xref: /xnu-8796.121.2/bsd/kern/kern_mman.c (revision c54f35ca767986246321eb901baf8f5ff7923f6a)
1 /*
2  * Copyright (c) 2007-2020 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 /*
29  * Copyright (c) 1988 University of Utah.
30  * Copyright (c) 1991, 1993
31  *	The Regents of the University of California.  All rights reserved.
32  *
33  * This code is derived from software contributed to Berkeley by
34  * the Systems Programming Group of the University of Utah Computer
35  * Science Department.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. All advertising materials mentioning features or use of this software
46  *    must display the following acknowledgement:
47  *	This product includes software developed by the University of
48  *	California, Berkeley and its contributors.
49  * 4. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
66  *
67  *	@(#)vm_mmap.c	8.10 (Berkeley) 2/19/95
68  */
69 /*
70  * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
71  * support for mandatory and extensible security protections.  This notice
72  * is included in support of clause 2.2 (b) of the Apple Public License,
73  * Version 2.0.
74  */
75 
76 /*
77  * Mapped file (mmap) interface to VM
78  */
79 
80 #include <sys/param.h>
81 #include <sys/systm.h>
82 #include <sys/filedesc.h>
83 #include <sys/proc_internal.h>
84 #include <sys/kauth.h>
85 #include <sys/resourcevar.h>
86 #include <sys/vnode_internal.h>
87 #include <sys/acct.h>
88 #include <sys/wait.h>
89 #include <sys/file_internal.h>
90 #include <sys/vadvise.h>
91 #include <sys/trace.h>
92 #include <sys/mman.h>
93 #include <sys/conf.h>
94 #include <sys/stat.h>
95 #include <sys/ubc.h>
96 #include <sys/ubc_internal.h>
97 #include <sys/sysproto.h>
98 
99 #include <sys/syscall.h>
100 #include <sys/kdebug.h>
101 #include <sys/bsdtask_info.h>
102 
103 #include <security/audit/audit.h>
104 #include <bsm/audit_kevents.h>
105 
106 #include <mach/mach_types.h>
107 #include <mach/mach_traps.h>
108 #include <mach/vm_sync.h>
109 #include <mach/vm_behavior.h>
110 #include <mach/vm_inherit.h>
111 #include <mach/vm_statistics.h>
112 #include <mach/mach_vm.h>
113 #include <mach/vm_map.h>
114 #include <mach/host_priv.h>
115 #include <mach/sdt.h>
116 #include <mach-o/loader.h>
117 
118 #include <machine/machine_routines.h>
119 
120 #include <kern/cpu_number.h>
121 #include <kern/host.h>
122 #include <kern/task.h>
123 #include <kern/page_decrypt.h>
124 
125 #include <IOKit/IOReturn.h>
126 #include <IOKit/IOBSD.h>
127 
128 #include <vm/vm_map.h>
129 #include <vm/vm_kern.h>
130 #include <vm/vm_pager.h>
131 #include <vm/vm_protos.h>
132 
133 #if CONFIG_MACF
134 #include <security/mac_framework.h>
135 #endif
136 #include <os/overflow.h>
137 
138 /*
139  * this function implements the same logic as dyld's "dyld_fall_2020_os_versions"
140  * from dyld_priv.h. Basically, we attempt to draw the line of: "was this code
141  * compiled with an SDK from fall of 2020 or later?""
142  */
143 static bool
proc_2020_fall_os_sdk_or_later(void)144 proc_2020_fall_os_sdk_or_later(void)
145 {
146 	const uint32_t proc_sdk_ver = proc_sdk(current_proc());
147 
148 	switch (proc_platform(current_proc())) {
149 	case PLATFORM_MACOS:
150 		return proc_sdk_ver >= 0x000a1000; // DYLD_MACOSX_VERSION_10_16
151 	case PLATFORM_IOS:
152 	case PLATFORM_IOSSIMULATOR:
153 	case PLATFORM_MACCATALYST:
154 		return proc_sdk_ver >= 0x000e0000; // DYLD_IOS_VERSION_14_0
155 	case PLATFORM_BRIDGEOS:
156 		return proc_sdk_ver >= 0x00050000; // DYLD_BRIDGEOS_VERSION_5_0
157 	case PLATFORM_TVOS:
158 	case PLATFORM_TVOSSIMULATOR:
159 		return proc_sdk_ver >= 0x000e0000; // DYLD_TVOS_VERSION_14_0
160 	case PLATFORM_WATCHOS:
161 	case PLATFORM_WATCHOSSIMULATOR:
162 		return proc_sdk_ver >= 0x00070000; // DYLD_WATCHOS_VERSION_7_0
163 	default:
164 		/*
165 		 * tough call, but let's give new platforms the benefit of the doubt
166 		 * to avoid a re-occurence of rdar://89843927
167 		 */
168 		return true;
169 	}
170 }
171 
172 #if MACH_ASSERT
173 vnode_t fbdp_vp = NULL;
174 #endif /* MACH_ASSERT */
175 
176 /*
177  * XXX Internally, we use VM_PROT_* somewhat interchangeably, but the correct
178  * XXX usage is PROT_* from an interface perspective.  Thus the values of
179  * XXX VM_PROT_* and PROT_* need to correspond.
180  */
181 int
mmap(proc_t p,struct mmap_args * uap,user_addr_t * retval)182 mmap(proc_t p, struct mmap_args *uap, user_addr_t *retval)
183 {
184 	/*
185 	 *	Map in special device (must be SHARED) or file
186 	 */
187 	struct fileproc *fp;
188 	struct                  vnode *vp;
189 	int                     flags;
190 	int                     prot;
191 	int                     err = 0;
192 	vm_map_t                user_map;
193 	kern_return_t           result;
194 	vm_map_offset_t         user_addr;
195 	vm_map_offset_t         sum;
196 	vm_map_size_t           user_size;
197 	vm_object_offset_t      pageoff;
198 	vm_object_offset_t      file_pos;
199 	vm_map_kernel_flags_t   vmk_flags = VM_MAP_KERNEL_FLAGS_NONE;
200 	boolean_t               docow;
201 	vm_prot_t               maxprot;
202 	void                    *handle;
203 	memory_object_t         pager = MEMORY_OBJECT_NULL;
204 	memory_object_control_t  control;
205 	int                     mapanon = 0;
206 	int                     fpref = 0;
207 	int error = 0;
208 	int fd = uap->fd;
209 	int num_retries = 0;
210 
211 	/*
212 	 * Note that for UNIX03 conformance, there is additional parameter checking for
213 	 * mmap() system call in libsyscall prior to entering the kernel.  The sanity
214 	 * checks and argument validation done in this function are not the only places
215 	 * one can get returned errnos.
216 	 */
217 
218 	user_map = current_map();
219 	user_addr = (vm_map_offset_t)uap->addr;
220 	user_size = (vm_map_size_t) uap->len;
221 
222 	AUDIT_ARG(addr, user_addr);
223 	AUDIT_ARG(len, user_size);
224 	AUDIT_ARG(fd, uap->fd);
225 
226 	if (vm_map_range_overflows(user_addr, user_size)) {
227 		return EINVAL;
228 	}
229 	prot = (uap->prot & VM_PROT_ALL);
230 #if 3777787
231 	/*
232 	 * Since the hardware currently does not support writing without
233 	 * read-before-write, or execution-without-read, if the request is
234 	 * for write or execute access, we must imply read access as well;
235 	 * otherwise programs expecting this to work will fail to operate.
236 	 */
237 	if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE)) {
238 		prot |= VM_PROT_READ;
239 	}
240 #endif  /* radar 3777787 */
241 
242 	flags = uap->flags;
243 	vp = NULLVP;
244 
245 	/*
246 	 * verify no unknown flags are passed in, and if any are,
247 	 * fail out early to make sure the logic below never has to deal
248 	 * with invalid flag values. only do so for processes compiled
249 	 * with Fall 2020 or later SDK, which is where we drew this
250 	 * line and documented it as such.
251 	 */
252 	if (flags & ~(MAP_SHARED |
253 	    MAP_PRIVATE |
254 	    MAP_COPY |
255 	    MAP_FIXED |
256 	    MAP_RENAME |
257 	    MAP_NORESERVE |
258 	    MAP_RESERVED0080 |                                  //grandfathered in as accepted and ignored
259 	    MAP_NOEXTEND |
260 	    MAP_HASSEMAPHORE |
261 	    MAP_NOCACHE |
262 	    MAP_JIT |
263 	    MAP_TPRO |
264 	    MAP_FILE |
265 	    MAP_ANON |
266 	    MAP_RESILIENT_CODESIGN |
267 	    MAP_RESILIENT_MEDIA |
268 #if XNU_TARGET_OS_OSX
269 	    MAP_32BIT |
270 #endif
271 	    MAP_TRANSLATED_ALLOW_EXECUTE |
272 	    MAP_UNIX03)) {
273 		if (proc_2020_fall_os_sdk_or_later()) {
274 			return EINVAL;
275 		}
276 	}
277 
278 
279 	/*
280 	 * The vm code does not have prototypes & compiler doesn't do
281 	 * the right thing when you cast 64bit value and pass it in function
282 	 * call. So here it is.
283 	 */
284 	file_pos = (vm_object_offset_t)uap->pos;
285 
286 
287 	/* make sure mapping fits into numeric range etc */
288 	if (os_add3_overflow(file_pos, user_size, vm_map_page_size(user_map) - 1, &sum)) {
289 		return EINVAL;
290 	}
291 
292 	if (flags & MAP_UNIX03) {
293 		vm_map_offset_t offset_alignment_mask;
294 
295 		/*
296 		 * Enforce UNIX03 compliance.
297 		 */
298 
299 		if (vm_map_is_exotic(current_map())) {
300 			offset_alignment_mask = 0xFFF;
301 		} else {
302 			offset_alignment_mask = vm_map_page_mask(current_map());
303 		}
304 		if (file_pos & offset_alignment_mask) {
305 			/* file offset should be page-aligned */
306 			return EINVAL;
307 		}
308 		if (!(flags & (MAP_PRIVATE | MAP_SHARED))) {
309 			/* need either MAP_PRIVATE or MAP_SHARED */
310 			return EINVAL;
311 		}
312 		if (user_size == 0) {
313 			/* mapping length should not be 0 */
314 			return EINVAL;
315 		}
316 	}
317 
318 	/*
319 	 * Align the file position to a page boundary,
320 	 * and save its page offset component.
321 	 */
322 	pageoff = (file_pos & vm_map_page_mask(user_map));
323 	file_pos -= (vm_object_offset_t)pageoff;
324 
325 
326 	/* Adjust size for rounding (on both ends). */
327 	user_size += pageoff;   /* low end... */
328 	user_size = vm_map_round_page(user_size,
329 	    vm_map_page_mask(user_map));                           /* hi end */
330 
331 
332 	if (flags & MAP_JIT) {
333 		if ((flags & MAP_FIXED) ||
334 		    (flags & MAP_SHARED) ||
335 		    !(flags & MAP_ANON) ||
336 		    (flags & MAP_RESILIENT_CODESIGN) ||
337 		    (flags & MAP_RESILIENT_MEDIA) ||
338 		    (flags & MAP_TPRO)) {
339 			return EINVAL;
340 		}
341 	}
342 
343 	if ((flags & MAP_RESILIENT_CODESIGN) ||
344 	    (flags & MAP_RESILIENT_MEDIA)) {
345 		if ((flags & MAP_ANON) ||
346 		    (flags & MAP_JIT) ||
347 		    (flags & MAP_TPRO)) {
348 			return EINVAL;
349 		}
350 	}
351 	if (flags & MAP_RESILIENT_CODESIGN) {
352 		int reject_prot = ((flags & MAP_PRIVATE) ? VM_PROT_EXECUTE : (VM_PROT_WRITE | VM_PROT_EXECUTE));
353 		if (prot & reject_prot) {
354 			/*
355 			 * Quick sanity check. maxprot is calculated below and
356 			 * we will test it again.
357 			 */
358 			return EPERM;
359 		}
360 	}
361 	if (flags & MAP_SHARED) {
362 		/*
363 		 * MAP_RESILIENT_MEDIA is not valid with MAP_SHARED because
364 		 * there is no place to inject zero-filled pages without
365 		 * actually adding them to the file.
366 		 * Since we didn't reject that combination before, there might
367 		 * already be callers using it and getting a valid MAP_SHARED
368 		 * mapping but without the resilience.
369 		 * For backwards compatibility's sake, let's keep ignoring
370 		 * MAP_RESILIENT_MEDIA in that case.
371 		 */
372 		flags &= ~MAP_RESILIENT_MEDIA;
373 	}
374 	if (flags & MAP_RESILIENT_MEDIA) {
375 		if ((flags & MAP_ANON) ||
376 		    (flags & MAP_SHARED)) {
377 			return EINVAL;
378 		}
379 	}
380 	if (flags & MAP_TPRO) {
381 		if ((prot & VM_PROT_EXECUTE) ||
382 		    !(prot & VM_PROT_WRITE) ||
383 		    (flags & MAP_SHARED) ||
384 		    !(flags & MAP_ANON)) {
385 			return EPERM;
386 		}
387 	}
388 
389 	/*
390 	 * Check for illegal addresses.  Watch out for address wrap... Note
391 	 * that VM_*_ADDRESS are not constants due to casts (argh).
392 	 */
393 	if (flags & MAP_FIXED) {
394 		/*
395 		 * The specified address must have the same remainder
396 		 * as the file offset taken modulo PAGE_SIZE, so it
397 		 * should be aligned after adjustment by pageoff.
398 		 */
399 		user_addr -= pageoff;
400 		if (user_addr & vm_map_page_mask(user_map)) {
401 			return EINVAL;
402 		}
403 	}
404 #ifdef notyet
405 	/* DO not have apis to get this info, need to wait till then*/
406 	/*
407 	 * XXX for non-fixed mappings where no hint is provided or
408 	 * the hint would fall in the potential heap space,
409 	 * place it after the end of the largest possible heap.
410 	 *
411 	 * There should really be a pmap call to determine a reasonable
412 	 * location.
413 	 */
414 	else if (addr < vm_map_round_page(p->p_vmspace->vm_daddr + MAXDSIZ,
415 	    vm_map_page_mask(user_map))) {
416 		addr = vm_map_round_page(p->p_vmspace->vm_daddr + MAXDSIZ,
417 		    vm_map_page_mask(user_map));
418 	}
419 
420 #endif
421 
422 #if CONFIG_MAP_RANGES
423 	/* default to placing mappings in the heap range. */
424 	vmk_flags.vmkf_range_id = UMEM_RANGE_ID_HEAP;
425 #endif /* CONFIG_MAP_RANGES */
426 
427 	if (flags & MAP_ANON) {
428 		maxprot = VM_PROT_ALL;
429 #if CONFIG_MACF
430 		/*
431 		 * Entitlement check.
432 		 */
433 		error = mac_proc_check_map_anon(p, user_addr, user_size, prot, flags, &maxprot);
434 		if (error) {
435 			return EINVAL;
436 		}
437 #endif /* MAC */
438 
439 		/*
440 		 * Mapping blank space is trivial.  Use positive fds as the alias
441 		 * value for memory tracking.
442 		 */
443 		if (fd != -1) {
444 			/*
445 			 * Use "fd" to pass (some) Mach VM allocation flags,
446 			 * (see the VM_FLAGS_* definitions).
447 			 */
448 			int vm_flags = fd & (VM_FLAGS_ALIAS_MASK |
449 			    VM_FLAGS_SUPERPAGE_MASK |
450 			    VM_FLAGS_PURGABLE |
451 			    VM_FLAGS_4GB_CHUNK);
452 
453 			if (vm_flags != fd) {
454 				/* reject if there are any extra flags */
455 				return EINVAL;
456 			}
457 
458 			/*
459 			 * vm_map_kernel_flags_set_vmflags() will assume that
460 			 * the full set of VM flags are passed, which is
461 			 * problematic for FIXED/ANYWHERE.
462 			 *
463 			 * The block handling MAP_FIXED below will do the same
464 			 * thing again which is fine because it's idempotent.
465 			 */
466 			if (flags & MAP_FIXED) {
467 				vm_flags |= VM_FLAGS_FIXED | VM_FLAGS_OVERWRITE;
468 			} else {
469 				vm_flags |= VM_FLAGS_ANYWHERE;
470 			}
471 			vm_map_kernel_flags_set_vmflags(&vmk_flags, vm_flags);
472 			if (vm_flags & VM_FLAGS_ALIAS_MASK) {
473 				/*
474 				 * if the client specified a tag,
475 				 * let the system policy apply.
476 				 */
477 				vmk_flags.vmkf_range_id = UMEM_RANGE_ID_DEFAULT;
478 				vm_map_kernel_flags_update_range_id(&vmk_flags, user_map);
479 			}
480 		}
481 
482 		handle = NULL;
483 		file_pos = 0;
484 		pageoff = 0;
485 		mapanon = 1;
486 	} else {
487 		struct vnode_attr va;
488 		vfs_context_t ctx = vfs_context_current();
489 
490 		if (flags & MAP_JIT) {
491 			return EINVAL;
492 		}
493 
494 		/*
495 		 * Mapping file, get fp for validation. Obtain vnode and make
496 		 * sure it is of appropriate type.
497 		 */
498 		err = fp_lookup(p, fd, &fp, 0);
499 		if (err) {
500 			return err;
501 		}
502 		fpref = 1;
503 		switch (FILEGLOB_DTYPE(fp->fp_glob)) {
504 		case DTYPE_PSXSHM:
505 			uap->addr = (user_addr_t)user_addr;
506 			uap->len = (user_size_t)user_size;
507 			uap->prot = prot;
508 			uap->flags = flags;
509 			uap->pos = file_pos;
510 			error = pshm_mmap(p, uap, retval, fp, (off_t)pageoff);
511 			goto bad;
512 		case DTYPE_VNODE:
513 			break;
514 		default:
515 			error = EINVAL;
516 			goto bad;
517 		}
518 		vp = (struct vnode *)fp_get_data(fp);
519 		error = vnode_getwithref(vp);
520 		if (error != 0) {
521 			goto bad;
522 		}
523 
524 		if (vp->v_type != VREG && vp->v_type != VCHR) {
525 			(void)vnode_put(vp);
526 			error = EINVAL;
527 			goto bad;
528 		}
529 
530 		AUDIT_ARG(vnpath, vp, ARG_VNODE1);
531 
532 		/*
533 		 * POSIX: mmap needs to update access time for mapped files
534 		 */
535 		if ((vnode_vfsvisflags(vp) & MNT_NOATIME) == 0) {
536 			VATTR_INIT(&va);
537 			nanotime(&va.va_access_time);
538 			VATTR_SET_ACTIVE(&va, va_access_time);
539 			vnode_setattr(vp, &va, ctx);
540 		}
541 
542 		/*
543 		 * XXX hack to handle use of /dev/zero to map anon memory (ala
544 		 * SunOS).
545 		 */
546 		if (vp->v_type == VCHR || vp->v_type == VSTR) {
547 			(void)vnode_put(vp);
548 			error = ENODEV;
549 			goto bad;
550 		} else {
551 			/*
552 			 * Ensure that file and memory protections are
553 			 * compatible.  Note that we only worry about
554 			 * writability if mapping is shared; in this case,
555 			 * current and max prot are dictated by the open file.
556 			 * XXX use the vnode instead?  Problem is: what
557 			 * credentials do we use for determination? What if
558 			 * proc does a setuid?
559 			 */
560 			maxprot = VM_PROT_EXECUTE;      /* TODO: Remove this and restrict maxprot? */
561 			if (fp->fp_glob->fg_flag & FREAD) {
562 				maxprot |= VM_PROT_READ;
563 			} else if (prot & PROT_READ) {
564 				(void)vnode_put(vp);
565 				error = EACCES;
566 				goto bad;
567 			}
568 			/*
569 			 * If we are sharing potential changes (either via
570 			 * MAP_SHARED or via the implicit sharing of character
571 			 * device mappings), and we are trying to get write
572 			 * permission although we opened it without asking
573 			 * for it, bail out.
574 			 */
575 
576 			if ((flags & MAP_SHARED) != 0) {
577 				if ((fp->fp_glob->fg_flag & FWRITE) != 0 &&
578 				    /*
579 				     * Do not allow writable mappings of
580 				     * swap files (see vm_swapfile_pager.c).
581 				     */
582 				    !vnode_isswap(vp)) {
583 					/*
584 					 * check for write access
585 					 *
586 					 * Note that we already made this check when granting FWRITE
587 					 * against the file, so it seems redundant here.
588 					 */
589 					error = vnode_authorize(vp, NULL, KAUTH_VNODE_CHECKIMMUTABLE, ctx);
590 
591 					/* if not granted for any reason, but we wanted it, bad */
592 					if ((prot & PROT_WRITE) && (error != 0)) {
593 						vnode_put(vp);
594 						goto bad;
595 					}
596 
597 					/* if writable, remember */
598 					if (error == 0) {
599 						maxprot |= VM_PROT_WRITE;
600 					}
601 				} else if ((prot & PROT_WRITE) != 0) {
602 					(void)vnode_put(vp);
603 					error = EACCES;
604 					goto bad;
605 				}
606 			} else {
607 				maxprot |= VM_PROT_WRITE;
608 			}
609 
610 			handle = (void *)vp;
611 #if CONFIG_MACF
612 			error = mac_file_check_mmap(vfs_context_ucred(ctx),
613 			    fp->fp_glob, prot, flags, file_pos + pageoff,
614 			    &maxprot);
615 			if (error) {
616 				(void)vnode_put(vp);
617 				goto bad;
618 			}
619 #endif /* MAC */
620 			/*
621 			 * Consult the file system to determine if this
622 			 * particular file object can be mapped.
623 			 *
624 			 * N.B. If MAP_PRIVATE (i.e. CoW) has been specified,
625 			 * then we don't check for writeability on the file
626 			 * object, because it will only ever see reads.
627 			 */
628 			error = VNOP_MMAP_CHECK(vp, (flags & MAP_PRIVATE) ?
629 			    (prot & ~PROT_WRITE) : prot, ctx);
630 			if (error) {
631 				(void)vnode_put(vp);
632 				goto bad;
633 			}
634 		}
635 
636 		/*
637 		 * No copy-on-read for mmap() mappings themselves.
638 		 */
639 		vmk_flags.vmkf_no_copy_on_read = 1;
640 	}
641 
642 	if (user_size == 0) {
643 		if (!mapanon) {
644 			(void)vnode_put(vp);
645 		}
646 		error = 0;
647 		goto bad;
648 	}
649 
650 	/*
651 	 *	We bend a little - round the start and end addresses
652 	 *	to the nearest page boundary.
653 	 */
654 	user_size = vm_map_round_page(user_size,
655 	    vm_map_page_mask(user_map));
656 
657 	if (file_pos & vm_map_page_mask(user_map)) {
658 		if (!mapanon) {
659 			(void)vnode_put(vp);
660 		}
661 		error = EINVAL;
662 		goto bad;
663 	}
664 
665 	if ((flags & MAP_FIXED) == 0) {
666 		user_addr = vm_map_round_page(user_addr,
667 		    vm_map_page_mask(user_map));
668 	} else {
669 		if (user_addr != vm_map_trunc_page(user_addr,
670 		    vm_map_page_mask(user_map))) {
671 			if (!mapanon) {
672 				(void)vnode_put(vp);
673 			}
674 			error = EINVAL;
675 			goto bad;
676 		}
677 		/*
678 		 * mmap(MAP_FIXED) will replace any existing mappings in the
679 		 * specified range, if the new mapping is successful.
680 		 * If we just deallocate the specified address range here,
681 		 * another thread might jump in and allocate memory in that
682 		 * range before we get a chance to establish the new mapping,
683 		 * and we won't have a chance to restore the old mappings.
684 		 * So we use VM_FLAGS_OVERWRITE to let Mach VM know that it
685 		 * has to deallocate the existing mappings and establish the
686 		 * new ones atomically.
687 		 */
688 		vmk_flags.vmf_fixed = true;
689 		vmk_flags.vmf_overwrite = true;
690 	}
691 
692 	if (flags & MAP_NOCACHE) {
693 		vmk_flags.vmf_no_cache = true;
694 	}
695 
696 	if (flags & MAP_JIT) {
697 		vmk_flags.vmkf_map_jit = TRUE;
698 	}
699 
700 	if (flags & MAP_TPRO) {
701 		vmk_flags.vmf_tpro = true;
702 	}
703 
704 #if CONFIG_ROSETTA
705 	if (flags & MAP_TRANSLATED_ALLOW_EXECUTE) {
706 		if (!proc_is_translated(p)) {
707 			if (!mapanon) {
708 				(void)vnode_put(vp);
709 			}
710 			error = EINVAL;
711 			goto bad;
712 		}
713 		vmk_flags.vmkf_translated_allow_execute = TRUE;
714 	}
715 #endif
716 
717 	if (flags & MAP_RESILIENT_CODESIGN) {
718 		vmk_flags.vmf_resilient_codesign = true;
719 	}
720 	if (flags & MAP_RESILIENT_MEDIA) {
721 		vmk_flags.vmf_resilient_media = true;
722 	}
723 
724 #if XNU_TARGET_OS_OSX
725 	/* macOS-specific MAP_32BIT flag handling */
726 	if (flags & MAP_32BIT) {
727 		vmk_flags.vmkf_32bit_map_va = TRUE;
728 	}
729 #endif
730 
731 	/*
732 	 * Lookup/allocate object.
733 	 */
734 	if (handle == NULL) {
735 		control = NULL;
736 #ifdef notyet
737 /* Hmm .. */
738 #if defined(VM_PROT_READ_IS_EXEC)
739 		if (prot & VM_PROT_READ) {
740 			prot |= VM_PROT_EXECUTE;
741 		}
742 		if (maxprot & VM_PROT_READ) {
743 			maxprot |= VM_PROT_EXECUTE;
744 		}
745 #endif
746 #endif
747 
748 #if 3777787
749 		if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE)) {
750 			prot |= VM_PROT_READ;
751 		}
752 		if (maxprot & (VM_PROT_EXECUTE | VM_PROT_WRITE)) {
753 			maxprot |= VM_PROT_READ;
754 		}
755 #endif  /* radar 3777787 */
756 map_anon_retry:
757 
758 		result = vm_map_enter_mem_object(user_map,
759 		    &user_addr, user_size,
760 		    0, vmk_flags,
761 		    IPC_PORT_NULL, 0, FALSE,
762 		    prot, maxprot,
763 		    (flags & MAP_SHARED) ?
764 		    VM_INHERIT_SHARE :
765 		    VM_INHERIT_DEFAULT);
766 
767 		/* If a non-binding address was specified for this anonymous
768 		 * mapping, retry the mapping with a zero base
769 		 * in the event the mapping operation failed due to
770 		 * lack of space between the address and the map's maximum.
771 		 */
772 		if ((result == KERN_NO_SPACE) && ((flags & MAP_FIXED) == 0) && user_addr && (num_retries++ == 0)) {
773 			user_addr = vm_map_page_size(user_map);
774 			goto map_anon_retry;
775 		}
776 	} else {
777 		if (vnode_isswap(vp)) {
778 			/*
779 			 * Map swap files with a special pager
780 			 * that returns obfuscated contents.
781 			 */
782 			control = NULL;
783 			pager = swapfile_pager_setup(vp);
784 			if (pager != MEMORY_OBJECT_NULL) {
785 				control = swapfile_pager_control(pager);
786 			}
787 		} else {
788 			control = ubc_getobject(vp, UBC_FLAGS_NONE);
789 		}
790 
791 		if (control == NULL) {
792 			(void)vnode_put(vp);
793 			error = ENOMEM;
794 			goto bad;
795 		}
796 
797 #if MACH_ASSERT
798 #define FBDP_PATH_NAME "/private/var/db/timezone/tz/2022a.1.1/icutz/"
799 #define FBDP_FILE_NAME "icutz44l.dat"
800 		if (fbdp_vp == NULL &&
801 		    !strncmp(vp->v_name, FBDP_FILE_NAME, strlen(FBDP_FILE_NAME))) {
802 			char *path;
803 			int len;
804 			len = MAXPATHLEN;
805 			path = zalloc_flags(ZV_NAMEI, Z_WAITOK | Z_NOFAIL);
806 			vn_getpath(vp, path, &len);
807 			if (!strncmp(path, FBDP_PATH_NAME, strlen(FBDP_PATH_NAME))) {
808 				fbdp_vp = vp;
809 				memory_object_mark_for_fbdp(control);
810 				printf("FBDP %s:%d marked vp %p \"%s\" moc %p as 'fbdp'\n", __FUNCTION__, __LINE__, vp, path, control);
811 			}
812 			zfree(ZV_NAMEI, path);
813 		}
814 #endif /* MACH_ASSERT */
815 
816 		/*
817 		 *  Set credentials:
818 		 *	FIXME: if we're writing the file we need a way to
819 		 *      ensure that someone doesn't replace our R/W creds
820 		 *      with ones that only work for read.
821 		 */
822 
823 		ubc_setthreadcred(vp, p, current_thread());
824 		docow = FALSE;
825 		if ((flags & (MAP_ANON | MAP_SHARED)) == 0) {
826 			docow = TRUE;
827 		}
828 
829 #ifdef notyet
830 /* Hmm .. */
831 #if defined(VM_PROT_READ_IS_EXEC)
832 		if (prot & VM_PROT_READ) {
833 			prot |= VM_PROT_EXECUTE;
834 		}
835 		if (maxprot & VM_PROT_READ) {
836 			maxprot |= VM_PROT_EXECUTE;
837 		}
838 #endif
839 #endif /* notyet */
840 
841 #if 3777787
842 		if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE)) {
843 			prot |= VM_PROT_READ;
844 		}
845 		if (maxprot & (VM_PROT_EXECUTE | VM_PROT_WRITE)) {
846 			maxprot |= VM_PROT_READ;
847 		}
848 #endif  /* radar 3777787 */
849 
850 map_file_retry:
851 		if (flags & MAP_RESILIENT_CODESIGN) {
852 			int reject_prot = ((flags & MAP_PRIVATE) ? VM_PROT_EXECUTE : (VM_PROT_WRITE | VM_PROT_EXECUTE));
853 			if (prot & reject_prot) {
854 				/*
855 				 * Would like to use (prot | maxprot) here
856 				 * but the assignment of VM_PROT_EXECUTE
857 				 * to maxprot above would always fail the test.
858 				 *
859 				 * Skipping the check is ok, however, because we
860 				 * restrict maxprot to prot just below in this
861 				 * block.
862 				 */
863 				assert(!mapanon);
864 				vnode_put(vp);
865 				error = EPERM;
866 				goto bad;
867 			}
868 			/* strictly limit access to "prot" */
869 			maxprot &= prot;
870 		}
871 
872 		vm_object_offset_t end_pos = 0;
873 		if (os_add_overflow(user_size, file_pos, &end_pos)) {
874 			vnode_put(vp);
875 			error = EINVAL;
876 			goto bad;
877 		}
878 
879 		result = vm_map_enter_mem_object_control(user_map,
880 		    &user_addr, user_size,
881 		    0, vmk_flags,
882 		    control, file_pos,
883 		    docow, prot, maxprot,
884 		    (flags & MAP_SHARED) ?
885 		    VM_INHERIT_SHARE :
886 		    VM_INHERIT_DEFAULT);
887 
888 		/* If a non-binding address was specified for this file backed
889 		 * mapping, retry the mapping with a zero base
890 		 * in the event the mapping operation failed due to
891 		 * lack of space between the address and the map's maximum.
892 		 */
893 		if ((result == KERN_NO_SPACE) && ((flags & MAP_FIXED) == 0) && user_addr && (num_retries++ == 0)) {
894 			user_addr = vm_map_page_size(user_map);
895 			goto map_file_retry;
896 		}
897 	}
898 
899 	if (!mapanon) {
900 		(void)vnode_put(vp);
901 	}
902 
903 	switch (result) {
904 	case KERN_SUCCESS:
905 		*retval = user_addr + pageoff;
906 		error = 0;
907 		break;
908 	case KERN_INVALID_ADDRESS:
909 	case KERN_NO_SPACE:
910 		error =  ENOMEM;
911 		break;
912 	case KERN_PROTECTION_FAILURE:
913 		error =  EACCES;
914 		break;
915 	default:
916 		error =  EINVAL;
917 		break;
918 	}
919 bad:
920 	if (pager != MEMORY_OBJECT_NULL) {
921 		/*
922 		 * Release the reference on the pager.
923 		 * If the mapping was successful, it now holds
924 		 * an extra reference.
925 		 */
926 		memory_object_deallocate(pager);
927 	}
928 	if (fpref) {
929 		fp_drop(p, fd, fp, 0);
930 	}
931 
932 	KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_SC_EXTENDED_INFO, SYS_mmap) | DBG_FUNC_NONE), fd, (uint32_t)(*retval), (uint32_t)user_size, error, 0);
933 #if XNU_TARGET_OS_OSX
934 	KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_SC_EXTENDED_INFO2, SYS_mmap) | DBG_FUNC_NONE), (uint32_t)(*retval >> 32), (uint32_t)(user_size >> 32),
935 	    (uint32_t)(file_pos >> 32), (uint32_t)file_pos, 0);
936 #endif /* XNU_TARGET_OS_OSX */
937 	return error;
938 }
939 
940 int
msync(__unused proc_t p,struct msync_args * uap,int32_t * retval)941 msync(__unused proc_t p, struct msync_args *uap, int32_t *retval)
942 {
943 	__pthread_testcancel(1);
944 	return msync_nocancel(p, (struct msync_nocancel_args *)uap, retval);
945 }
946 
947 int
msync_nocancel(__unused proc_t p,struct msync_nocancel_args * uap,__unused int32_t * retval)948 msync_nocancel(__unused proc_t p, struct msync_nocancel_args *uap, __unused int32_t *retval)
949 {
950 	mach_vm_offset_t addr;
951 	mach_vm_size_t size;
952 	int flags;
953 	vm_map_t user_map;
954 	int rv;
955 	vm_sync_t sync_flags = 0;
956 
957 	user_map = current_map();
958 	addr = (mach_vm_offset_t) uap->addr;
959 	size = (mach_vm_size_t) uap->len;
960 #if XNU_TARGET_OS_OSX
961 	KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_SC_EXTENDED_INFO, SYS_msync) | DBG_FUNC_NONE), (uint32_t)(addr >> 32), (uint32_t)(size >> 32), 0, 0, 0);
962 #endif /* XNU_TARGET_OS_OSX */
963 	if (mach_vm_range_overflows(addr, size)) {
964 		return EINVAL;
965 	}
966 	if (addr & vm_map_page_mask(user_map)) {
967 		/* UNIX SPEC: user address is not page-aligned, return EINVAL */
968 		return EINVAL;
969 	}
970 	if (size == 0) {
971 		/*
972 		 * We cannot support this properly without maintaining
973 		 * list all mmaps done. Cannot use vm_map_entry as they could be
974 		 * split or coalesced by indepenedant actions. So instead of
975 		 * inaccurate results, lets just return error as invalid size
976 		 * specified
977 		 */
978 		return EINVAL; /* XXX breaks posix apps */
979 	}
980 
981 	flags = uap->flags;
982 	/* disallow contradictory flags */
983 	if ((flags & (MS_SYNC | MS_ASYNC)) == (MS_SYNC | MS_ASYNC)) {
984 		return EINVAL;
985 	}
986 
987 	if (flags & MS_KILLPAGES) {
988 		sync_flags |= VM_SYNC_KILLPAGES;
989 	}
990 	if (flags & MS_DEACTIVATE) {
991 		sync_flags |= VM_SYNC_DEACTIVATE;
992 	}
993 	if (flags & MS_INVALIDATE) {
994 		sync_flags |= VM_SYNC_INVALIDATE;
995 	}
996 
997 	if (!(flags & (MS_KILLPAGES | MS_DEACTIVATE))) {
998 		if (flags & MS_ASYNC) {
999 			sync_flags |= VM_SYNC_ASYNCHRONOUS;
1000 		} else {
1001 			sync_flags |= VM_SYNC_SYNCHRONOUS;
1002 		}
1003 	}
1004 
1005 	sync_flags |= VM_SYNC_CONTIGUOUS;       /* complain if holes */
1006 
1007 	rv = mach_vm_msync(user_map, addr, size, sync_flags);
1008 
1009 	switch (rv) {
1010 	case KERN_SUCCESS:
1011 		break;
1012 	case KERN_INVALID_ADDRESS:      /* hole in region being sync'ed */
1013 		return ENOMEM;
1014 	case KERN_FAILURE:
1015 		return EIO;
1016 	default:
1017 		return EINVAL;
1018 	}
1019 	return 0;
1020 }
1021 
1022 
1023 int
munmap(__unused proc_t p,struct munmap_args * uap,__unused int32_t * retval)1024 munmap(__unused proc_t p, struct munmap_args *uap, __unused int32_t *retval)
1025 {
1026 	mach_vm_offset_t        user_addr;
1027 	mach_vm_size_t          user_size;
1028 	kern_return_t           result;
1029 	vm_map_t                user_map;
1030 
1031 	user_map = current_map();
1032 	user_addr = (mach_vm_offset_t) uap->addr;
1033 	user_size = (mach_vm_size_t) uap->len;
1034 
1035 	AUDIT_ARG(addr, user_addr);
1036 	AUDIT_ARG(len, user_size);
1037 
1038 	if (user_addr & vm_map_page_mask(user_map)) {
1039 		/* UNIX SPEC: user address is not page-aligned, return EINVAL */
1040 		return EINVAL;
1041 	}
1042 
1043 	if (mach_vm_range_overflows(user_addr, user_size)) {
1044 		return EINVAL;
1045 	}
1046 
1047 	if (user_size == 0) {
1048 		/* UNIX SPEC: size is 0, return EINVAL */
1049 		return EINVAL;
1050 	}
1051 
1052 	result = mach_vm_deallocate(user_map, user_addr, user_size);
1053 	if (result != KERN_SUCCESS) {
1054 		return EINVAL;
1055 	}
1056 	return 0;
1057 }
1058 
1059 int
mprotect(__unused proc_t p,struct mprotect_args * uap,__unused int32_t * retval)1060 mprotect(__unused proc_t p, struct mprotect_args *uap, __unused int32_t *retval)
1061 {
1062 	vm_prot_t prot;
1063 	mach_vm_offset_t        user_addr;
1064 	mach_vm_size_t  user_size;
1065 	kern_return_t   result;
1066 	vm_map_t        user_map;
1067 #if CONFIG_MACF
1068 	int error;
1069 #endif
1070 
1071 	AUDIT_ARG(addr, uap->addr);
1072 	AUDIT_ARG(len, uap->len);
1073 	AUDIT_ARG(value32, uap->prot);
1074 
1075 	user_map = current_map();
1076 	user_addr = (mach_vm_offset_t) uap->addr;
1077 	user_size = (mach_vm_size_t) uap->len;
1078 	prot = (vm_prot_t)(uap->prot & (VM_PROT_ALL | VM_PROT_TRUSTED | VM_PROT_STRIP_READ));
1079 
1080 	if (mach_vm_range_overflows(user_addr, user_size)) {
1081 		return EINVAL;
1082 	}
1083 	if (user_addr & vm_map_page_mask(user_map)) {
1084 		/* UNIX SPEC: user address is not page-aligned, return EINVAL */
1085 		return EINVAL;
1086 	}
1087 
1088 #ifdef notyet
1089 /* Hmm .. */
1090 #if defined(VM_PROT_READ_IS_EXEC)
1091 	if (prot & VM_PROT_READ) {
1092 		prot |= VM_PROT_EXECUTE;
1093 	}
1094 #endif
1095 #endif /* notyet */
1096 
1097 #if 3936456
1098 	if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE)) {
1099 		prot |= VM_PROT_READ;
1100 	}
1101 #endif  /* 3936456 */
1102 
1103 #if CONFIG_MACF
1104 	/*
1105 	 * The MAC check for mprotect is of limited use for 2 reasons:
1106 	 * Without mmap revocation, the caller could have asked for the max
1107 	 * protections initially instead of a reduced set, so a mprotect
1108 	 * check would offer no new security.
1109 	 * It is not possible to extract the vnode from the pager object(s)
1110 	 * of the target memory range.
1111 	 * However, the MAC check may be used to prevent a process from,
1112 	 * e.g., making the stack executable.
1113 	 */
1114 	error = mac_proc_check_mprotect(p, user_addr,
1115 	    user_size, prot);
1116 	if (error) {
1117 		return error;
1118 	}
1119 #endif
1120 
1121 	if (prot & VM_PROT_TRUSTED) {
1122 #if CONFIG_DYNAMIC_CODE_SIGNING
1123 		/* CODE SIGNING ENFORCEMENT - JIT support */
1124 		/* The special protection value VM_PROT_TRUSTED requests that we treat
1125 		 * this page as if it had a valid code signature.
1126 		 * If this is enabled, there MUST be a MAC policy implementing the
1127 		 * mac_proc_check_mprotect() hook above. Otherwise, Codesigning will be
1128 		 * compromised because the check would always succeed and thusly any
1129 		 * process could sign dynamically. */
1130 		result = vm_map_sign(
1131 			user_map,
1132 			vm_map_trunc_page(user_addr,
1133 			vm_map_page_mask(user_map)),
1134 			vm_map_round_page(user_addr + user_size,
1135 			vm_map_page_mask(user_map)));
1136 		switch (result) {
1137 		case KERN_SUCCESS:
1138 			break;
1139 		case KERN_INVALID_ADDRESS:
1140 			/* UNIX SPEC: for an invalid address range, return ENOMEM */
1141 			return ENOMEM;
1142 		default:
1143 			return EINVAL;
1144 		}
1145 #else
1146 		return ENOTSUP;
1147 #endif
1148 	}
1149 	prot &= ~VM_PROT_TRUSTED;
1150 
1151 	result = mach_vm_protect(user_map, user_addr, user_size,
1152 	    FALSE, prot);
1153 	switch (result) {
1154 	case KERN_SUCCESS:
1155 		return 0;
1156 	case KERN_PROTECTION_FAILURE:
1157 		return EACCES;
1158 	case KERN_INVALID_ADDRESS:
1159 		/* UNIX SPEC: for an invalid address range, return ENOMEM */
1160 		return ENOMEM;
1161 	}
1162 	return EINVAL;
1163 }
1164 
1165 
1166 int
minherit(__unused proc_t p,struct minherit_args * uap,__unused int32_t * retval)1167 minherit(__unused proc_t p, struct minherit_args *uap, __unused int32_t *retval)
1168 {
1169 	mach_vm_offset_t addr;
1170 	mach_vm_size_t size;
1171 	vm_inherit_t inherit;
1172 	vm_map_t        user_map;
1173 	kern_return_t   result;
1174 
1175 	AUDIT_ARG(addr, uap->addr);
1176 	AUDIT_ARG(len, uap->len);
1177 	AUDIT_ARG(value32, uap->inherit);
1178 
1179 	addr = (mach_vm_offset_t)uap->addr;
1180 	size = (mach_vm_size_t)uap->len;
1181 	inherit = uap->inherit;
1182 	if (mach_vm_range_overflows(addr, size)) {
1183 		return EINVAL;
1184 	}
1185 	user_map = current_map();
1186 	result = mach_vm_inherit(user_map, addr, size,
1187 	    inherit);
1188 	switch (result) {
1189 	case KERN_SUCCESS:
1190 		return 0;
1191 	case KERN_PROTECTION_FAILURE:
1192 		return EACCES;
1193 	}
1194 	return EINVAL;
1195 }
1196 
1197 int
madvise(__unused proc_t p,struct madvise_args * uap,__unused int32_t * retval)1198 madvise(__unused proc_t p, struct madvise_args *uap, __unused int32_t *retval)
1199 {
1200 	vm_map_t user_map;
1201 	mach_vm_offset_t start;
1202 	mach_vm_size_t size;
1203 	vm_behavior_t new_behavior;
1204 	kern_return_t   result;
1205 
1206 	/*
1207 	 * Since this routine is only advisory, we default to conservative
1208 	 * behavior.
1209 	 */
1210 	switch (uap->behav) {
1211 	case MADV_RANDOM:
1212 		new_behavior = VM_BEHAVIOR_RANDOM;
1213 		break;
1214 	case MADV_SEQUENTIAL:
1215 		new_behavior = VM_BEHAVIOR_SEQUENTIAL;
1216 		break;
1217 	case MADV_NORMAL:
1218 		new_behavior = VM_BEHAVIOR_DEFAULT;
1219 		break;
1220 	case MADV_WILLNEED:
1221 		new_behavior = VM_BEHAVIOR_WILLNEED;
1222 		break;
1223 	case MADV_DONTNEED:
1224 		new_behavior = VM_BEHAVIOR_DONTNEED;
1225 		break;
1226 	case MADV_FREE:
1227 		new_behavior = VM_BEHAVIOR_FREE;
1228 		break;
1229 	case MADV_ZERO_WIRED_PAGES:
1230 		new_behavior = VM_BEHAVIOR_ZERO_WIRED_PAGES;
1231 		break;
1232 	case MADV_FREE_REUSABLE:
1233 		new_behavior = VM_BEHAVIOR_REUSABLE;
1234 		break;
1235 	case MADV_FREE_REUSE:
1236 		new_behavior = VM_BEHAVIOR_REUSE;
1237 		break;
1238 	case MADV_CAN_REUSE:
1239 		new_behavior = VM_BEHAVIOR_CAN_REUSE;
1240 		break;
1241 	case MADV_PAGEOUT:
1242 #if MACH_ASSERT
1243 		new_behavior = VM_BEHAVIOR_PAGEOUT;
1244 		break;
1245 #else /* MACH_ASSERT */
1246 		return ENOTSUP;
1247 #endif /* MACH_ASSERT */
1248 	default:
1249 		return EINVAL;
1250 	}
1251 
1252 	start = (mach_vm_offset_t) uap->addr;
1253 	size = (mach_vm_size_t) uap->len;
1254 	if (mach_vm_range_overflows(start, size)) {
1255 		return EINVAL;
1256 	}
1257 #if __arm64__
1258 	if (start == 0 &&
1259 	    size != 0 &&
1260 	    (uap->behav == MADV_FREE ||
1261 	    uap->behav == MADV_FREE_REUSABLE)) {
1262 		printf("** FOURK_COMPAT: %d[%s] "
1263 		    "failing madvise(0x%llx,0x%llx,%s)\n",
1264 		    proc_getpid(p), p->p_comm, start, size,
1265 		    ((uap->behav == MADV_FREE_REUSABLE)
1266 		    ? "MADV_FREE_REUSABLE"
1267 		    : "MADV_FREE"));
1268 		DTRACE_VM3(fourk_compat_madvise,
1269 		    uint64_t, start,
1270 		    uint64_t, size,
1271 		    int, uap->behav);
1272 		return EINVAL;
1273 	}
1274 #endif /* __arm64__ */
1275 
1276 	user_map = current_map();
1277 
1278 	result = mach_vm_behavior_set(user_map, start, size, new_behavior);
1279 	switch (result) {
1280 	case KERN_SUCCESS:
1281 		return 0;
1282 	case KERN_INVALID_ADDRESS:
1283 		return EINVAL;
1284 	case KERN_NO_SPACE:
1285 		return ENOMEM;
1286 	}
1287 
1288 	return EINVAL;
1289 }
1290 
1291 int
mincore(__unused proc_t p,struct mincore_args * uap,__unused int32_t * retval)1292 mincore(__unused proc_t p, struct mincore_args *uap, __unused int32_t *retval)
1293 {
1294 	mach_vm_offset_t addr = 0, first_addr = 0, end = 0, cur_end = 0;
1295 	vm_map_t map = VM_MAP_NULL;
1296 	user_addr_t vec = 0;
1297 	int error = 0;
1298 	int64_t lastvecindex = 0;
1299 	int mincoreinfo = 0;
1300 	int pqueryinfo = 0;
1301 	uint64_t pqueryinfo_vec_size = 0;
1302 	vm_page_info_basic_t info = NULL;
1303 	mach_msg_type_number_t count = 0;
1304 	char *kernel_vec = NULL;
1305 	uint64_t req_vec_size_pages = 0, cur_vec_size_pages = 0, vecindex = 0;
1306 	kern_return_t kr = KERN_SUCCESS;
1307 	int effective_page_shift, effective_page_size;
1308 
1309 	map = current_map();
1310 
1311 	/*
1312 	 * On systems with 4k kernel space and 16k user space, we will
1313 	 * use the kernel page size to report back the residency information.
1314 	 * This is for backwards compatibility since we already have
1315 	 * processes that depend on this behavior.
1316 	 */
1317 	if (vm_map_page_shift(map) < PAGE_SHIFT) {
1318 		effective_page_shift = vm_map_page_shift(map);
1319 		effective_page_size = vm_map_page_size(map);
1320 	} else {
1321 		effective_page_shift = PAGE_SHIFT;
1322 		effective_page_size = PAGE_SIZE;
1323 	}
1324 
1325 	/*
1326 	 * Make sure that the addresses presented are valid for user
1327 	 * mode.
1328 	 */
1329 	first_addr = addr = vm_map_trunc_page(uap->addr,
1330 	    vm_map_page_mask(map));
1331 	end = vm_map_round_page(uap->addr + uap->len,
1332 	    vm_map_page_mask(map));
1333 
1334 	if (end < addr) {
1335 		return EINVAL;
1336 	}
1337 
1338 	if (end == addr) {
1339 		return 0;
1340 	}
1341 
1342 	/*
1343 	 * We are going to loop through the whole 'req_vec_size' pages
1344 	 * range in chunks of 'cur_vec_size'.
1345 	 */
1346 
1347 	req_vec_size_pages = (end - addr) >> effective_page_shift;
1348 	cur_vec_size_pages = MIN(req_vec_size_pages, (MAX_PAGE_RANGE_QUERY >> effective_page_shift));
1349 	size_t kernel_vec_size = cur_vec_size_pages;
1350 
1351 	kernel_vec = (char *)kalloc_data(kernel_vec_size, Z_WAITOK | Z_ZERO);
1352 
1353 	if (kernel_vec == NULL) {
1354 		return ENOMEM;
1355 	}
1356 
1357 	/*
1358 	 * Address of byte vector
1359 	 */
1360 	vec = uap->vec;
1361 
1362 	pqueryinfo_vec_size = cur_vec_size_pages * sizeof(struct vm_page_info_basic);
1363 
1364 	info = (struct vm_page_info_basic *)kalloc_data(pqueryinfo_vec_size, Z_WAITOK);
1365 
1366 	if (info == NULL) {
1367 		kfree_data(kernel_vec, kernel_vec_size);
1368 		return ENOMEM;
1369 	}
1370 
1371 	while (addr < end) {
1372 		cur_end = addr + (cur_vec_size_pages * effective_page_size);
1373 
1374 		count =  VM_PAGE_INFO_BASIC_COUNT;
1375 		kr = vm_map_page_range_info_internal(map,
1376 		    addr,
1377 		    cur_end,
1378 		    effective_page_shift,
1379 		    VM_PAGE_INFO_BASIC,
1380 		    (vm_page_info_t) info,
1381 		    &count);
1382 
1383 		assert(kr == KERN_SUCCESS);
1384 
1385 		/*
1386 		 * Do this on a map entry basis so that if the pages are not
1387 		 * in the current processes address space, we can easily look
1388 		 * up the pages elsewhere.
1389 		 */
1390 		lastvecindex = -1;
1391 
1392 		for (; addr < cur_end; addr += effective_page_size) {
1393 			pqueryinfo = info[lastvecindex + 1].disposition;
1394 
1395 			mincoreinfo = 0;
1396 
1397 			if (pqueryinfo & VM_PAGE_QUERY_PAGE_PRESENT) {
1398 				mincoreinfo |= MINCORE_INCORE;
1399 			}
1400 			if (pqueryinfo & VM_PAGE_QUERY_PAGE_REF) {
1401 				mincoreinfo |= MINCORE_REFERENCED;
1402 			}
1403 			if (pqueryinfo & VM_PAGE_QUERY_PAGE_DIRTY) {
1404 				mincoreinfo |= MINCORE_MODIFIED;
1405 			}
1406 			if (pqueryinfo & VM_PAGE_QUERY_PAGE_PAGED_OUT) {
1407 				mincoreinfo |= MINCORE_PAGED_OUT;
1408 			}
1409 			if (pqueryinfo & VM_PAGE_QUERY_PAGE_COPIED) {
1410 				mincoreinfo |= MINCORE_COPIED;
1411 			}
1412 			if ((pqueryinfo & VM_PAGE_QUERY_PAGE_EXTERNAL) == 0) {
1413 				mincoreinfo |= MINCORE_ANONYMOUS;
1414 			}
1415 			/*
1416 			 * calculate index into user supplied byte vector
1417 			 */
1418 			vecindex = (addr - first_addr) >> effective_page_shift;
1419 			kernel_vec[vecindex] = (char)mincoreinfo;
1420 			lastvecindex = vecindex;
1421 		}
1422 
1423 
1424 		assert(vecindex == (cur_vec_size_pages - 1));
1425 
1426 		error = copyout(kernel_vec, vec, cur_vec_size_pages * sizeof(char) /* a char per page */);
1427 
1428 		if (error) {
1429 			break;
1430 		}
1431 
1432 		/*
1433 		 * For the next chunk, we'll need:
1434 		 * - bump the location in the user buffer for our next disposition.
1435 		 * - new length
1436 		 * - starting address
1437 		 */
1438 		vec += cur_vec_size_pages * sizeof(char);
1439 		req_vec_size_pages = (end - addr) >> effective_page_shift;
1440 		cur_vec_size_pages = MIN(req_vec_size_pages, (MAX_PAGE_RANGE_QUERY >> effective_page_shift));
1441 
1442 		first_addr = addr;
1443 	}
1444 
1445 	kfree_data(info, pqueryinfo_vec_size);
1446 	kfree_data(kernel_vec, kernel_vec_size);
1447 
1448 	if (error) {
1449 		return EFAULT;
1450 	}
1451 
1452 	return 0;
1453 }
1454 
1455 int
mlock(__unused proc_t p,struct mlock_args * uap,__unused int32_t * retvalval)1456 mlock(__unused proc_t p, struct mlock_args *uap, __unused int32_t *retvalval)
1457 {
1458 	vm_map_t user_map;
1459 	vm_map_offset_t addr;
1460 	vm_map_size_t size, pageoff;
1461 	kern_return_t   result;
1462 
1463 	AUDIT_ARG(addr, uap->addr);
1464 	AUDIT_ARG(len, uap->len);
1465 
1466 	addr = (vm_map_offset_t) uap->addr;
1467 	size = (vm_map_size_t)uap->len;
1468 
1469 	if (vm_map_range_overflows(addr, size)) {
1470 		return EINVAL;
1471 	}
1472 
1473 	if (size == 0) {
1474 		return 0;
1475 	}
1476 
1477 	user_map = current_map();
1478 	pageoff = (addr & vm_map_page_mask(user_map));
1479 	addr -= pageoff;
1480 	size = vm_map_round_page(size + pageoff, vm_map_page_mask(user_map));
1481 
1482 	/* have to call vm_map_wire directly to pass "I don't know" protections */
1483 	result = vm_map_wire_kernel(user_map, addr, addr + size, VM_PROT_NONE, VM_KERN_MEMORY_MLOCK, TRUE);
1484 
1485 	if (result == KERN_RESOURCE_SHORTAGE) {
1486 		return EAGAIN;
1487 	} else if (result == KERN_PROTECTION_FAILURE) {
1488 		return EACCES;
1489 	} else if (result != KERN_SUCCESS) {
1490 		return ENOMEM;
1491 	}
1492 
1493 	return 0;       /* KERN_SUCCESS */
1494 }
1495 
1496 int
munlock(__unused proc_t p,struct munlock_args * uap,__unused int32_t * retval)1497 munlock(__unused proc_t p, struct munlock_args *uap, __unused int32_t *retval)
1498 {
1499 	mach_vm_offset_t addr;
1500 	mach_vm_size_t size;
1501 	vm_map_t user_map;
1502 	kern_return_t   result;
1503 
1504 	AUDIT_ARG(addr, uap->addr);
1505 	AUDIT_ARG(len, uap->len);
1506 
1507 	addr = (mach_vm_offset_t) uap->addr;
1508 	size = (mach_vm_size_t)uap->len;
1509 	user_map = current_map();
1510 	if (mach_vm_range_overflows(addr, size)) {
1511 		return EINVAL;
1512 	}
1513 	/* JMM - need to remove all wirings by spec - this just removes one */
1514 	result = mach_vm_wire_kernel(user_map, addr, size, VM_PROT_NONE, VM_KERN_MEMORY_MLOCK);
1515 	return result == KERN_SUCCESS ? 0 : ENOMEM;
1516 }
1517 
1518 
1519 int
mlockall(__unused proc_t p,__unused struct mlockall_args * uap,__unused int32_t * retval)1520 mlockall(__unused proc_t p, __unused struct mlockall_args *uap, __unused int32_t *retval)
1521 {
1522 	return ENOSYS;
1523 }
1524 
1525 int
munlockall(__unused proc_t p,__unused struct munlockall_args * uap,__unused int32_t * retval)1526 munlockall(__unused proc_t p, __unused struct munlockall_args *uap, __unused int32_t *retval)
1527 {
1528 	return ENOSYS;
1529 }
1530 
1531 #if CONFIG_CODE_DECRYPTION
1532 int
mremap_encrypted(__unused struct proc * p,struct mremap_encrypted_args * uap,__unused int32_t * retval)1533 mremap_encrypted(__unused struct proc *p, struct mremap_encrypted_args *uap, __unused int32_t *retval)
1534 {
1535 	mach_vm_offset_t    user_addr;
1536 	mach_vm_size_t      user_size;
1537 	kern_return_t       result;
1538 	vm_map_t    user_map;
1539 	uint32_t    cryptid;
1540 	cpu_type_t  cputype;
1541 	cpu_subtype_t       cpusubtype;
1542 	pager_crypt_info_t  crypt_info;
1543 	const char * cryptname = 0;
1544 	char *vpath;
1545 	int len, ret;
1546 	struct proc_regioninfo_internal pinfo;
1547 	vnode_t vp;
1548 	uintptr_t vnodeaddr;
1549 	uint32_t vid;
1550 
1551 	AUDIT_ARG(addr, uap->addr);
1552 	AUDIT_ARG(len, uap->len);
1553 
1554 	user_map = current_map();
1555 	user_addr = (mach_vm_offset_t) uap->addr;
1556 	user_size = (mach_vm_size_t) uap->len;
1557 
1558 	cryptid = uap->cryptid;
1559 	cputype = uap->cputype;
1560 	cpusubtype = uap->cpusubtype;
1561 
1562 	if (mach_vm_range_overflows(user_addr, user_size)) {
1563 		return EINVAL;
1564 	}
1565 	if (user_addr & vm_map_page_mask(user_map)) {
1566 		/* UNIX SPEC: user address is not page-aligned, return EINVAL */
1567 		return EINVAL;
1568 	}
1569 
1570 	switch (cryptid) {
1571 	case CRYPTID_NO_ENCRYPTION:
1572 		/* not encrypted, just an empty load command */
1573 		return 0;
1574 	case CRYPTID_APP_ENCRYPTION:
1575 	case CRYPTID_MODEL_ENCRYPTION:
1576 		cryptname = "com.apple.unfree";
1577 		break;
1578 	case 0x10:
1579 		/* some random cryptid that you could manually put into
1580 		 * your binary if you want NULL */
1581 		cryptname = "com.apple.null";
1582 		break;
1583 	default:
1584 		return EINVAL;
1585 	}
1586 
1587 	if (NULL == text_crypter_create) {
1588 		return ENOTSUP;
1589 	}
1590 
1591 	ret = fill_procregioninfo_onlymappedvnodes( proc_task(p), user_addr, &pinfo, &vnodeaddr, &vid);
1592 	if (ret == 0 || !vnodeaddr) {
1593 		/* No really, this returns 0 if the memory address is not backed by a file */
1594 		return EINVAL;
1595 	}
1596 
1597 	vp = (vnode_t)vnodeaddr;
1598 	if ((vnode_getwithvid(vp, vid)) == 0) {
1599 		vpath = zalloc(ZV_NAMEI);
1600 
1601 		len = MAXPATHLEN;
1602 		ret = vn_getpath(vp, vpath, &len);
1603 		if (ret) {
1604 			zfree(ZV_NAMEI, vpath);
1605 			vnode_put(vp);
1606 			return ret;
1607 		}
1608 
1609 		vnode_put(vp);
1610 	} else {
1611 		return EINVAL;
1612 	}
1613 
1614 #if 0
1615 	kprintf("%s vpath %s cryptid 0x%08x cputype 0x%08x cpusubtype 0x%08x range 0x%016llx size 0x%016llx\n",
1616 	    __FUNCTION__, vpath, cryptid, cputype, cpusubtype, (uint64_t)user_addr, (uint64_t)user_size);
1617 #endif
1618 
1619 	if (user_size == 0) {
1620 		printf("%s:%d '%s': user_addr 0x%llx user_size 0x%llx cryptid 0x%x ignored\n", __FUNCTION__, __LINE__, vpath, user_addr, user_size, cryptid);
1621 		zfree(ZV_NAMEI, vpath);
1622 		return 0;
1623 	}
1624 
1625 	/* set up decrypter first */
1626 	crypt_file_data_t crypt_data = {
1627 		.filename = vpath,
1628 		.cputype = cputype,
1629 		.cpusubtype = cpusubtype,
1630 		.origin = CRYPT_ORIGIN_LIBRARY_LOAD,
1631 	};
1632 	result = text_crypter_create(&crypt_info, cryptname, (void*)&crypt_data);
1633 #if VM_MAP_DEBUG_APPLE_PROTECT
1634 	if (vm_map_debug_apple_protect) {
1635 		printf("APPLE_PROTECT: %d[%s] map %p [0x%llx:0x%llx] %s(%s) -> 0x%x\n",
1636 		    proc_getpid(p), p->p_comm,
1637 		    user_map,
1638 		    (uint64_t) user_addr,
1639 		    (uint64_t) (user_addr + user_size),
1640 		    __FUNCTION__, vpath, result);
1641 	}
1642 #endif /* VM_MAP_DEBUG_APPLE_PROTECT */
1643 	zfree(ZV_NAMEI, vpath);
1644 
1645 	if (result) {
1646 		printf("%s: unable to create decrypter %s, kr=%d\n",
1647 		    __FUNCTION__, cryptname, result);
1648 		if (result == kIOReturnNotPrivileged) {
1649 			/* text encryption returned decryption failure */
1650 			return EPERM;
1651 		} else {
1652 			return ENOMEM;
1653 		}
1654 	}
1655 
1656 	/* now remap using the decrypter */
1657 	vm_object_offset_t crypto_backing_offset;
1658 	crypto_backing_offset = -1; /* i.e. use map entry's offset */
1659 	result = vm_map_apple_protected(user_map,
1660 	    user_addr,
1661 	    user_addr + user_size,
1662 	    crypto_backing_offset,
1663 	    &crypt_info,
1664 	    cryptid);
1665 	if (result) {
1666 		printf("%s: mapping failed with %d\n", __FUNCTION__, result);
1667 	}
1668 
1669 	if (result) {
1670 		return EPERM;
1671 	}
1672 	return 0;
1673 }
1674 #endif /* CONFIG_CODE_DECRYPTION */
1675