap->name, &sel->ibits[(x) * nw], ni) : 0)
1323
1324 if ((error = getbits(in, 0))) {
1325 return error;
1326 }
1327 if ((error = getbits(ou, 1))) {
1328 return error;
1329 }
1330 if ((error = getbits(ex, 2))) {
1331 return error;
1332 }
1333 #undef getbits
1334
1335 if ((error = selcount(p, sel->ibits, uap->nd, &seldata->count))) {
1336 return error;
1337 }
1338
1339 if (uth->uu_selset == NULL) {
1340 uth->uu_selset = select_set_alloc();
1341 }
1342 return selprocess(p, 0, SEL_FIRSTPASS);
1343 }
1344
1345 static int
selcontinue(int error)1346 selcontinue(int error)
1347 {
1348 return selprocess(current_proc(), error, SEL_SECONDPASS);
1349 }
1350
1351
1352 /*
1353 * selprocess
1354 *
1355 * Parameters: error The error code from our caller
1356 * sel_pass The pass we are on
1357 */
1358 int
selprocess(struct proc * p,int error,int sel_pass)1359 selprocess(struct proc *p, int error, int sel_pass)
1360 {
1361 struct uthread *uth = current_uthread();
1362 struct _select *sel = &uth->uu_select;
1363 struct _select_data *seldata = &uth->uu_save.uus_select_data;
1364 struct select_nocancel_args *uap = seldata->args;
1365 int *retval = seldata->retval;
1366
1367 int unwind = 1;
1368 int prepost = 0;
1369 int somewakeup = 0;
1370 int doretry = 0;
1371 wait_result_t wait_result;
1372
1373 if ((error != 0) && (sel_pass == SEL_FIRSTPASS)) {
1374 unwind = 0;
1375 }
1376 if (seldata->count == 0) {
1377 unwind = 0;
1378 }
1379 retry:
1380 if (error != 0) {
1381 goto done;
1382 }
1383
1384 OSBitOrAtomic(P_SELECT, &p->p_flag);
1385
1386 /* skip scans if the select is just for timeouts */
1387 if (seldata->count) {
1388 error = selscan(p, sel, seldata, uap->nd, retval, sel_pass,
1389 uth->uu_selset);
1390 if (error || *retval) {
1391 goto done;
1392 }
1393 if (prepost || somewakeup) {
1394 /*
1395 * if the select of log, then we can wakeup and
1396 * discover some one else already read the data;
1397 * go to select again if time permits
1398 */
1399 prepost = 0;
1400 somewakeup = 0;
1401 doretry = 1;
1402 }
1403 }
1404
1405 if (uap->tv) {
1406 uint64_t now;
1407
1408 clock_get_uptime(&now);
1409 if (now >= seldata->abstime) {
1410 goto done;
1411 }
1412 }
1413
1414 if (doretry) {
1415 /* cleanup obits and try again */
1416 doretry = 0;
1417 sel_pass = SEL_FIRSTPASS;
1418 goto retry;
1419 }
1420
1421 /*
1422 * To effect a poll, the timeout argument should be
1423 * non-nil, pointing to a zero-valued timeval structure.
1424 */
1425 if (uap->tv && seldata->abstime == 0) {
1426 goto done;
1427 }
1428
1429 /* No spurious wakeups due to colls,no need to check for them */
1430 if ((sel_pass == SEL_SECONDPASS) || ((p->p_flag & P_SELECT) == 0)) {
1431 sel_pass = SEL_FIRSTPASS;
1432 goto retry;
1433 }
1434
1435 OSBitAndAtomic(~((uint32_t)P_SELECT), &p->p_flag);
1436
1437 /* if the select is just for timeout skip check */
1438 if (seldata->count && (sel_pass == SEL_SECONDPASS)) {
1439 panic("selprocess: 2nd pass assertwaiting");
1440 }
1441
1442 wait_result = waitq_assert_wait64_leeway(uth->uu_selset,
1443 NO_EVENT64, THREAD_ABORTSAFE,
1444 TIMEOUT_URGENCY_USER_NORMAL,
1445 seldata->abstime,
1446 TIMEOUT_NO_LEEWAY);
1447 if (wait_result != THREAD_AWAKENED) {
1448 /* there are no preposted events */
1449 error = tsleep1(NULL, PSOCK | PCATCH,
1450 "select", 0, selcontinue);
1451 } else {
1452 prepost = 1;
1453 error = 0;
1454 }
1455
1456 if (error == 0) {
1457 sel_pass = SEL_SECONDPASS;
1458 if (!prepost) {
1459 somewakeup = 1;
1460 }
1461 goto retry;
1462 }
1463 done:
1464 if (unwind) {
1465 seldrop(p, sel->ibits, uap->nd, seldata->count);
1466 select_set_reset(uth->uu_selset);
1467 }
1468 OSBitAndAtomic(~((uint32_t)P_SELECT), &p->p_flag);
1469 /* select is not restarted after signals... */
1470 if (error == ERESTART) {
1471 error = EINTR;
1472 }
1473 if (error == EWOULDBLOCK) {
1474 error = 0;
1475 }
1476
1477 if (error == 0) {
1478 uint32_t nw = howmany(uap->nd, NFDBITS);
1479 uint32_t ni = nw * sizeof(fd_mask);
1480
1481 #define putbits(name, x) \
1482 (uap->name ? copyout(&sel->obits[(x) * nw], uap->name, ni) : 0)
1483 int e0 = putbits(in, 0);
1484 int e1 = putbits(ou, 1);
1485 int e2 = putbits(ex, 2);
1486
1487 error = e0 ?: e1 ?: e2;
1488 #undef putbits
1489 }
1490
1491 if (error != EINTR && sel_pass == SEL_SECONDPASS && uth->uu_flag & UT_SAS_OLDMASK) {
1492 /* restore signal mask - continuation case */
1493 uth->uu_sigmask = uth->uu_oldmask;
1494 uth->uu_oldmask = 0;
1495 uth->uu_flag &= ~UT_SAS_OLDMASK;
1496 }
1497
1498 return error;
1499 }
1500
1501
1502 /**
1503 * remove the fileproc's underlying waitq from the supplied waitq set;
1504 * clear FP_INSELECT when appropriate
1505 *
1506 * Parameters:
1507 * fp File proc that is potentially currently in select
1508 * selset Waitq set to which the fileproc may belong
1509 * (usually this is the thread's private waitq set)
1510 * Conditions:
1511 * proc_fdlock is held
1512 */
1513 static void
selunlinkfp(struct fileproc * fp,struct select_set * selset)1514 selunlinkfp(struct fileproc *fp, struct select_set *selset)
1515 {
1516 if (fp->fp_flags & FP_INSELECT) {
1517 if (fp->fp_guard_attrs) {
1518 if (fp->fp_guard->fpg_wset == selset) {
1519 fp->fp_guard->fpg_wset = NULL;
1520 fp->fp_flags &= ~FP_INSELECT;
1521 }
1522 } else {
1523 if (fp->fp_wset == selset) {
1524 fp->fp_wset = NULL;
1525 fp->fp_flags &= ~FP_INSELECT;
1526 }
1527 }
1528 }
1529 }
1530
1531 /**
1532 * connect a fileproc to the given selset, potentially bridging to a waitq
1533 * pointed to indirectly by wq_data
1534 *
1535 * Parameters:
1536 * fp File proc potentially currently in select
1537 * selset Waitq set to which the fileproc should now belong
1538 * (usually this is the thread's private waitq set)
1539 *
1540 * Conditions:
1541 * proc_fdlock is held
1542 */
1543 static void
sellinkfp(struct fileproc * fp,struct select_set * selset,waitq_link_t * linkp)1544 sellinkfp(struct fileproc *fp, struct select_set *selset, waitq_link_t *linkp)
1545 {
1546 if ((fp->fp_flags & FP_INSELECT) == 0) {
1547 if (fp->fp_guard_attrs) {
1548 fp->fp_guard->fpg_wset = selset;
1549 } else {
1550 fp->fp_wset = selset;
1551 }
1552 fp->fp_flags |= FP_INSELECT;
1553 } else {
1554 fp->fp_flags |= FP_SELCONFLICT;
1555 if (linkp->wqlh == NULL) {
1556 *linkp = waitq_link_alloc(WQT_SELECT_SET);
1557 }
1558 select_set_link(&select_conflict_queue, selset, linkp);
1559 }
1560 }
1561
1562
1563 /*
1564 * selscan
1565 *
1566 * Parameters: p Process performing the select
1567 * sel The per-thread select context structure
1568 * nfd The number of file descriptors to scan
1569 * retval The per thread system call return area
1570 * sel_pass Which pass this is; allowed values are
1571 * SEL_FIRSTPASS and SEL_SECONDPASS
1572 * selset The per thread wait queue set
1573 *
1574 * Returns: 0 Success
1575 * EIO Invalid p->p_fd field XXX Obsolete?
1576 * EBADF One of the files in the bit vector is
1577 * invalid.
1578 */
1579 static int
selscan(struct proc * p,struct _select * sel,struct _select_data * seldata,int nfd,int32_t * retval,int sel_pass,struct select_set * selset)1580 selscan(struct proc *p, struct _select *sel, struct _select_data * seldata,
1581 int nfd, int32_t *retval, int sel_pass, struct select_set *selset)
1582 {
1583 int msk, i, j, fd;
1584 u_int32_t bits;
1585 struct fileproc *fp;
1586 int n = 0; /* count of bits */
1587 int nc = 0; /* bit vector offset (nc'th bit) */
1588 static int flag[3] = { FREAD, FWRITE, 0 };
1589 u_int32_t *iptr, *optr;
1590 u_int nw;
1591 u_int32_t *ibits, *obits;
1592 int count;
1593 struct vfs_context context = {
1594 .vc_thread = current_thread(),
1595 };
1596 waitq_link_t link = WQL_NULL;
1597 void *s_data;
1598
1599 ibits = sel->ibits;
1600 obits = sel->obits;
1601
1602 nw = howmany(nfd, NFDBITS);
1603
1604 count = seldata->count;
1605
1606 nc = 0;
1607 if (!count) {
1608 *retval = 0;
1609 return 0;
1610 }
1611
1612 if (sel_pass == SEL_FIRSTPASS) {
1613 /*
1614 * Make sure the waitq-set is all clean:
1615 *
1616 * select loops until it finds at least one event, however it
1617 * doesn't mean that the event that woke up select is still
1618 * fired by the time the second pass runs, and then
1619 * select_internal will loop back to a first pass.
1620 */
1621 select_set_reset(selset);
1622 s_data = &link;
1623 } else {
1624 s_data = NULL;
1625 }
1626
1627 proc_fdlock(p);
1628 for (msk = 0; msk < 3; msk++) {
1629 iptr = (u_int32_t *)&ibits[msk * nw];
1630 optr = (u_int32_t *)&obits[msk * nw];
1631
1632 for (i = 0; i < nfd; i += NFDBITS) {
1633 bits = iptr[i / NFDBITS];
1634
1635 while ((j = ffs(bits)) && (fd = i + --j) < nfd) {
1636 bits &= ~(1U << j);
1637
1638 fp = fp_get_noref_locked(p, fd);
1639 if (fp == NULL) {
1640 /*
1641 * If we abort because of a bad
1642 * fd, let the caller unwind...
1643 */
1644 proc_fdunlock(p);
1645 return EBADF;
1646 }
1647 if (sel_pass == SEL_SECONDPASS) {
1648 selunlinkfp(fp, selset);
1649 } else if (link.wqlh == NULL) {
1650 link = waitq_link_alloc(WQT_SELECT_SET);
1651 }
1652
1653 context.vc_ucred = fp->f_cred;
1654
1655 /* The select; set the bit, if true */
1656 if (fo_select(fp, flag[msk], s_data, &context)) {
1657 optr[fd / NFDBITS] |= (1U << (fd % NFDBITS));
1658 n++;
1659 }
1660 if (sel_pass == SEL_FIRSTPASS) {
1661 /*
1662 * Hook up the thread's waitq set either to
1663 * the fileproc structure, or to the global
1664 * conflict queue: but only on the first
1665 * select pass.
1666 */
1667 sellinkfp(fp, selset, &link);
1668 }
1669 nc++;
1670 }
1671 }
1672 }
1673 proc_fdunlock(p);
1674
1675 if (link.wqlh) {
1676 waitq_link_free(WQT_SELECT_SET, link);
1677 }
1678
1679 *retval = n;
1680 return 0;
1681 }
1682
1683 static int poll_callback(struct kevent_qos_s *, kevent_ctx_t);
1684
1685 int
poll(struct proc * p,struct poll_args * uap,int32_t * retval)1686 poll(struct proc *p, struct poll_args *uap, int32_t *retval)
1687 {
1688 __pthread_testcancel(1);
1689 return poll_nocancel(p, (struct poll_nocancel_args *)uap, retval);
1690 }
1691
1692
1693 int
poll_nocancel(struct proc * p,struct poll_nocancel_args * uap,int32_t * retval)1694 poll_nocancel(struct proc *p, struct poll_nocancel_args *uap, int32_t *retval)
1695 {
1696 struct pollfd *fds = NULL;
1697 struct kqueue *kq = NULL;
1698 int error = 0;
1699 u_int nfds = uap->nfds;
1700 u_int rfds = 0;
1701 rlim_t nofile = proc_limitgetcur(p, RLIMIT_NOFILE);
1702 size_t ni = nfds * sizeof(struct pollfd);
1703
1704 /*
1705 * This is kinda bogus. We have fd limits, but that is not
1706 * really related to the size of the pollfd array. Make sure
1707 * we let the process use at least FD_SETSIZE entries and at
1708 * least enough for the current limits. We want to be reasonably
1709 * safe, but not overly restrictive.
1710 */
1711 if (nfds > OPEN_MAX ||
1712 (nfds > nofile && (proc_suser(p) || nfds > FD_SETSIZE))) {
1713 return EINVAL;
1714 }
1715
1716 kq = kqueue_alloc(p);
1717 if (kq == NULL) {
1718 return EAGAIN;
1719 }
1720
1721 if (nfds) {
1722 fds = (struct pollfd *)kalloc_data(ni, Z_WAITOK);
1723 if (NULL == fds) {
1724 error = EAGAIN;
1725 goto out;
1726 }
1727
1728 error = copyin(uap->fds, fds, nfds * sizeof(struct pollfd));
1729 if (error) {
1730 goto out;
1731 }
1732 }
1733
1734 /* JMM - all this P_SELECT stuff is bogus */
1735 OSBitOrAtomic(P_SELECT, &p->p_flag);
1736 for (u_int i = 0; i < nfds; i++) {
1737 short events = fds[i].events;
1738 __assert_only int rc;
1739
1740 /* per spec, ignore fd values below zero */
1741 if (fds[i].fd < 0) {
1742 fds[i].revents = 0;
1743 continue;
1744 }
1745
1746 /* convert the poll event into a kqueue kevent */
1747 struct kevent_qos_s kev = {
1748 .ident = fds[i].fd,
1749 .flags = EV_ADD | EV_ONESHOT | EV_POLL,
1750 .udata = CAST_USER_ADDR_T(&fds[i])
1751 };
1752
1753 /* Handle input events */
1754 if (events & (POLLIN | POLLRDNORM | POLLPRI | POLLRDBAND | POLLHUP)) {
1755 kev.filter = EVFILT_READ;
1756 if (events & (POLLPRI | POLLRDBAND)) {
1757 kev.flags |= EV_OOBAND;
1758 }
1759 rc = kevent_register(kq, &kev, NULL);
1760 assert((rc & FILTER_REGISTER_WAIT) == 0);
1761 }
1762
1763 /* Handle output events */
1764 if ((kev.flags & EV_ERROR) == 0 &&
1765 (events & (POLLOUT | POLLWRNORM | POLLWRBAND))) {
1766 kev.filter = EVFILT_WRITE;
1767 rc = kevent_register(kq, &kev, NULL);
1768 assert((rc & FILTER_REGISTER_WAIT) == 0);
1769 }
1770
1771 /* Handle BSD extension vnode events */
1772 if ((kev.flags & EV_ERROR) == 0 &&
1773 (events & (POLLEXTEND | POLLATTRIB | POLLNLINK | POLLWRITE))) {
1774 kev.filter = EVFILT_VNODE;
1775 kev.fflags = 0;
1776 if (events & POLLEXTEND) {
1777 kev.fflags |= NOTE_EXTEND;
1778 }
1779 if (events & POLLATTRIB) {
1780 kev.fflags |= NOTE_ATTRIB;
1781 }
1782 if (events & POLLNLINK) {
1783 kev.fflags |= NOTE_LINK;
1784 }
1785 if (events & POLLWRITE) {
1786 kev.fflags |= NOTE_WRITE;
1787 }
1788 rc = kevent_register(kq, &kev, NULL);
1789 assert((rc & FILTER_REGISTER_WAIT) == 0);
1790 }
1791
1792 if (kev.flags & EV_ERROR) {
1793 fds[i].revents = POLLNVAL;
1794 rfds++;
1795 } else {
1796 fds[i].revents = 0;
1797 }
1798 }
1799
1800 /*
1801 * Did we have any trouble registering?
1802 * If user space passed 0 FDs, then respect any timeout value passed.
1803 * This is an extremely inefficient sleep. If user space passed one or
1804 * more FDs, and we had trouble registering _all_ of them, then bail
1805 * out. If a subset of the provided FDs failed to register, then we
1806 * will still call the kqueue_scan function.
1807 */
1808 if (nfds && (rfds == nfds)) {
1809 goto done;
1810 }
1811
1812 /* scan for, and possibly wait for, the kevents to trigger */
1813 kevent_ctx_t kectx = kevent_get_context(current_thread());
1814 *kectx = (struct kevent_ctx_s){
1815 .kec_process_noutputs = rfds,
1816 .kec_process_flags = KEVENT_FLAG_POLL,
1817 .kec_deadline = 0, /* wait forever */
1818 };
1819
1820 /*
1821 * If any events have trouble registering, an event has fired and we
1822 * shouldn't wait for events in kqueue_scan.
1823 */
1824 if (rfds) {
1825 kectx->kec_process_flags |= KEVENT_FLAG_IMMEDIATE;
1826 } else if (uap->timeout != -1) {
1827 clock_interval_to_deadline(uap->timeout, NSEC_PER_MSEC,
1828 &kectx->kec_deadline);
1829 }
1830
1831 error = kqueue_scan(kq, kectx->kec_process_flags, kectx, poll_callback);
1832 rfds = kectx->kec_process_noutputs;
1833
1834 done:
1835 OSBitAndAtomic(~((uint32_t)P_SELECT), &p->p_flag);
1836 /* poll is not restarted after signals... */
1837 if (error == ERESTART) {
1838 error = EINTR;
1839 }
1840 if (error == 0) {
1841 error = copyout(fds, uap->fds, nfds * sizeof(struct pollfd));
1842 *retval = rfds;
1843 }
1844
1845 out:
1846 kfree_data(fds, ni);
1847
1848 kqueue_dealloc(kq);
1849 return error;
1850 }
1851
1852 static int
poll_callback(struct kevent_qos_s * kevp,kevent_ctx_t kectx)1853 poll_callback(struct kevent_qos_s *kevp, kevent_ctx_t kectx)
1854 {
1855 struct pollfd *fds = CAST_DOWN(struct pollfd *, kevp->udata);
1856 short prev_revents = fds->revents;
1857 short mask = 0;
1858
1859 /* convert the results back into revents */
1860 if (kevp->flags & EV_EOF) {
1861 fds->revents |= POLLHUP;
1862 }
1863 if (kevp->flags & EV_ERROR) {
1864 fds->revents |= POLLERR;
1865 }
1866
1867 switch (kevp->filter) {
1868 case EVFILT_READ:
1869 if (fds->revents & POLLHUP) {
1870 mask = (POLLIN | POLLRDNORM | POLLPRI | POLLRDBAND);
1871 } else {
1872 mask = (POLLIN | POLLRDNORM);
1873 if (kevp->flags & EV_OOBAND) {
1874 mask |= (POLLPRI | POLLRDBAND);
1875 }
1876 }
1877 fds->revents |= (fds->events & mask);
1878 break;
1879
1880 case EVFILT_WRITE:
1881 if (!(fds->revents & POLLHUP)) {
1882 fds->revents |= (fds->events & (POLLOUT | POLLWRNORM | POLLWRBAND));
1883 }
1884 break;
1885
1886 case EVFILT_VNODE:
1887 if (kevp->fflags & NOTE_EXTEND) {
1888 fds->revents |= (fds->events & POLLEXTEND);
1889 }
1890 if (kevp->fflags & NOTE_ATTRIB) {
1891 fds->revents |= (fds->events & POLLATTRIB);
1892 }
1893 if (kevp->fflags & NOTE_LINK) {
1894 fds->revents |= (fds->events & POLLNLINK);
1895 }
1896 if (kevp->fflags & NOTE_WRITE) {
1897 fds->revents |= (fds->events & POLLWRITE);
1898 }
1899 break;
1900 }
1901
1902 if (fds->revents != 0 && prev_revents == 0) {
1903 kectx->kec_process_noutputs++;
1904 }
1905
1906 return 0;
1907 }
1908
1909 int
seltrue(__unused dev_t dev,__unused int flag,__unused struct proc * p)1910 seltrue(__unused dev_t dev, __unused int flag, __unused struct proc *p)
1911 {
1912 return 1;
1913 }
1914
1915 /*
1916 * selcount
1917 *
1918 * Count the number of bits set in the input bit vector, and establish an
1919 * outstanding fp->fp_iocount for each of the descriptors which will be in
1920 * use in the select operation.
1921 *
1922 * Parameters: p The process doing the select
1923 * ibits The input bit vector
1924 * nfd The number of fd's in the vector
1925 * countp Pointer to where to store the bit count
1926 *
1927 * Returns: 0 Success
1928 * EIO Bad per process open file table
1929 * EBADF One of the bits in the input bit vector
1930 * references an invalid fd
1931 *
1932 * Implicit: *countp (modified) Count of fd's
1933 *
1934 * Notes: This function is the first pass under the proc_fdlock() that
1935 * permits us to recognize invalid descriptors in the bit vector;
1936 * the may, however, not remain valid through the drop and
1937 * later reacquisition of the proc_fdlock().
1938 */
1939 static int
selcount(struct proc * p,u_int32_t * ibits,int nfd,int * countp)1940 selcount(struct proc *p, u_int32_t *ibits, int nfd, int *countp)
1941 {
1942 int msk, i, j, fd;
1943 u_int32_t bits;
1944 struct fileproc *fp;
1945 int n = 0;
1946 u_int32_t *iptr;
1947 u_int nw;
1948 int error = 0;
1949 int need_wakeup = 0;
1950
1951 nw = howmany(nfd, NFDBITS);
1952
1953 proc_fdlock(p);
1954 for (msk = 0; msk < 3; msk++) {
1955 iptr = (u_int32_t *)&ibits[msk * nw];
1956 for (i = 0; i < nfd; i += NFDBITS) {
1957 bits = iptr[i / NFDBITS];
1958 while ((j = ffs(bits)) && (fd = i + --j) < nfd) {
1959 bits &= ~(1U << j);
1960
1961 fp = fp_get_noref_locked(p, fd);
1962 if (fp == NULL) {
1963 *countp = 0;
1964 error = EBADF;
1965 goto bad;
1966 }
1967 os_ref_retain_locked(&fp->fp_iocount);
1968 n++;
1969 }
1970 }
1971 }
1972 proc_fdunlock(p);
1973
1974 *countp = n;
1975 return 0;
1976
1977 bad:
1978 if (n == 0) {
1979 goto out;
1980 }
1981 /* Ignore error return; it's already EBADF */
1982 (void)seldrop_locked(p, ibits, nfd, n, &need_wakeup);
1983
1984 out:
1985 proc_fdunlock(p);
1986 if (need_wakeup) {
1987 wakeup(&p->p_fd.fd_fpdrainwait);
1988 }
1989 return error;
1990 }
1991
1992
1993 /*
1994 * seldrop_locked
1995 *
1996 * Drop outstanding wait queue references set up during selscan(); drop the
1997 * outstanding per fileproc fp_iocount picked up during the selcount().
1998 *
1999 * Parameters: p Process performing the select
2000 * ibits Input bit bector of fd's
2001 * nfd Number of fd's
2002 * lim Limit to number of vector entries to
2003 * consider, or -1 for "all"
2004 * inselect True if
2005 * need_wakeup Pointer to flag to set to do a wakeup
2006 * if f_iocont on any descriptor goes to 0
2007 *
2008 * Returns: 0 Success
2009 * EBADF One or more fds in the bit vector
2010 * were invalid, but the rest
2011 * were successfully dropped
2012 *
2013 * Notes: An fd make become bad while the proc_fdlock() is not held,
2014 * if a multithreaded application closes the fd out from under
2015 * the in progress select. In this case, we still have to
2016 * clean up after the set up on the remaining fds.
2017 */
2018 static int
seldrop_locked(struct proc * p,u_int32_t * ibits,int nfd,int lim,int * need_wakeup)2019 seldrop_locked(struct proc *p, u_int32_t *ibits, int nfd, int lim, int *need_wakeup)
2020 {
2021 int msk, i, j, nc, fd;
2022 u_int32_t bits;
2023 struct fileproc *fp;
2024 u_int32_t *iptr;
2025 u_int nw;
2026 int error = 0;
2027 uthread_t uth = current_uthread();
2028 struct _select_data *seldata;
2029
2030 *need_wakeup = 0;
2031
2032 nw = howmany(nfd, NFDBITS);
2033 seldata = &uth->uu_save.uus_select_data;
2034
2035 nc = 0;
2036 for (msk = 0; msk < 3; msk++) {
2037 iptr = (u_int32_t *)&ibits[msk * nw];
2038 for (i = 0; i < nfd; i += NFDBITS) {
2039 bits = iptr[i / NFDBITS];
2040 while ((j = ffs(bits)) && (fd = i + --j) < nfd) {
2041 bits &= ~(1U << j);
2042 /*
2043 * If we've already dropped as many as were
2044 * counted/scanned, then we are done.
2045 */
2046 if (nc >= lim) {
2047 goto done;
2048 }
2049
2050 /*
2051 * We took an I/O reference in selcount,
2052 * so the fp can't possibly be NULL.
2053 */
2054 fp = fp_get_noref_locked_with_iocount(p, fd);
2055 selunlinkfp(fp, uth->uu_selset);
2056
2057 nc++;
2058
2059 const os_ref_count_t refc = os_ref_release_locked(&fp->fp_iocount);
2060 if (0 == refc) {
2061 panic("fp_iocount overdecrement!");
2062 }
2063
2064 if (1 == refc) {
2065 /*
2066 * The last iocount is responsible for clearing
2067 * selconfict flag - even if we didn't set it -
2068 * and is also responsible for waking up anyone
2069 * waiting on iocounts to drain.
2070 */
2071 if (fp->fp_flags & FP_SELCONFLICT) {
2072 fp->fp_flags &= ~FP_SELCONFLICT;
2073 }
2074 if (p->p_fd.fd_fpdrainwait) {
2075 p->p_fd.fd_fpdrainwait = 0;
2076 *need_wakeup = 1;
2077 }
2078 }
2079 }
2080 }
2081 }
2082 done:
2083 return error;
2084 }
2085
2086
2087 static int
seldrop(struct proc * p,u_int32_t * ibits,int nfd,int lim)2088 seldrop(struct proc *p, u_int32_t *ibits, int nfd, int lim)
2089 {
2090 int error;
2091 int need_wakeup = 0;
2092
2093 proc_fdlock(p);
2094 error = seldrop_locked(p, ibits, nfd, lim, &need_wakeup);
2095 proc_fdunlock(p);
2096 if (need_wakeup) {
2097 wakeup(&p->p_fd.fd_fpdrainwait);
2098 }
2099 return error;
2100 }
2101
2102 /*
2103 * Record a select request.
2104 */
2105 void
selrecord(__unused struct proc * selector,struct selinfo * sip,void * s_data)2106 selrecord(__unused struct proc *selector, struct selinfo *sip, void *s_data)
2107 {
2108 struct select_set *selset = current_uthread()->uu_selset;
2109
2110 /* do not record if this is second pass of select */
2111 if (!s_data) {
2112 return;
2113 }
2114
2115 if (selset == SELSPEC_RECORD_MARKER) {
2116 /*
2117 * The kevent subsystem is trying to sniff
2118 * the selinfo::si_note to attach to.
2119 */
2120 ((selspec_record_hook_t)s_data)(sip);
2121 } else {
2122 waitq_link_t *linkp = s_data;
2123
2124 if (!waitq_is_valid(&sip->si_waitq)) {
2125 waitq_init(&sip->si_waitq, WQT_SELECT, SYNC_POLICY_FIFO);
2126 }
2127
2128 /* note: this checks for pre-existing linkage */
2129 select_set_link(&sip->si_waitq, selset, linkp);
2130 }
2131 }
2132
2133 static void
selwakeup_internal(struct selinfo * sip,long hint,wait_result_t wr)2134 selwakeup_internal(struct selinfo *sip, long hint, wait_result_t wr)
2135 {
2136 if (sip->si_flags & SI_SELSPEC) {
2137 /*
2138 * The "primitive" lock is held.
2139 * The knote lock is not held.
2140 *
2141 * All knotes will transition their kn_hook to NULL and we will
2142 * reeinitialize the primitive's klist
2143 */
2144 lck_spin_lock(&selspec_lock);
2145 knote(&sip->si_note, hint, /*autodetach=*/ true);
2146 lck_spin_unlock(&selspec_lock);
2147 sip->si_flags &= ~SI_SELSPEC;
2148 }
2149
2150 /*
2151 * After selrecord() has been called, selinfo owners must call
2152 * at least one of selwakeup() or selthreadclear().
2153 *
2154 * Use this opportunity to deinit the waitq
2155 * so that all linkages are garbage collected
2156 * in a combined wakeup-all + unlink + deinit call.
2157 */
2158 select_waitq_wakeup_and_deinit(&sip->si_waitq, NO_EVENT64, wr);
2159 }
2160
2161
2162 void
selwakeup(struct selinfo * sip)2163 selwakeup(struct selinfo *sip)
2164 {
2165 selwakeup_internal(sip, 0, THREAD_AWAKENED);
2166 }
2167
2168 void
selthreadclear(struct selinfo * sip)2169 selthreadclear(struct selinfo *sip)
2170 {
2171 selwakeup_internal(sip, NOTE_REVOKE, THREAD_RESTART);
2172 }
2173
2174
2175 /*
2176 * gethostuuid
2177 *
2178 * Description: Get the host UUID from IOKit and return it to user space.
2179 *
2180 * Parameters: uuid_buf Pointer to buffer to receive UUID
2181 * timeout Timespec for timout
2182 *
2183 * Returns: 0 Success
2184 * EWOULDBLOCK Timeout is too short
2185 * copyout:EFAULT Bad user buffer
2186 * mac_system_check_info:EPERM Client not allowed to perform this operation
2187 *
2188 * Notes: A timeout seems redundant, since if it's tolerable to not
2189 * have a system UUID in hand, then why ask for one?
2190 */
2191 int
gethostuuid(struct proc * p,struct gethostuuid_args * uap,__unused int32_t * retval)2192 gethostuuid(struct proc *p, struct gethostuuid_args *uap, __unused int32_t *retval)
2193 {
2194 kern_return_t kret;
2195 int error;
2196 mach_timespec_t mach_ts; /* for IOKit call */
2197 __darwin_uuid_t uuid_kern = {}; /* for IOKit call */
2198
2199 /* Check entitlement */
2200 if (!IOCurrentTaskHasEntitlement("com.apple.private.getprivatesysid")) {
2201 #if !defined(XNU_TARGET_OS_OSX)
2202 #if CONFIG_MACF
2203 if ((error = mac_system_check_info(kauth_cred_get(), "hw.uuid")) != 0) {
2204 /* EPERM invokes userspace upcall if present */
2205 return error;
2206 }
2207 #endif
2208 #endif
2209 }
2210
2211 /* Convert the 32/64 bit timespec into a mach_timespec_t */
2212 if (proc_is64bit(p)) {
2213 struct user64_timespec ts;
2214 error = copyin(uap->timeoutp, &ts, sizeof(ts));
2215 if (error) {
2216 return error;
2217 }
2218 mach_ts.tv_sec = (unsigned int)ts.tv_sec;
2219 mach_ts.tv_nsec = (clock_res_t)ts.tv_nsec;
2220 } else {
2221 struct user32_timespec ts;
2222 error = copyin(uap->timeoutp, &ts, sizeof(ts));
2223 if (error) {
2224 return error;
2225 }
2226 mach_ts.tv_sec = ts.tv_sec;
2227 mach_ts.tv_nsec = ts.tv_nsec;
2228 }
2229
2230 /* Call IOKit with the stack buffer to get the UUID */
2231 kret = IOBSDGetPlatformUUID(uuid_kern, mach_ts);
2232
2233 /*
2234 * If we get it, copy out the data to the user buffer; note that a
2235 * uuid_t is an array of characters, so this is size invariant for
2236 * 32 vs. 64 bit.
2237 */
2238 if (kret == KERN_SUCCESS) {
2239 error = copyout(uuid_kern, uap->uuid_buf, sizeof(uuid_kern));
2240 } else {
2241 error = EWOULDBLOCK;
2242 }
2243
2244 return error;
2245 }
2246
2247 /*
2248 * ledger
2249 *
2250 * Description: Omnibus system call for ledger operations
2251 */
2252 int
ledger(struct proc * p,struct ledger_args * args,__unused int32_t * retval)2253 ledger(struct proc *p, struct ledger_args *args, __unused int32_t *retval)
2254 {
2255 #if !CONFIG_MACF
2256 #pragma unused(p)
2257 #endif
2258 int rval, pid, len, error;
2259 #ifdef LEDGER_DEBUG
2260 struct ledger_limit_args lla;
2261 #endif
2262 task_t task;
2263 proc_t proc;
2264
2265 /* Finish copying in the necessary args before taking the proc lock */
2266 error = 0;
2267 len = 0;
2268 if (args->cmd == LEDGER_ENTRY_INFO) {
2269 error = copyin(args->arg3, (char *)&len, sizeof(len));
2270 } else if (args->cmd == LEDGER_TEMPLATE_INFO) {
2271 error = copyin(args->arg2, (char *)&len, sizeof(len));
2272 } else if (args->cmd == LEDGER_LIMIT)
2273 #ifdef LEDGER_DEBUG
2274 { error = copyin(args->arg2, (char *)&lla, sizeof(lla));}
2275 #else
2276 { return EINVAL; }
2277 #endif
2278 else if ((args->cmd < 0) || (args->cmd > LEDGER_MAX_CMD)) {
2279 return EINVAL;
2280 }
2281
2282 if (error) {
2283 return error;
2284 }
2285 if (len < 0) {
2286 return EINVAL;
2287 }
2288
2289 rval = 0;
2290 if (args->cmd != LEDGER_TEMPLATE_INFO) {
2291 pid = (int)args->arg1;
2292 proc = proc_find(pid);
2293 if (proc == NULL) {
2294 return ESRCH;
2295 }
2296
2297 #if CONFIG_MACF
2298 error = mac_proc_check_ledger(p, proc, args->cmd);
2299 if (error) {
2300 proc_rele(proc);
2301 return error;
2302 }
2303 #endif
2304
2305 task = proc_task(proc);
2306 }
2307
2308 switch (args->cmd) {
2309 #ifdef LEDGER_DEBUG
2310 case LEDGER_LIMIT: {
2311 if (!kauth_cred_issuser(kauth_cred_get())) {
2312 rval = EPERM;
2313 }
2314 rval = ledger_limit(task, &lla);
2315 proc_rele(proc);
2316 break;
2317 }
2318 #endif
2319 case LEDGER_INFO: {
2320 struct ledger_info info = {};
2321
2322 rval = ledger_info(task, &info);
2323 proc_rele(proc);
2324 if (rval == 0) {
2325 rval = copyout(&info, args->arg2,
2326 sizeof(info));
2327 }
2328 break;
2329 }
2330
2331 case LEDGER_ENTRY_INFO: {
2332 void *buf;
2333 int sz;
2334
2335 #if CONFIG_MEMORYSTATUS
2336 task_ledger_settle_dirty_time(task);
2337 #endif /* CONFIG_MEMORYSTATUS */
2338
2339 rval = ledger_get_task_entry_info_multiple(task, &buf, &len);
2340 proc_rele(proc);
2341 if ((rval == 0) && (len >= 0)) {
2342 sz = len * sizeof(struct ledger_entry_info);
2343 rval = copyout(buf, args->arg2, sz);
2344 kfree_data(buf, sz);
2345 }
2346 if (rval == 0) {
2347 rval = copyout(&len, args->arg3, sizeof(len));
2348 }
2349 break;
2350 }
2351
2352 case LEDGER_TEMPLATE_INFO: {
2353 void *buf;
2354 int sz;
2355
2356 rval = ledger_template_info(&buf, &len);
2357 if ((rval == 0) && (len >= 0)) {
2358 sz = len * sizeof(struct ledger_template_info);
2359 rval = copyout(buf, args->arg1, sz);
2360 kfree_data(buf, sz);
2361 }
2362 if (rval == 0) {
2363 rval = copyout(&len, args->arg2, sizeof(len));
2364 }
2365 break;
2366 }
2367
2368 default:
2369 panic("ledger syscall logic error -- command type %d", args->cmd);
2370 proc_rele(proc);
2371 rval = EINVAL;
2372 }
2373
2374 return rval;
2375 }
2376
2377 int
telemetry(__unused struct proc * p,struct telemetry_args * args,__unused int32_t * retval)2378 telemetry(__unused struct proc *p, struct telemetry_args *args, __unused int32_t *retval)
2379 {
2380 int error = 0;
2381
2382 switch (args->cmd) {
2383 #if CONFIG_TELEMETRY
2384 case TELEMETRY_CMD_TIMER_EVENT:
2385 error = telemetry_timer_event(args->deadline, args->interval, args->leeway);
2386 break;
2387 case TELEMETRY_CMD_PMI_SETUP:
2388 error = telemetry_pmi_setup((enum telemetry_pmi)args->deadline, args->interval);
2389 break;
2390 #endif /* CONFIG_TELEMETRY */
2391 case TELEMETRY_CMD_VOUCHER_NAME:
2392 if (thread_set_voucher_name((mach_port_name_t)args->deadline)) {
2393 error = EINVAL;
2394 }
2395 break;
2396
2397 default:
2398 error = EINVAL;
2399 break;
2400 }
2401
2402 return error;
2403 }
2404
2405 /*
2406 * Logging
2407 *
2408 * Description: syscall to access kernel logging from userspace
2409 *
2410 * Args:
2411 * tag - used for syncing with userspace on the version.
2412 * flags - flags used by the syscall.
2413 * buffer - userspace address of string to copy.
2414 * size - size of buffer.
2415 */
2416 int
log_data(__unused struct proc * p,struct log_data_args * args,int * retval)2417 log_data(__unused struct proc *p, struct log_data_args *args, int *retval)
2418 {
2419 unsigned int tag = args->tag;
2420 unsigned int flags = args->flags;
2421 user_addr_t buffer = args->buffer;
2422 unsigned int size = args->size;
2423 int ret = 0;
2424 *retval = 0;
2425
2426 /* Only DEXTs are suppose to use this syscall. */
2427 if (!task_is_driver(current_task())) {
2428 return EPERM;
2429 }
2430
2431 /*
2432 * Tag synchronize the syscall version with userspace.
2433 * Tag == 0 => flags == OS_LOG_TYPE
2434 */
2435 if (tag != 0) {
2436 return EINVAL;
2437 }
2438
2439 /*
2440 * OS_LOG_TYPE are defined in libkern/os/log.h
2441 * In userspace they are defined in libtrace/os/log.h
2442 */
2443 if (flags != OS_LOG_TYPE_DEFAULT &&
2444 flags != OS_LOG_TYPE_INFO &&
2445 flags != OS_LOG_TYPE_DEBUG &&
2446 flags != OS_LOG_TYPE_ERROR &&
2447 flags != OS_LOG_TYPE_FAULT) {
2448 return EINVAL;
2449 }
2450
2451 if (size == 0) {
2452 return EINVAL;
2453 }
2454
2455 /* truncate to OS_LOG_DATA_MAX_SIZE */
2456 if (size > OS_LOG_DATA_MAX_SIZE) {
2457 printf("%s: WARNING msg is going to be truncated from %u to %u\n",
2458 __func__, size, OS_LOG_DATA_MAX_SIZE);
2459 size = OS_LOG_DATA_MAX_SIZE;
2460 }
2461
2462 char *log_msg = (char *)kalloc_data(size, Z_WAITOK);
2463 if (!log_msg) {
2464 return ENOMEM;
2465 }
2466
2467 if (copyin(buffer, log_msg, size) != 0) {
2468 ret = EFAULT;
2469 goto out;
2470 }
2471 log_msg[size - 1] = '\0';
2472
2473 /*
2474 * This will log to dmesg and logd.
2475 * The call will fail if the current
2476 * process is not a driverKit process.
2477 */
2478 os_log_driverKit(&ret, OS_LOG_DEFAULT, (os_log_type_t)flags, "%s", log_msg);
2479
2480 out:
2481 if (log_msg != NULL) {
2482 kfree_data(log_msg, size);
2483 }
2484
2485 return ret;
2486 }
2487
2488 #if DEVELOPMENT || DEBUG
2489
2490 static int
2491 sysctl_mpsc_test_pingpong SYSCTL_HANDLER_ARGS
2492 {
2493 #pragma unused(oidp, arg1, arg2)
2494 uint64_t value = 0;
2495 int error;
2496
2497 error = SYSCTL_IN(req, &value, sizeof(value));
2498 if (error) {
2499 return error;
2500 }
2501
2502 if (error == 0 && req->newptr) {
2503 error = mpsc_test_pingpong(value, &value);
2504 if (error == 0) {
2505 error = SYSCTL_OUT(req, &value, sizeof(value));
2506 }
2507 }
2508
2509 return error;
2510 }
2511 SYSCTL_PROC(_kern, OID_AUTO, mpsc_test_pingpong, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
2512 0, 0, sysctl_mpsc_test_pingpong, "Q", "MPSC tests: pingpong");
2513
2514 #endif /* DEVELOPMENT || DEBUG */
2515
2516 /* Telemetry, microstackshots */
2517
2518 SYSCTL_NODE(_kern, OID_AUTO, microstackshot, CTLFLAG_RD | CTLFLAG_LOCKED, 0,
2519 "microstackshot info");
2520
2521 extern uint32_t telemetry_sample_rate;
2522 SYSCTL_UINT(_kern_microstackshot, OID_AUTO, interrupt_sample_rate,
2523 CTLFLAG_RD | CTLFLAG_LOCKED, &telemetry_sample_rate, 0,
2524 "interrupt-based sampling rate in Hz");
2525
2526 #if defined(MT_CORE_INSTRS) && defined(MT_CORE_CYCLES)
2527
2528 extern uint64_t mt_microstackshot_period;
2529 SYSCTL_QUAD(_kern_microstackshot, OID_AUTO, pmi_sample_period,
2530 CTLFLAG_RD | CTLFLAG_LOCKED, &mt_microstackshot_period,
2531 "PMI sampling rate");
2532 extern unsigned int mt_microstackshot_ctr;
2533 SYSCTL_UINT(_kern_microstackshot, OID_AUTO, pmi_sample_counter,
2534 CTLFLAG_RD | CTLFLAG_LOCKED, &mt_microstackshot_ctr, 0,
2535 "PMI counter");
2536
2537 #endif /* defined(MT_CORE_INSTRS) && defined(MT_CORE_CYCLES) */
2538
2539 /*Remote Time api*/
2540 SYSCTL_NODE(_machdep, OID_AUTO, remotetime, CTLFLAG_RD | CTLFLAG_LOCKED, 0, "Remote time api");
2541
2542 #if DEVELOPMENT || DEBUG
2543 #if CONFIG_MACH_BRIDGE_SEND_TIME
2544 extern _Atomic uint32_t bt_init_flag;
2545 extern uint32_t mach_bridge_timer_enable(uint32_t, int);
2546
2547 SYSCTL_INT(_machdep_remotetime, OID_AUTO, bridge_timer_init_flag,
2548 CTLFLAG_RD | CTLFLAG_LOCKED, &bt_init_flag, 0, "");
2549
2550 static int sysctl_mach_bridge_timer_enable SYSCTL_HANDLER_ARGS
2551 {
2552 #pragma unused(oidp, arg1, arg2)
2553 uint32_t value = 0;
2554 int error = 0;
2555 /* User is querying buffer size */
2556 if (req->oldptr == USER_ADDR_NULL && req->newptr == USER_ADDR_NULL) {
2557 req->oldidx = sizeof(value);
2558 return 0;
2559 }
2560 if (os_atomic_load(&bt_init_flag, acquire)) {
2561 if (req->newptr) {
2562 int new_value = 0;
2563 error = SYSCTL_IN(req, &new_value, sizeof(new_value));
2564 if (error) {
2565 return error;
2566 }
2567 if (new_value == 0 || new_value == 1) {
2568 value = mach_bridge_timer_enable(new_value, 1);
2569 } else {
2570 return EPERM;
2571 }
2572 } else {
2573 value = mach_bridge_timer_enable(0, 0);
2574 }
2575 }
2576 error = SYSCTL_OUT(req, &value, sizeof(value));
2577 return error;
2578 }
2579
2580 SYSCTL_PROC(_machdep_remotetime, OID_AUTO, bridge_timer_enable,
2581 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
2582 0, 0, sysctl_mach_bridge_timer_enable, "I", "");
2583
2584 #endif /* CONFIG_MACH_BRIDGE_SEND_TIME */
2585
2586 static int sysctl_mach_bridge_remote_time SYSCTL_HANDLER_ARGS
2587 {
2588 #pragma unused(oidp, arg1, arg2)
2589 uint64_t ltime = 0, rtime = 0;
2590 if (req->oldptr == USER_ADDR_NULL) {
2591 req->oldidx = sizeof(rtime);
2592 return 0;
2593 }
2594 if (req->newptr) {
2595 int error = SYSCTL_IN(req, <ime, sizeof(ltime));
2596 if (error) {
2597 return error;
2598 }
2599 }
2600 rtime = mach_bridge_remote_time(ltime);
2601 return SYSCTL_OUT(req, &rtime, sizeof(rtime));
2602 }
2603 SYSCTL_PROC(_machdep_remotetime, OID_AUTO, mach_bridge_remote_time,
2604 CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED,
2605 0, 0, sysctl_mach_bridge_remote_time, "Q", "");
2606
2607 #endif /* DEVELOPMENT || DEBUG */
2608
2609 #if CONFIG_MACH_BRIDGE_RECV_TIME
2610 extern struct bt_params bt_params_get_latest(void);
2611
2612 static int sysctl_mach_bridge_conversion_params SYSCTL_HANDLER_ARGS
2613 {
2614 #pragma unused(oidp, arg1, arg2)
2615 struct bt_params params = {};
2616 if (req->oldptr == USER_ADDR_NULL) {
2617 req->oldidx = sizeof(struct bt_params);
2618 return 0;
2619 }
2620 if (req->newptr) {
2621 return EPERM;
2622 }
2623 params = bt_params_get_latest();
2624 return SYSCTL_OUT(req, ¶ms, MIN(sizeof(params), req->oldlen));
2625 }
2626
2627 SYSCTL_PROC(_machdep_remotetime, OID_AUTO, conversion_params,
2628 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0,
2629 0, sysctl_mach_bridge_conversion_params, "S,bt_params", "");
2630
2631 #endif /* CONFIG_MACH_BRIDGE_RECV_TIME */
2632
2633 #if DEVELOPMENT || DEBUG
2634
2635 #include <pexpert/pexpert.h>
2636 extern int32_t sysctl_get_bound_cpuid(void);
2637 extern kern_return_t sysctl_thread_bind_cpuid(int32_t cpuid);
2638 static int
2639 sysctl_kern_sched_thread_bind_cpu SYSCTL_HANDLER_ARGS
2640 {
2641 #pragma unused(oidp, arg1, arg2)
2642
2643 /*
2644 * DO NOT remove this bootarg guard or make this non-development.
2645 * This kind of binding should only be used for tests and
2646 * experiments in a custom configuration, never shipping code.
2647 */
2648
2649 if (!PE_parse_boot_argn("enable_skstb", NULL, 0)) {
2650 return ENOENT;
2651 }
2652
2653 int32_t cpuid = sysctl_get_bound_cpuid();
2654
2655 int32_t new_value;
2656 int changed;
2657 int error = sysctl_io_number(req, cpuid, sizeof cpuid, &new_value, &changed);
2658 if (error) {
2659 return error;
2660 }
2661
2662 if (changed) {
2663 kern_return_t kr = sysctl_thread_bind_cpuid(new_value);
2664
2665 if (kr == KERN_NOT_SUPPORTED) {
2666 return ENOTSUP;
2667 }
2668
2669 if (kr == KERN_INVALID_VALUE) {
2670 return ERANGE;
2671 }
2672 }
2673
2674 return error;
2675 }
2676
2677 SYSCTL_PROC(_kern, OID_AUTO, sched_thread_bind_cpu, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
2678 0, 0, sysctl_kern_sched_thread_bind_cpu, "I", "");
2679
2680 #if __AMP__
2681 extern char sysctl_get_bound_cluster_type(void);
2682 extern void sysctl_thread_bind_cluster_type(char cluster_type);
2683 static int
2684 sysctl_kern_sched_thread_bind_cluster_type SYSCTL_HANDLER_ARGS
2685 {
2686 #pragma unused(oidp, arg1, arg2)
2687 char buff[4];
2688
2689 if (!PE_parse_boot_argn("enable_skstb", NULL, 0)) {
2690 return ENOENT;
2691 }
2692
2693 int error = SYSCTL_IN(req, buff, 1);
2694 if (error) {
2695 return error;
2696 }
2697 char cluster_type = buff[0];
2698
2699 if (!req->newptr) {
2700 goto out;
2701 }
2702
2703 sysctl_thread_bind_cluster_type(cluster_type);
2704 out:
2705 cluster_type = sysctl_get_bound_cluster_type();
2706 buff[0] = cluster_type;
2707
2708 return SYSCTL_OUT(req, buff, 1);
2709 }
2710
2711 SYSCTL_PROC(_kern, OID_AUTO, sched_thread_bind_cluster_type, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_LOCKED,
2712 0, 0, sysctl_kern_sched_thread_bind_cluster_type, "A", "");
2713
2714 extern char sysctl_get_task_cluster_type(void);
2715 extern void sysctl_task_set_cluster_type(char cluster_type);
2716 static int
2717 sysctl_kern_sched_task_set_cluster_type SYSCTL_HANDLER_ARGS
2718 {
2719 #pragma unused(oidp, arg1, arg2)
2720 char buff[4];
2721
2722 if (!PE_parse_boot_argn("enable_skstsct", NULL, 0)) {
2723 return ENOENT;
2724 }
2725
2726 int error = SYSCTL_IN(req, buff, 1);
2727 if (error) {
2728 return error;
2729 }
2730 char cluster_type = buff[0];
2731
2732 if (!req->newptr) {
2733 goto out;
2734 }
2735
2736 sysctl_task_set_cluster_type(cluster_type);
2737 out:
2738 cluster_type = sysctl_get_task_cluster_type();
2739 buff[0] = cluster_type;
2740
2741 return SYSCTL_OUT(req, buff, 1);
2742 }
2743
2744 SYSCTL_PROC(_kern, OID_AUTO, sched_task_set_cluster_type, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_LOCKED,
2745 0, 0, sysctl_kern_sched_task_set_cluster_type, "A", "");
2746
2747 extern kern_return_t thread_bind_cluster_id(thread_t thread, uint32_t cluster_id, thread_bind_option_t options);
2748 extern uint32_t thread_bound_cluster_id(thread_t);
2749 static int
2750 sysctl_kern_sched_thread_bind_cluster_id SYSCTL_HANDLER_ARGS
2751 {
2752 #pragma unused(oidp, arg1, arg2)
2753 if (!PE_parse_boot_argn("enable_skstb", NULL, 0)) {
2754 return ENOENT;
2755 }
2756
2757 thread_t self = current_thread();
2758 uint32_t old_value = thread_bound_cluster_id(self);
2759 uint32_t new_value;
2760
2761 int error = SYSCTL_IN(req, &new_value, sizeof(new_value));
2762 if (error) {
2763 return error;
2764 }
2765 if (new_value != old_value) {
2766 /*
2767 * This sysctl binds the thread to the cluster without any flags,
2768 * which means it will be hard bound and not check eligibility.
2769 */
2770 thread_bind_cluster_id(self, new_value, 0);
2771 }
2772 return SYSCTL_OUT(req, &old_value, sizeof(old_value));
2773 }
2774
2775 SYSCTL_PROC(_kern, OID_AUTO, sched_thread_bind_cluster_id, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
2776 0, 0, sysctl_kern_sched_thread_bind_cluster_id, "I", "");
2777
2778 #if CONFIG_SCHED_EDGE
2779
2780 extern int sched_edge_restrict_ut;
2781 SYSCTL_INT(_kern, OID_AUTO, sched_edge_restrict_ut, CTLFLAG_RW | CTLFLAG_LOCKED, &sched_edge_restrict_ut, 0, "Edge Scheduler Restrict UT Threads");
2782 extern int sched_edge_restrict_bg;
2783 SYSCTL_INT(_kern, OID_AUTO, sched_edge_restrict_bg, CTLFLAG_RW | CTLFLAG_LOCKED, &sched_edge_restrict_ut, 0, "Edge Scheduler Restrict BG Threads");
2784 extern int sched_edge_migrate_ipi_immediate;
2785 SYSCTL_INT(_kern, OID_AUTO, sched_edge_migrate_ipi_immediate, CTLFLAG_RW | CTLFLAG_LOCKED, &sched_edge_migrate_ipi_immediate, 0, "Edge Scheduler uses immediate IPIs for migration event based on execution latency");
2786
2787 #endif /* CONFIG_SCHED_EDGE */
2788
2789 #endif /* __AMP__ */
2790
2791 #if SCHED_HYGIENE_DEBUG
2792
2793 SYSCTL_QUAD(_kern, OID_AUTO, interrupt_masked_threshold_mt, CTLFLAG_RW | CTLFLAG_LOCKED,
2794 &interrupt_masked_timeout,
2795 "Interrupt masked duration after which a tracepoint is emitted or the device panics (in mach timebase units)");
2796
2797 SYSCTL_INT(_kern, OID_AUTO, interrupt_masked_debug_mode, CTLFLAG_RW | CTLFLAG_LOCKED,
2798 &interrupt_masked_debug_mode, 0,
2799 "Enable interrupt masked tracing or panic (0: off, 1: trace, 2: panic)");
2800
2801 SYSCTL_QUAD(_kern, OID_AUTO, sched_preemption_disable_threshold_mt, CTLFLAG_RW | CTLFLAG_LOCKED,
2802 &sched_preemption_disable_threshold_mt,
2803 "Preemption disablement duration after which a tracepoint is emitted or the device panics (in mach timebase units)");
2804
2805 SYSCTL_INT(_kern, OID_AUTO, sched_preemption_disable_debug_mode, CTLFLAG_RW | CTLFLAG_LOCKED,
2806 &sched_preemption_disable_debug_mode, 0,
2807 "Enable preemption disablement tracing or panic (0: off, 1: trace, 2: panic)");
2808
2809 PERCPU_DECL(uint64_t _Atomic, preemption_disable_max_mt);
2810
2811 static int
sysctl_sched_preemption_disable_stats(__unused struct sysctl_oid * oidp,__unused void * arg1,__unused int arg2,struct sysctl_req * req)2812 sysctl_sched_preemption_disable_stats(__unused struct sysctl_oid *oidp, __unused void *arg1, __unused int arg2, struct sysctl_req *req)
2813 {
2814 uint64_t stats[MAX_CPUS]; // maximum per CPU
2815
2816 /*
2817 * No synchronization here. The individual values are pretty much
2818 * independent, and reading/writing them is atomic.
2819 */
2820
2821 int cpu = 0;
2822 percpu_foreach(max_stat, preemption_disable_max_mt) {
2823 stats[cpu++] = os_atomic_load(max_stat, relaxed);
2824 }
2825
2826 if (req->newlen > 0) {
2827 // writing just resets all stats.
2828 percpu_foreach(max_stat, preemption_disable_max_mt) {
2829 os_atomic_store(max_stat, 0, relaxed);
2830 }
2831 }
2832
2833 return sysctl_io_opaque(req, stats, cpu * sizeof(uint64_t), NULL);
2834 }
2835
2836 SYSCTL_PROC(_kern, OID_AUTO, sched_preemption_disable_stats,
2837 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_LOCKED,
2838 0, 0, sysctl_sched_preemption_disable_stats, "I", "Preemption disablement statistics");
2839
2840 #endif /* SCHED_HYGIENE_DEBUG */
2841
2842 /* used for testing by exception_tests */
2843 extern uint32_t ipc_control_port_options;
2844 SYSCTL_INT(_kern, OID_AUTO, ipc_control_port_options,
2845 CTLFLAG_RD | CTLFLAG_LOCKED, &ipc_control_port_options, 0, "");
2846
2847 #endif /* DEVELOPMENT || DEBUG */
2848
2849 extern uint32_t task_exc_guard_default;
2850
2851 SYSCTL_INT(_kern, OID_AUTO, task_exc_guard_default,
2852 CTLFLAG_RD | CTLFLAG_LOCKED, &task_exc_guard_default, 0, "");
2853
2854
2855 static int
2856 sysctl_kern_tcsm_available SYSCTL_HANDLER_ARGS
2857 {
2858 #pragma unused(oidp, arg1, arg2)
2859 uint32_t value = machine_csv(CPUVN_CI) ? 1 : 0;
2860
2861 if (req->newptr) {
2862 return EINVAL;
2863 }
2864
2865 return SYSCTL_OUT(req, &value, sizeof(value));
2866 }
2867 SYSCTL_PROC(_kern, OID_AUTO, tcsm_available,
2868 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_LOCKED | CTLFLAG_MASKED | CTLFLAG_ANYBODY,
2869 0, 0, sysctl_kern_tcsm_available, "I", "");
2870
2871
2872 static int
2873 sysctl_kern_tcsm_enable SYSCTL_HANDLER_ARGS
2874 {
2875 #pragma unused(oidp, arg1, arg2)
2876 uint32_t soflags = 0;
2877 uint32_t old_value = thread_get_no_smt() ? 1 : 0;
2878
2879 int error = SYSCTL_IN(req, &soflags, sizeof(soflags));
2880 if (error) {
2881 return error;
2882 }
2883
2884 if (soflags && machine_csv(CPUVN_CI)) {
2885 thread_set_no_smt(true);
2886 machine_tecs(current_thread());
2887 }
2888
2889 return SYSCTL_OUT(req, &old_value, sizeof(old_value));
2890 }
2891 SYSCTL_PROC(_kern, OID_AUTO, tcsm_enable,
2892 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_MASKED | CTLFLAG_ANYBODY,
2893 0, 0, sysctl_kern_tcsm_enable, "I", "");
2894
2895 static int
2896 sysctl_kern_debug_get_preoslog SYSCTL_HANDLER_ARGS
2897 {
2898 #pragma unused(oidp, arg1, arg2)
2899 static bool oneshot_executed = false;
2900 size_t preoslog_size = 0;
2901 const char *preoslog = NULL;
2902 int ret = 0;
2903
2904 // DumpPanic passes a non-zero write value when it needs oneshot behaviour
2905 if (req->newptr != USER_ADDR_NULL) {
2906 uint8_t oneshot = 0;
2907 int error = SYSCTL_IN(req, &oneshot, sizeof(oneshot));
2908 if (error) {
2909 return error;
2910 }
2911
2912 if (oneshot) {
2913 if (!os_atomic_cmpxchg(&oneshot_executed, false, true, acq_rel)) {
2914 return EPERM;
2915 }
2916 }
2917 }
2918
2919 preoslog = sysctl_debug_get_preoslog(&preoslog_size);
2920 if (preoslog != NULL && preoslog_size == 0) {
2921 sysctl_debug_free_preoslog();
2922 return 0;
2923 }
2924
2925 if (preoslog == NULL || preoslog_size == 0) {
2926 return 0;
2927 }
2928
2929 if (req->oldptr == USER_ADDR_NULL) {
2930 req->oldidx = preoslog_size;
2931 return 0;
2932 }
2933
2934 ret = SYSCTL_OUT(req, preoslog, preoslog_size);
2935 sysctl_debug_free_preoslog();
2936 return ret;
2937 }
2938
2939 SYSCTL_PROC(_kern, OID_AUTO, preoslog, CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_LOCKED,
2940 0, 0, sysctl_kern_debug_get_preoslog, "-", "");
2941
2942 #if DEVELOPMENT || DEBUG
2943 extern void sysctl_task_set_no_smt(char no_smt);
2944 extern char sysctl_task_get_no_smt(void);
2945
2946 static int
2947 sysctl_kern_sched_task_set_no_smt SYSCTL_HANDLER_ARGS
2948 {
2949 #pragma unused(oidp, arg1, arg2)
2950 char buff[4];
2951
2952 int error = SYSCTL_IN(req, buff, 1);
2953 if (error) {
2954 return error;
2955 }
2956 char no_smt = buff[0];
2957
2958 if (!req->newptr) {
2959 goto out;
2960 }
2961
2962 sysctl_task_set_no_smt(no_smt);
2963 out:
2964 no_smt = sysctl_task_get_no_smt();
2965 buff[0] = no_smt;
2966
2967 return SYSCTL_OUT(req, buff, 1);
2968 }
2969
2970 SYSCTL_PROC(_kern, OID_AUTO, sched_task_set_no_smt, CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY,
2971 0, 0, sysctl_kern_sched_task_set_no_smt, "A", "");
2972
2973 static int
sysctl_kern_sched_thread_set_no_smt(__unused struct sysctl_oid * oidp,__unused void * arg1,__unused int arg2,struct sysctl_req * req)2974 sysctl_kern_sched_thread_set_no_smt(__unused struct sysctl_oid *oidp, __unused void *arg1, __unused int arg2, struct sysctl_req *req)
2975 {
2976 int new_value, changed;
2977 int old_value = thread_get_no_smt() ? 1 : 0;
2978 int error = sysctl_io_number(req, old_value, sizeof(int), &new_value, &changed);
2979
2980 if (changed) {
2981 thread_set_no_smt(!!new_value);
2982 }
2983
2984 return error;
2985 }
2986
2987 SYSCTL_PROC(_kern, OID_AUTO, sched_thread_set_no_smt,
2988 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY,
2989 0, 0, sysctl_kern_sched_thread_set_no_smt, "I", "");
2990
2991 #if CONFIG_SCHED_RT_ALLOW
2992
2993 #if DEVELOPMENT || DEBUG
2994 #define RT_ALLOW_CTLFLAGS CTLFLAG_RW
2995 #else
2996 #define RT_ALLOW_CTLFLAGS CTLFLAG_RD
2997 #endif /* DEVELOPMENT || DEBUG */
2998
2999 static int
sysctl_kern_rt_allow_limit_percent(__unused struct sysctl_oid * oidp,__unused void * arg1,__unused int arg2,struct sysctl_req * req)3000 sysctl_kern_rt_allow_limit_percent(__unused struct sysctl_oid *oidp,
3001 __unused void *arg1, __unused int arg2, struct sysctl_req *req)
3002 {
3003 extern uint8_t rt_allow_limit_percent;
3004
3005 int new_value = 0;
3006 int old_value = rt_allow_limit_percent;
3007 int changed = 0;
3008
3009 int error = sysctl_io_number(req, old_value, sizeof(old_value),
3010 &new_value, &changed);
3011 if (error != 0) {
3012 return error;
3013 }
3014
3015 /* Only accept a percentage between 1 and 99 inclusive. */
3016 if (changed) {
3017 if (new_value >= 100 || new_value <= 0) {
3018 return EINVAL;
3019 }
3020
3021 rt_allow_limit_percent = (uint8_t)new_value;
3022 }
3023
3024 return 0;
3025 }
3026
3027 SYSCTL_PROC(_kern, OID_AUTO, rt_allow_limit_percent,
3028 RT_ALLOW_CTLFLAGS | CTLTYPE_INT | CTLFLAG_LOCKED,
3029 0, 0, sysctl_kern_rt_allow_limit_percent, "I", "");
3030
3031 static int
sysctl_kern_rt_allow_limit_interval_ms(__unused struct sysctl_oid * oidp,__unused void * arg1,__unused int arg2,struct sysctl_req * req)3032 sysctl_kern_rt_allow_limit_interval_ms(__unused struct sysctl_oid *oidp,
3033 __unused void *arg1, __unused int arg2, struct sysctl_req *req)
3034 {
3035 extern uint16_t rt_allow_limit_interval_ms;
3036
3037 uint64_t new_value = 0;
3038 uint64_t old_value = rt_allow_limit_interval_ms;
3039 int changed = 0;
3040
3041 int error = sysctl_io_number(req, old_value, sizeof(old_value),
3042 &new_value, &changed);
3043 if (error != 0) {
3044 return error;
3045 }
3046
3047 /* Value is in ns. Must be at least 1ms. */
3048 if (changed) {
3049 if (new_value < 1 || new_value > UINT16_MAX) {
3050 return EINVAL;
3051 }
3052
3053 rt_allow_limit_interval_ms = (uint16_t)new_value;
3054 }
3055
3056 return 0;
3057 }
3058
3059 SYSCTL_PROC(_kern, OID_AUTO, rt_allow_limit_interval_ms,
3060 RT_ALLOW_CTLFLAGS | CTLTYPE_QUAD | CTLFLAG_LOCKED,
3061 0, 0, sysctl_kern_rt_allow_limit_interval_ms, "Q", "");
3062
3063 #endif /* CONFIG_SCHED_RT_ALLOW */
3064
3065
3066 static int
3067 sysctl_kern_task_set_filter_msg_flag SYSCTL_HANDLER_ARGS
3068 {
3069 #pragma unused(oidp, arg1, arg2)
3070 int new_value, changed;
3071 int old_value = task_get_filter_msg_flag(current_task()) ? 1 : 0;
3072 int error = sysctl_io_number(req, old_value, sizeof(int), &new_value, &changed);
3073
3074 if (changed) {
3075 task_set_filter_msg_flag(current_task(), !!new_value);
3076 }
3077
3078 return error;
3079 }
3080
3081 SYSCTL_PROC(_kern, OID_AUTO, task_set_filter_msg_flag, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
3082 0, 0, sysctl_kern_task_set_filter_msg_flag, "I", "");
3083
3084 #if CONFIG_PROC_RESOURCE_LIMITS
3085
3086 extern mach_port_name_t current_task_get_fatal_port_name(void);
3087
3088 static int
3089 sysctl_kern_task_get_fatal_port SYSCTL_HANDLER_ARGS
3090 {
3091 #pragma unused(oidp, arg1, arg2)
3092 int port = 0;
3093 int flag = 0;
3094
3095 if (req->oldptr == USER_ADDR_NULL) {
3096 req->oldidx = sizeof(mach_port_t);
3097 return 0;
3098 }
3099
3100 int error = SYSCTL_IN(req, &flag, sizeof(flag));
3101 if (error) {
3102 return error;
3103 }
3104
3105 if (flag == 1) {
3106 port = (int)current_task_get_fatal_port_name();
3107 }
3108 return SYSCTL_OUT(req, &port, sizeof(port));
3109 }
3110
3111 SYSCTL_PROC(_machdep, OID_AUTO, task_get_fatal_port, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
3112 0, 0, sysctl_kern_task_get_fatal_port, "I", "");
3113
3114 #endif /* CONFIG_PROC_RESOURCE_LIMITS */
3115
3116 extern unsigned int ipc_entry_table_count_max(void);
3117
3118 static int
3119 sysctl_mach_max_port_table_size SYSCTL_HANDLER_ARGS
3120 {
3121 #pragma unused(oidp, arg1, arg2)
3122 int old_value = ipc_entry_table_count_max();
3123 int error = sysctl_io_number(req, old_value, sizeof(int), NULL, NULL);
3124
3125 return error;
3126 }
3127
3128 SYSCTL_PROC(_machdep, OID_AUTO, max_port_table_size, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
3129 0, 0, sysctl_mach_max_port_table_size, "I", "");
3130
3131 #endif /* DEVELOPMENT || DEBUG */
3132
3133 #if defined(CONFIG_KDP_INTERACTIVE_DEBUGGING) && defined(CONFIG_KDP_COREDUMP_ENCRYPTION)
3134
3135 #define COREDUMP_ENCRYPTION_KEY_ENTITLEMENT "com.apple.private.coredump-encryption-key"
3136
3137 static int
3138 sysctl_coredump_encryption_key_update SYSCTL_HANDLER_ARGS
3139 {
3140 kern_return_t ret = KERN_SUCCESS;
3141 int error = 0;
3142 struct kdp_core_encryption_key_descriptor key_descriptor = {
3143 .kcekd_format = MACH_CORE_FILEHEADER_V2_FLAG_NEXT_COREFILE_KEY_FORMAT_NIST_P256,
3144 };
3145
3146 /* Need to be root and have entitlement */
3147 if (!kauth_cred_issuser(kauth_cred_get()) && !IOCurrentTaskHasEntitlement(COREDUMP_ENCRYPTION_KEY_ENTITLEMENT)) {
3148 return EPERM;
3149 }
3150
3151 // Sanity-check the given key length
3152 if (req->newlen > UINT16_MAX) {
3153 return EINVAL;
3154 }
3155
3156 // It is allowed for the caller to pass in a NULL buffer.
3157 // This indicates that they want us to forget about any public key we might have.
3158 if (req->newptr) {
3159 key_descriptor.kcekd_size = (uint16_t) req->newlen;
3160 key_descriptor.kcekd_key = kalloc_data(key_descriptor.kcekd_size, Z_WAITOK);
3161
3162 if (key_descriptor.kcekd_key == NULL) {
3163 return ENOMEM;
3164 }
3165
3166 error = SYSCTL_IN(req, key_descriptor.kcekd_key, key_descriptor.kcekd_size);
3167 if (error) {
3168 goto out;
3169 }
3170 }
3171
3172 ret = IOProvideCoreFileAccess(kdp_core_handle_new_encryption_key, (void *)&key_descriptor);
3173 if (KERN_SUCCESS != ret) {
3174 printf("Failed to handle the new encryption key. Error 0x%x", ret);
3175 error = EFAULT;
3176 }
3177
3178 out:
3179 kfree_data(key_descriptor.kcekd_key, key_descriptor.kcekd_size);
3180 return 0;
3181 }
3182
3183 SYSCTL_PROC(_kern, OID_AUTO, coredump_encryption_key, CTLTYPE_OPAQUE | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
3184 0, 0, &sysctl_coredump_encryption_key_update, "-", "Set a new encryption key for coredumps");
3185
3186 #endif /* CONFIG_KDP_INTERACTIVE_DEBUGGING && CONFIG_KDP_COREDUMP_ENCRYPTION*/
3187