xref: /xnu-11215/bsd/kern/mach_loader.c (revision 4f1223e8)
1 /*
2  * Copyright (c) 2000-2020 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, 1989,  NeXT, Inc.
30  *
31  *	File:	kern/mach_loader.c
32  *	Author:	Avadis Tevanian, Jr.
33  *
34  *	Mach object file loader (kernel version, for now).
35  *
36  * 21-Jul-88  Avadis Tevanian, Jr. (avie) at NeXT
37  *	Started.
38  */
39 
40 #include <sys/param.h>
41 #include <sys/vnode_internal.h>
42 #include <sys/uio.h>
43 #include <sys/namei.h>
44 #include <sys/proc_internal.h>
45 #include <sys/kauth.h>
46 #include <sys/stat.h>
47 #include <sys/malloc.h>
48 #include <sys/mount_internal.h>
49 #include <sys/fcntl.h>
50 #include <sys/file_internal.h>
51 #include <sys/ubc_internal.h>
52 #include <sys/imgact.h>
53 #include <sys/codesign.h>
54 #include <sys/proc_uuid_policy.h>
55 #include <sys/reason.h>
56 #include <sys/kdebug.h>
57 #include <sys/spawn_internal.h>
58 
59 #include <mach/mach_types.h>
60 #include <mach/vm_map.h>        /* vm_allocate() */
61 #include <mach/mach_vm.h>       /* mach_vm_allocate() */
62 #include <mach/vm_statistics.h>
63 #include <mach/task.h>
64 #include <mach/thread_act.h>
65 
66 #include <machine/vmparam.h>
67 #include <machine/exec.h>
68 #include <machine/pal_routines.h>
69 
70 #include <kern/ast.h>
71 #include <kern/kern_types.h>
72 #include <kern/cpu_number.h>
73 #include <kern/mach_loader.h>
74 #include <kern/mach_fat.h>
75 #include <kern/kalloc.h>
76 #include <kern/task.h>
77 #include <kern/thread.h>
78 #include <kern/page_decrypt.h>
79 
80 #include <mach-o/fat.h>
81 #include <mach-o/loader.h>
82 
83 #include <vm/pmap.h>
84 #include <vm/vm_map_xnu.h>
85 #include <vm/vm_kern_xnu.h>
86 #include <vm/vm_pager_xnu.h>
87 #include <vm/vnode_pager.h>
88 #include <vm/vm_protos.h>
89 #include <vm/vm_shared_region.h>
90 #include <IOKit/IOReturn.h>     /* for kIOReturnNotPrivileged */
91 #include <IOKit/IOBSD.h>        /* for IOVnodeHasEntitlement */
92 
93 #include <os/log.h>
94 #include <os/overflow.h>
95 
96 #include "kern_exec_internal.h"
97 
98 
99 
100 /* An empty load_result_t */
101 static const load_result_t load_result_null = {
102 	.mach_header = MACH_VM_MIN_ADDRESS,
103 	.entry_point = MACH_VM_MIN_ADDRESS,
104 	.user_stack = MACH_VM_MIN_ADDRESS,
105 	.user_stack_size = 0,
106 	.user_stack_alloc = MACH_VM_MIN_ADDRESS,
107 	.user_stack_alloc_size = 0,
108 	.all_image_info_addr = MACH_VM_MIN_ADDRESS,
109 	.all_image_info_size = 0,
110 	.thread_count = 0,
111 	.unixproc = 0,
112 	.dynlinker = 0,
113 	.needs_dynlinker = 0,
114 	.validentry = 0,
115 	.using_lcmain = 0,
116 	.is_64bit_addr = 0,
117 	.is_64bit_data = 0,
118 	.custom_stack = 0,
119 	.csflags = 0,
120 	.has_pagezero = 0,
121 	.uuid = { 0 },
122 	.min_vm_addr = MACH_VM_MAX_ADDRESS,
123 	.max_vm_addr = MACH_VM_MIN_ADDRESS,
124 	.ro_vm_start = MACH_VM_MIN_ADDRESS,
125 	.ro_vm_end = MACH_VM_MIN_ADDRESS,
126 	.cs_end_offset = 0,
127 	.threadstate = NULL,
128 	.threadstate_sz = 0,
129 	.is_rosetta = 0,
130 	.dynlinker_ro_vm_start = 0,
131 	.dynlinker_ro_vm_end = 0,
132 	.dynlinker_mach_header = MACH_VM_MIN_ADDRESS,
133 	.dynlinker_fd = -1,
134 };
135 
136 /*
137  * Prototypes of static functions.
138  */
139 static load_return_t
140 parse_machfile(
141 	struct vnode            *vp,
142 	vm_map_t                map,
143 	thread_t                thread,
144 	struct mach_header      *header,
145 	off_t                   file_offset,
146 	off_t                   macho_size,
147 	int                     depth,
148 	int64_t                 slide,
149 	int64_t                 dyld_slide,
150 	load_result_t           *result,
151 	load_result_t           *binresult,
152 	struct image_params     *imgp
153 	);
154 
155 static load_return_t
156 load_segment(
157 	struct load_command             *lcp,
158 	uint32_t                        filetype,
159 	void                            *control,
160 	off_t                           pager_offset,
161 	off_t                           macho_size,
162 	struct vnode                    *vp,
163 	vm_map_t                        map,
164 	int64_t                         slide,
165 	load_result_t                   *result,
166 	struct image_params             *imgp
167 	);
168 
169 static load_return_t
170 load_uuid(
171 	struct uuid_command             *uulp,
172 	char                            *command_end,
173 	load_result_t                   *result
174 	);
175 
176 static load_return_t
177 load_version(
178 	struct version_min_command     *vmc,
179 	boolean_t               *found_version_cmd,
180 	struct image_params             *imgp,
181 	load_result_t           *result
182 	);
183 
184 static load_return_t
185 load_code_signature(
186 	struct linkedit_data_command    *lcp,
187 	struct vnode                    *vp,
188 	off_t                           macho_offset,
189 	off_t                           macho_size,
190 	cpu_type_t                      cputype,
191 	cpu_subtype_t                   cpusubtype,
192 	load_result_t                   *result,
193 	struct image_params             *imgp);
194 
195 #if CONFIG_CODE_DECRYPTION
196 static load_return_t
197 set_code_unprotect(
198 	struct encryption_info_command  *lcp,
199 	caddr_t                         addr,
200 	vm_map_t                        map,
201 	int64_t                         slide,
202 	struct vnode                    *vp,
203 	off_t                           macho_offset,
204 	cpu_type_t                      cputype,
205 	cpu_subtype_t                   cpusubtype);
206 #endif
207 
208 static
209 load_return_t
210 load_main(
211 	struct entry_point_command      *epc,
212 	thread_t                thread,
213 	int64_t                         slide,
214 	load_result_t           *result
215 	);
216 
217 static
218 load_return_t
219 setup_driver_main(
220 	thread_t                thread,
221 	int64_t                         slide,
222 	load_result_t           *result
223 	);
224 
225 static load_return_t
226 load_unixthread(
227 	struct thread_command   *tcp,
228 	thread_t                        thread,
229 	int64_t                         slide,
230 	boolean_t                       is_x86_64_compat_binary,
231 	load_result_t                   *result
232 	);
233 
234 static load_return_t
235 load_threadstate(
236 	thread_t                thread,
237 	uint32_t        *ts,
238 	uint32_t        total_size,
239 	load_result_t *
240 	);
241 
242 static load_return_t
243 load_threadstack(
244 	thread_t                thread,
245 	uint32_t                *ts,
246 	uint32_t                total_size,
247 	mach_vm_offset_t        *user_stack,
248 	int                     *customstack,
249 	boolean_t               is_x86_64_compat_binary,
250 	load_result_t           *result
251 	);
252 
253 static load_return_t
254 load_threadentry(
255 	thread_t                thread,
256 	uint32_t        *ts,
257 	uint32_t        total_size,
258 	mach_vm_offset_t        *entry_point
259 	);
260 
261 static load_return_t
262 load_dylinker(
263 	struct dylinker_command *lcp,
264 	integer_t               archbits,
265 	vm_map_t                map,
266 	thread_t                thread,
267 	int                     depth,
268 	int64_t                 slide,
269 	load_result_t           *result,
270 	struct image_params     *imgp
271 	);
272 
273 
274 #if CONFIG_ROSETTA
275 static load_return_t
276 load_rosetta(
277 	vm_map_t                        map,
278 	thread_t                        thread,
279 	load_result_t           *result,
280 	struct image_params     *imgp
281 	);
282 #endif
283 
284 #if __x86_64__
285 extern int bootarg_no32exec;
286 static boolean_t
287 check_if_simulator_binary(
288 	struct image_params     *imgp,
289 	off_t                   file_offset,
290 	off_t                   macho_size);
291 #endif
292 
293 struct macho_data;
294 
295 static load_return_t
296 get_macho_vnode(
297 	const char                      *path,
298 	integer_t               archbits,
299 	struct mach_header      *mach_header,
300 	off_t                   *file_offset,
301 	off_t                   *macho_size,
302 	struct macho_data       *macho_data,
303 	struct vnode            **vpp,
304 	struct image_params     *imgp
305 	);
306 
307 static inline void
widen_segment_command(const struct segment_command * scp32,struct segment_command_64 * scp)308 widen_segment_command(const struct segment_command *scp32,
309     struct segment_command_64 *scp)
310 {
311 	scp->cmd = scp32->cmd;
312 	scp->cmdsize = scp32->cmdsize;
313 	bcopy(scp32->segname, scp->segname, sizeof(scp->segname));
314 	scp->vmaddr = scp32->vmaddr;
315 	scp->vmsize = scp32->vmsize;
316 	scp->fileoff = scp32->fileoff;
317 	scp->filesize = scp32->filesize;
318 	scp->maxprot = scp32->maxprot;
319 	scp->initprot = scp32->initprot;
320 	scp->nsects = scp32->nsects;
321 	scp->flags = scp32->flags;
322 }
323 
324 static void
note_all_image_info_section(const struct segment_command_64 * scp,boolean_t is64,size_t section_size,const void * sections,int64_t slide,load_result_t * result)325 note_all_image_info_section(const struct segment_command_64 *scp,
326     boolean_t is64, size_t section_size, const void *sections,
327     int64_t slide, load_result_t *result)
328 {
329 	const union {
330 		struct section s32;
331 		struct section_64 s64;
332 	} *sectionp;
333 	unsigned int i;
334 
335 
336 	if (strncmp(scp->segname, "__DATA_DIRTY", sizeof(scp->segname)) != 0 &&
337 	    strncmp(scp->segname, "__DATA", sizeof(scp->segname)) != 0) {
338 		return;
339 	}
340 	for (i = 0; i < scp->nsects; ++i) {
341 		sectionp = (const void *)
342 		    ((const char *)sections + section_size * i);
343 		if (0 == strncmp(sectionp->s64.sectname, "__all_image_info",
344 		    sizeof(sectionp->s64.sectname))) {
345 			result->all_image_info_addr =
346 			    is64 ? sectionp->s64.addr : sectionp->s32.addr;
347 			result->all_image_info_addr += slide;
348 			result->all_image_info_size =
349 			    is64 ? sectionp->s64.size : sectionp->s32.size;
350 			return;
351 		}
352 	}
353 }
354 
355 #if __arm64__
356 /*
357  * Allow bypassing some security rules (hard pagezero, no write+execute)
358  * in exchange for better binary compatibility for legacy apps built
359  * before 16KB-alignment was enforced.
360  */
361 const int fourk_binary_compatibility_unsafe = TRUE;
362 #endif /* __arm64__ */
363 
364 #if XNU_TARGET_OS_OSX
365 
366 /* Determines whether this process may host/run third party plugins. */
367 static inline bool
process_is_plugin_host(struct image_params * imgp,load_result_t * result)368 process_is_plugin_host(struct image_params *imgp, load_result_t *result)
369 {
370 	if (imgp->ip_flags & IMGPF_NOJOP) {
371 		return false;
372 	}
373 
374 	if (!result->platform_binary) {
375 		return false;
376 	}
377 
378 	struct cs_blob *csblob = csvnode_get_blob(imgp->ip_vp, imgp->ip_arch_offset);
379 	const char *identity = csblob_get_identity(csblob);
380 	if (!identity) {
381 		return false;
382 	}
383 
384 	/* Check if override host plugin entitlement is present and posix spawn attribute to disable A keys is passed */
385 	if (IOVnodeHasEntitlement(imgp->ip_vp, (int64_t)imgp->ip_arch_offset, OVERRIDE_PLUGIN_HOST_ENTITLEMENT)) {
386 		bool ret = imgp->ip_flags & IMGPF_PLUGIN_HOST_DISABLE_A_KEYS;
387 		if (ret) {
388 			proc_t p = vfs_context_proc(imgp->ip_vfs_context);
389 			set_proc_name(imgp, p);
390 			os_log(OS_LOG_DEFAULT, "%s: running binary \"%s\" in keys-off mode due to posix_spawnattr_disable_ptr_auth_a_keys_np", __func__, p->p_name);
391 		}
392 		return ret;
393 	}
394 
395 	/* Disabling library validation is a good signal that this process plans to host plugins */
396 	const char *const disable_lv_entitlements[] = {
397 		"com.apple.security.cs.disable-library-validation",
398 		"com.apple.private.cs.automator-plugins",
399 		CLEAR_LV_ENTITLEMENT,
400 	};
401 	for (size_t i = 0; i < ARRAY_COUNT(disable_lv_entitlements); i++) {
402 		const char *entitlement = disable_lv_entitlements[i];
403 		if (IOVnodeHasEntitlement(imgp->ip_vp, (int64_t)imgp->ip_arch_offset, entitlement)) {
404 			proc_t p = vfs_context_proc(imgp->ip_vfs_context);
405 			set_proc_name(imgp, p);
406 			os_log(OS_LOG_DEFAULT, "%s: running binary \"%s\" in keys-off mode due to entitlement: %s", __func__, p->p_name, entitlement);
407 			return true;
408 		}
409 	}
410 
411 	/* From /System/Library/Security/HardeningExceptions.plist */
412 	const char *const hardening_exceptions[] = {
413 		"com.apple.perl5", /* Scripting engines may load third party code and jit*/
414 		"com.apple.perl", /* Scripting engines may load third party code and jit*/
415 		"org.python.python", /* Scripting engines may load third party code and jit*/
416 		"com.apple.expect", /* Scripting engines may load third party code and jit*/
417 		"com.tcltk.wish", /* Scripting engines may load third party code and jit*/
418 		"com.tcltk.tclsh", /* Scripting engines may load third party code and jit*/
419 		"com.apple.ruby", /* Scripting engines may load third party code and jit*/
420 		"com.apple.bash", /* Required for the 'enable' command */
421 		"com.apple.zsh", /* Required for the 'zmodload' command */
422 		"com.apple.ksh", /* Required for 'builtin' command */
423 	};
424 	for (size_t i = 0; i < ARRAY_COUNT(hardening_exceptions); i++) {
425 		if (strncmp(hardening_exceptions[i], identity, strlen(hardening_exceptions[i])) == 0) {
426 			proc_t p = vfs_context_proc(imgp->ip_vfs_context);
427 			set_proc_name(imgp, p);
428 			os_log(OS_LOG_DEFAULT, "%s: running binary \"%s\" in keys-off mode due to identity: %s", __func__, p->p_name, identity);
429 			return true;
430 		}
431 	}
432 
433 	return false;
434 }
435 #endif /* XNU_TARGET_OS_OSX */
436 
437 load_return_t
load_machfile(struct image_params * imgp,struct mach_header * header,thread_t thread,vm_map_t * mapp,load_result_t * result)438 load_machfile(
439 	struct image_params     *imgp,
440 	struct mach_header      *header,
441 	thread_t                thread,
442 	vm_map_t                *mapp,
443 	load_result_t           *result
444 	)
445 {
446 	struct vnode            *vp = imgp->ip_vp;
447 	off_t                   file_offset = imgp->ip_arch_offset;
448 	off_t                   macho_size = imgp->ip_arch_size;
449 	off_t                   total_size = 0;
450 	off_t                   file_size = imgp->ip_vattr->va_data_size;
451 	pmap_t                  pmap = 0;       /* protected by create_map */
452 	vm_map_t                map;
453 	load_result_t           myresult;
454 	load_return_t           lret;
455 	boolean_t enforce_hard_pagezero = TRUE;
456 	int in_exec = (imgp->ip_flags & IMGPF_EXEC);
457 	task_t task = current_task();
458 	int64_t                 aslr_page_offset = 0;
459 	int64_t                 dyld_aslr_page_offset = 0;
460 	int64_t                 aslr_section_size = 0;
461 	int64_t                 aslr_section_offset = 0;
462 	kern_return_t           kret;
463 	unsigned int            pmap_flags = 0;
464 
465 	if (os_add_overflow(file_offset, macho_size, &total_size) ||
466 	    total_size > file_size) {
467 		return LOAD_BADMACHO;
468 	}
469 
470 	result->is_64bit_addr = ((imgp->ip_flags & IMGPF_IS_64BIT_ADDR) == IMGPF_IS_64BIT_ADDR);
471 	result->is_64bit_data = ((imgp->ip_flags & IMGPF_IS_64BIT_DATA) == IMGPF_IS_64BIT_DATA);
472 #if defined(HAS_APPLE_PAC)
473 	pmap_flags |= (imgp->ip_flags & IMGPF_NOJOP) ? PMAP_CREATE_DISABLE_JOP : 0;
474 #endif /* defined(HAS_APPLE_PAC) */
475 #if CONFIG_ROSETTA
476 	pmap_flags |= (imgp->ip_flags & IMGPF_ROSETTA) ? PMAP_CREATE_ROSETTA : 0;
477 #endif
478 	pmap_flags |= result->is_64bit_addr ? PMAP_CREATE_64BIT : 0;
479 
480 	task_t ledger_task;
481 	if (imgp->ip_new_thread) {
482 		ledger_task = get_threadtask(imgp->ip_new_thread);
483 	} else {
484 		ledger_task = task;
485 	}
486 
487 #if XNU_TARGET_OS_OSX && _POSIX_SPAWN_FORCE_4K_PAGES && PMAP_CREATE_FORCE_4K_PAGES
488 	if (imgp->ip_px_sa != NULL) {
489 		struct _posix_spawnattr* psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
490 		if (psa->psa_flags & _POSIX_SPAWN_FORCE_4K_PAGES) {
491 			pmap_flags |= PMAP_CREATE_FORCE_4K_PAGES;
492 		}
493 	}
494 #endif /* XNU_TARGET_OS_OSX && _POSIX_SPAWN_FORCE_4K_PAGES && PMAP_CREATE_FORCE_4K_PAGE */
495 
496 	pmap = pmap_create_options(get_task_ledger(ledger_task),
497 	    (vm_map_size_t) 0,
498 	    pmap_flags);
499 	if (pmap == NULL) {
500 		return LOAD_RESOURCE;
501 	}
502 	map = vm_map_create_options(pmap, 0,
503 	    vm_compute_max_offset(result->is_64bit_addr),
504 	    VM_MAP_CREATE_PAGEABLE);
505 
506 #if defined(__arm64__)
507 	if (result->is_64bit_addr) {
508 		/* enforce 16KB alignment of VM map entries */
509 		vm_map_set_page_shift(map, SIXTEENK_PAGE_SHIFT);
510 	} else {
511 		vm_map_set_page_shift(map, page_shift_user32);
512 	}
513 #endif /* __arm64__ */
514 
515 #if PMAP_CREATE_FORCE_4K_PAGES
516 	if (pmap_flags & PMAP_CREATE_FORCE_4K_PAGES) {
517 		DEBUG4K_LIFE("***** launching '%s' as 4k *****\n", vp->v_name);
518 		vm_map_set_page_shift(map, FOURK_PAGE_SHIFT);
519 	}
520 #endif /* PMAP_CREATE_FORCE_4K_PAGES */
521 
522 #ifndef CONFIG_ENFORCE_SIGNED_CODE
523 	/* This turns off faulting for executable pages, which allows
524 	 * to circumvent Code Signing Enforcement. The per process
525 	 * flag (CS_ENFORCEMENT) is not set yet, but we can use the
526 	 * global flag.
527 	 */
528 	if (!cs_process_global_enforcement() && (header->flags & MH_ALLOW_STACK_EXECUTION)) {
529 		vm_map_disable_NX(map);
530 		// TODO: Message Trace or log that this is happening
531 	}
532 #endif
533 
534 	/* Forcibly disallow execution from data pages on even if the arch
535 	 * normally permits it. */
536 	if ((header->flags & MH_NO_HEAP_EXECUTION) && !(imgp->ip_flags & IMGPF_ALLOW_DATA_EXEC)) {
537 		vm_map_disallow_data_exec(map);
538 	}
539 
540 	/*
541 	 * Compute a random offset for ASLR, and an independent random offset for dyld.
542 	 */
543 	if (!(imgp->ip_flags & IMGPF_DISABLE_ASLR)) {
544 		vm_map_get_max_aslr_slide_section(map, &aslr_section_offset, &aslr_section_size);
545 		aslr_section_offset = (random() % aslr_section_offset) * aslr_section_size;
546 
547 		aslr_page_offset = random();
548 		aslr_page_offset = (aslr_page_offset % (vm_map_get_max_aslr_slide_pages(map) - 1)) + 1;
549 		aslr_page_offset <<= vm_map_page_shift(map);
550 
551 		dyld_aslr_page_offset = random();
552 		dyld_aslr_page_offset = (dyld_aslr_page_offset %
553 		    (vm_map_get_max_loader_aslr_slide_pages(map) - 1)) + 1;
554 		dyld_aslr_page_offset <<= vm_map_page_shift(map);
555 
556 		aslr_page_offset += aslr_section_offset;
557 	}
558 	if (vm_map_page_shift(map) < (int)PAGE_SHIFT) {
559 		DEBUG4K_LOAD("slide=0x%llx dyld_slide=0x%llx\n", aslr_page_offset, dyld_aslr_page_offset);
560 	}
561 
562 	if (!result) {
563 		result = &myresult;
564 	}
565 
566 	*result = load_result_null;
567 
568 	/*
569 	 * re-set the bitness on the load result since we cleared the load result above.
570 	 */
571 	result->is_64bit_addr = ((imgp->ip_flags & IMGPF_IS_64BIT_ADDR) == IMGPF_IS_64BIT_ADDR);
572 	result->is_64bit_data = ((imgp->ip_flags & IMGPF_IS_64BIT_DATA) == IMGPF_IS_64BIT_DATA);
573 
574 	lret = parse_machfile(vp, map, thread, header, file_offset, macho_size,
575 	    0, aslr_page_offset, dyld_aslr_page_offset, result,
576 	    NULL, imgp);
577 
578 	if (lret != LOAD_SUCCESS) {
579 		imgp->ip_free_map = map;
580 		return lret;
581 	}
582 
583 #if __x86_64__
584 	/*
585 	 * On x86, for compatibility, don't enforce the hard page-zero restriction for 32-bit binaries.
586 	 */
587 	if (!result->is_64bit_addr) {
588 		enforce_hard_pagezero = FALSE;
589 	}
590 
591 	/*
592 	 * For processes with IMGPF_HIGH_BITS_ASLR, add a few random high bits
593 	 * to the start address for "anywhere" memory allocations.
594 	 */
595 #define VM_MAP_HIGH_START_BITS_COUNT 8
596 #define VM_MAP_HIGH_START_BITS_SHIFT 27
597 	if (result->is_64bit_addr &&
598 	    (imgp->ip_flags & IMGPF_HIGH_BITS_ASLR)) {
599 		int random_bits;
600 		vm_map_offset_t high_start;
601 
602 		random_bits = random();
603 		random_bits &= (1 << VM_MAP_HIGH_START_BITS_COUNT) - 1;
604 		high_start = (((vm_map_offset_t)random_bits)
605 		        << VM_MAP_HIGH_START_BITS_SHIFT);
606 		vm_map_set_high_start(map, high_start);
607 	}
608 #endif /* __x86_64__ */
609 
610 	/*
611 	 * Check to see if the page zero is enforced by the map->min_offset.
612 	 */
613 	if (enforce_hard_pagezero &&
614 	    (vm_map_has_hard_pagezero(map, 0x1000) == FALSE)) {
615 #if __arm64__
616 		if (
617 			!result->is_64bit_addr && /* not 64-bit address space */
618 			!(header->flags & MH_PIE) &&      /* not PIE */
619 			(vm_map_page_shift(map) != FOURK_PAGE_SHIFT ||
620 			PAGE_SHIFT != FOURK_PAGE_SHIFT) && /* page size != 4KB */
621 			result->has_pagezero && /* has a "soft" page zero */
622 			fourk_binary_compatibility_unsafe) {
623 			/*
624 			 * For backwards compatibility of "4K" apps on
625 			 * a 16K system, do not enforce a hard page zero...
626 			 */
627 		} else
628 #endif /* __arm64__ */
629 		{
630 			imgp->ip_free_map = map;
631 			return LOAD_BADMACHO;
632 		}
633 	}
634 
635 #if __arm64__
636 	if (enforce_hard_pagezero && result->is_64bit_addr && (header->cputype == CPU_TYPE_ARM64)) {
637 		/* 64 bit ARM binary must have "hard page zero" of 4GB to cover the lower 32 bit address space */
638 		if (vm_map_has_hard_pagezero(map, 0x100000000) == FALSE) {
639 			imgp->ip_free_map = map;
640 			return LOAD_BADMACHO;
641 		}
642 	}
643 #endif
644 
645 	vm_commit_pagezero_status(map);
646 
647 	/*
648 	 * If this is an exec, then we are going to destroy the old
649 	 * task, and it's correct to halt it; if it's spawn, the
650 	 * task is not yet running, and it makes no sense.
651 	 */
652 	if (in_exec) {
653 		proc_t p = current_proc();
654 		/*
655 		 * Mark the task as halting and start the other
656 		 * threads towards terminating themselves.  Then
657 		 * make sure any threads waiting for a process
658 		 * transition get informed that we are committed to
659 		 * this transition, and then finally complete the
660 		 * task halting (wait for threads and then cleanup
661 		 * task resources).
662 		 *
663 		 * NOTE: task_start_halt() makes sure that no new
664 		 * threads are created in the task during the transition.
665 		 * We need to mark the workqueue as exiting before we
666 		 * wait for threads to terminate (at the end of which
667 		 * we no longer have a prohibition on thread creation).
668 		 *
669 		 * Finally, clean up any lingering workqueue data structures
670 		 * that may have been left behind by the workqueue threads
671 		 * as they exited (and then clean up the work queue itself).
672 		 */
673 		kret = task_start_halt(task);
674 		if (kret != KERN_SUCCESS) {
675 			imgp->ip_free_map = map;
676 			return LOAD_FAILURE;
677 		}
678 		proc_transcommit(p, 0);
679 		workq_mark_exiting(p);
680 		task_complete_halt(task);
681 		workq_exit(p);
682 
683 		/*
684 		 * Roll up accounting info to new task. The roll up is done after
685 		 * task_complete_halt to make sure the thread accounting info is
686 		 * rolled up to current_task.
687 		 */
688 		task_rollup_accounting_info(get_threadtask(thread), task);
689 	}
690 	*mapp = map;
691 
692 #if XNU_TARGET_OS_OSX
693 	if (process_is_plugin_host(imgp, result)) {
694 		/*
695 		 * We need to disable security policies for processes
696 		 * that run third party plugins.
697 		 */
698 		imgp->ip_flags |= IMGPF_3P_PLUGINS;
699 	}
700 
701 #if __has_feature(ptrauth_calls)
702 	/*
703 	 * arm64e plugin hosts currently run with JOP keys disabled, since they
704 	 * may need to run arm64 plugins.
705 	 */
706 	if (imgp->ip_flags & IMGPF_3P_PLUGINS) {
707 		imgp->ip_flags |= IMGPF_NOJOP;
708 		pmap_disable_user_jop(pmap);
709 	}
710 
711 #if CONFIG_ROSETTA
712 	/* Disable JOP keys if the Rosetta runtime being used isn't arm64e */
713 	if (result->is_rosetta && (imgp->ip_flags & IMGPF_NOJOP)) {
714 		pmap_disable_user_jop(pmap);
715 	}
716 #endif /* CONFIG_ROSETTA */
717 #endif /* __has_feature(ptrauth_calls)*/
718 #endif /* XNU_TARGET_OS_OSX */
719 
720 
721 	return LOAD_SUCCESS;
722 }
723 
724 int macho_printf = 0;
725 #define MACHO_PRINTF(args)                              \
726 	do {                                            \
727 	        if (macho_printf) {                     \
728 	                printf args;                    \
729 	        }                                       \
730 	} while (0)
731 
732 
733 static boolean_t
pie_required(cpu_type_t exectype,cpu_subtype_t execsubtype)734 pie_required(
735 	cpu_type_t exectype,
736 	cpu_subtype_t execsubtype)
737 {
738 	switch (exectype) {
739 	case CPU_TYPE_X86_64:
740 		return FALSE;
741 	case CPU_TYPE_ARM64:
742 		return TRUE;
743 	case CPU_TYPE_ARM:
744 		switch (execsubtype) {
745 		case CPU_SUBTYPE_ARM_V7K:
746 			return TRUE;
747 		}
748 		break;
749 	}
750 	return FALSE;
751 }
752 
753 /*
754  * The file size of a mach-o file is limited to 32 bits; this is because
755  * this is the limit on the kalloc() of enough bytes for a mach_header and
756  * the contents of its sizeofcmds, which is currently constrained to 32
757  * bits in the file format itself.  We read into the kernel buffer the
758  * commands section, and then parse it in order to parse the mach-o file
759  * format load_command segment(s).  We are only interested in a subset of
760  * the total set of possible commands. If "map"==VM_MAP_NULL or
761  * "thread"==THREAD_NULL, do not make permament VM modifications,
762  * just preflight the parse.
763  */
764 static
765 load_return_t
parse_machfile(struct vnode * vp,vm_map_t map,thread_t thread,struct mach_header * header,off_t file_offset,off_t macho_size,int depth,int64_t aslr_offset,int64_t dyld_aslr_offset,load_result_t * result,load_result_t * binresult,struct image_params * imgp)766 parse_machfile(
767 	struct vnode            *vp,
768 	vm_map_t                map,
769 	thread_t                thread,
770 	struct mach_header      *header,
771 	off_t                   file_offset,
772 	off_t                   macho_size,
773 	int                     depth,
774 	int64_t                 aslr_offset,
775 	int64_t                 dyld_aslr_offset,
776 	load_result_t           *result,
777 	load_result_t           *binresult,
778 	struct image_params     *imgp
779 	)
780 {
781 	uint32_t                ncmds;
782 	struct load_command     *lcp;
783 	struct dylinker_command *dlp = 0;
784 	void *                  control;
785 	load_return_t           ret = LOAD_SUCCESS;
786 	void *                  addr;
787 	vm_size_t               alloc_size, cmds_size;
788 	size_t                  offset;
789 	size_t                  oldoffset;      /* for overflow check */
790 	int                     pass;
791 	proc_t                  p = vfs_context_proc(imgp->ip_vfs_context);
792 	int                     error;
793 	int                     resid = 0;
794 	int                     spawn = (imgp->ip_flags & IMGPF_SPAWN);
795 	size_t                  mach_header_sz = sizeof(struct mach_header);
796 	boolean_t               abi64;
797 	boolean_t               got_code_signatures = FALSE;
798 	boolean_t               found_header_segment = FALSE;
799 	boolean_t               found_xhdr = FALSE;
800 	boolean_t               found_version_cmd = FALSE;
801 	int64_t                 slide = 0;
802 	boolean_t               dyld_no_load_addr = FALSE;
803 	boolean_t               is_dyld = FALSE;
804 	vm_map_offset_t         effective_page_mask = PAGE_MASK;
805 #if __arm64__
806 	uint64_t                pagezero_end = 0;
807 	uint64_t                executable_end = 0;
808 	uint64_t                writable_start = 0;
809 	vm_map_size_t           effective_page_size;
810 
811 	effective_page_mask = vm_map_page_mask(map);
812 	effective_page_size = vm_map_page_size(map);
813 #endif /* __arm64__ */
814 
815 	if (header->magic == MH_MAGIC_64 ||
816 	    header->magic == MH_CIGAM_64) {
817 		mach_header_sz = sizeof(struct mach_header_64);
818 	}
819 
820 	/*
821 	 *	Break infinite recursion
822 	 */
823 	if (depth > 2) {
824 		return LOAD_FAILURE;
825 	}
826 
827 	depth++;
828 
829 	/*
830 	 * Set CS_NO_UNTRUSTED_HELPERS by default; load_dylinker and load_rosetta
831 	 * will unset it if necessary.
832 	 */
833 	if (depth == 1) {
834 		result->csflags |= CS_NO_UNTRUSTED_HELPERS;
835 	}
836 
837 	/*
838 	 *	Check to see if right machine type.
839 	 */
840 	if (((cpu_type_t)(header->cputype & ~CPU_ARCH_MASK) != (cpu_type() & ~CPU_ARCH_MASK))
841 	    ) {
842 		return LOAD_BADARCH;
843 	}
844 
845 	if (!grade_binary(header->cputype,
846 	    header->cpusubtype & ~CPU_SUBTYPE_MASK,
847 	    header->cpusubtype & CPU_SUBTYPE_MASK, TRUE)) {
848 		return LOAD_BADARCH;
849 	}
850 
851 	abi64 = ((header->cputype & CPU_ARCH_ABI64) == CPU_ARCH_ABI64);
852 
853 	switch (header->filetype) {
854 	case MH_EXECUTE:
855 		if (depth != 1 && depth != 3) {
856 			return LOAD_FAILURE;
857 		}
858 		if (header->flags & MH_DYLDLINK) {
859 			/* Check properties of dynamic executables */
860 			if (!(header->flags & MH_PIE) && pie_required(header->cputype, header->cpusubtype & ~CPU_SUBTYPE_MASK)) {
861 				return LOAD_FAILURE;
862 			}
863 			result->needs_dynlinker = TRUE;
864 		} else if (header->cputype == CPU_TYPE_X86_64) {
865 			/* x86_64 static binaries allowed */
866 #if CONFIG_ROSETTA
867 		} else if (imgp->ip_flags & IMGPF_ROSETTA) {
868 			/* Rosetta runtime allowed */
869 #endif /* CONFIG_X86_64_COMPAT */
870 		} else {
871 			/* Check properties of static executables (disallowed except for development) */
872 #if !(DEVELOPMENT || DEBUG)
873 			return LOAD_FAILURE;
874 #endif
875 		}
876 		break;
877 	case MH_DYLINKER:
878 		if (depth != 2) {
879 			return LOAD_FAILURE;
880 		}
881 		is_dyld = TRUE;
882 		break;
883 
884 	default:
885 		return LOAD_FAILURE;
886 	}
887 
888 	/*
889 	 *	For PIE and dyld, slide everything by the ASLR offset.
890 	 */
891 	if ((header->flags & MH_PIE) || is_dyld) {
892 		slide = aslr_offset;
893 	}
894 
895 	/*
896 	 *	Get the pager for the file.
897 	 */
898 	control = ubc_getobject(vp, UBC_FLAGS_NONE);
899 
900 	/* ensure header + sizeofcmds falls within the file */
901 	if (os_add_overflow(mach_header_sz, header->sizeofcmds, &cmds_size) ||
902 	    (off_t)cmds_size > macho_size ||
903 	    round_page_overflow(cmds_size, &alloc_size) ||
904 	    alloc_size > INT_MAX) {
905 		return LOAD_BADMACHO;
906 	}
907 
908 	/*
909 	 * Map the load commands into kernel memory.
910 	 */
911 	addr = kalloc_data(alloc_size, Z_WAITOK);
912 	if (addr == NULL) {
913 		return LOAD_NOSPACE;
914 	}
915 
916 	error = vn_rdwr(UIO_READ, vp, addr, (int)alloc_size, file_offset,
917 	    UIO_SYSSPACE, 0, vfs_context_ucred(imgp->ip_vfs_context), &resid, p);
918 	if (error) {
919 		kfree_data(addr, alloc_size);
920 		return LOAD_IOERROR;
921 	}
922 
923 	if (resid) {
924 		{
925 			/* We must be able to read in as much as the mach_header indicated */
926 			kfree_data(addr, alloc_size);
927 			return LOAD_BADMACHO;
928 		}
929 	}
930 
931 	/*
932 	 *  Scan through the commands, processing each one as necessary.
933 	 *  We parse in three passes through the headers:
934 	 *  0: determine if TEXT and DATA boundary can be page-aligned, load platform version
935 	 *  1: thread state, uuid, code signature
936 	 *  2: segments
937 	 *  3: dyld, encryption, check entry point
938 	 */
939 
940 	boolean_t slide_realign = FALSE;
941 #if __arm64__
942 	if (!abi64) {
943 		slide_realign = TRUE;
944 	}
945 #endif
946 
947 	for (pass = 0; pass <= 3; pass++) {
948 		if (pass == 1) {
949 #if __arm64__
950 			boolean_t       is_pie;
951 			int64_t         adjust;
952 
953 			is_pie = ((header->flags & MH_PIE) != 0);
954 			if (pagezero_end != 0 &&
955 			    pagezero_end < effective_page_size) {
956 				/* need at least 1 page for PAGEZERO */
957 				adjust = effective_page_size;
958 				MACHO_PRINTF(("pagezero boundary at "
959 				    "0x%llx; adjust slide from "
960 				    "0x%llx to 0x%llx%s\n",
961 				    (uint64_t) pagezero_end,
962 				    slide,
963 				    slide + adjust,
964 				    (is_pie
965 				    ? ""
966 				    : " BUT NO PIE ****** :-(")));
967 				if (is_pie) {
968 					slide += adjust;
969 					pagezero_end += adjust;
970 					executable_end += adjust;
971 					writable_start += adjust;
972 				}
973 			}
974 			if (pagezero_end != 0) {
975 				result->has_pagezero = TRUE;
976 			}
977 			if (executable_end == writable_start &&
978 			    (executable_end & effective_page_mask) != 0 &&
979 			    (executable_end & FOURK_PAGE_MASK) == 0) {
980 				/*
981 				 * The TEXT/DATA boundary is 4K-aligned but
982 				 * not page-aligned.  Adjust the slide to make
983 				 * it page-aligned and avoid having a page
984 				 * with both write and execute permissions.
985 				 */
986 				adjust =
987 				    (effective_page_size -
988 				    (executable_end & effective_page_mask));
989 				MACHO_PRINTF(("page-unaligned X-W boundary at "
990 				    "0x%llx; adjust slide from "
991 				    "0x%llx to 0x%llx%s\n",
992 				    (uint64_t) executable_end,
993 				    slide,
994 				    slide + adjust,
995 				    (is_pie
996 				    ? ""
997 				    : " BUT NO PIE ****** :-(")));
998 				if (is_pie) {
999 					slide += adjust;
1000 				}
1001 			}
1002 #endif /* __arm64__ */
1003 
1004 			if (dyld_no_load_addr && binresult) {
1005 				/*
1006 				 * The dyld Mach-O does not specify a load address. Try to locate
1007 				 * it right after the main binary. If binresult == NULL, load
1008 				 * directly to the given slide.
1009 				 */
1010 				mach_vm_address_t max_vm_addr = binresult->max_vm_addr;
1011 				slide = vm_map_round_page(slide + max_vm_addr, effective_page_mask);
1012 			}
1013 		}
1014 
1015 		/*
1016 		 * Check that the entry point is contained in an executable segment
1017 		 */
1018 		if ((pass == 3) && (thread != THREAD_NULL)) {
1019 			if (depth == 1 && imgp && (imgp->ip_flags & IMGPF_DRIVER)) {
1020 				/* Driver binaries must have driverkit platform */
1021 				if (result->ip_platform == PLATFORM_DRIVERKIT) {
1022 					/* Driver binaries have no entry point */
1023 					ret = setup_driver_main(thread, slide, result);
1024 				} else {
1025 					ret = LOAD_FAILURE;
1026 				}
1027 			} else if (!result->using_lcmain && result->validentry == 0) {
1028 				ret = LOAD_FAILURE;
1029 			}
1030 			if (ret != KERN_SUCCESS) {
1031 				thread_state_initialize(thread);
1032 				break;
1033 			}
1034 		}
1035 
1036 		/*
1037 		 * Check that some segment maps the start of the mach-o file, which is
1038 		 * needed by the dynamic loader to read the mach headers, etc.
1039 		 */
1040 		if ((pass == 3) && (found_header_segment == FALSE)) {
1041 			ret = LOAD_BADMACHO;
1042 			break;
1043 		}
1044 
1045 		/*
1046 		 * Loop through each of the load_commands indicated by the
1047 		 * Mach-O header; if an absurd value is provided, we just
1048 		 * run off the end of the reserved section by incrementing
1049 		 * the offset too far, so we are implicitly fail-safe.
1050 		 */
1051 		offset = mach_header_sz;
1052 		ncmds = header->ncmds;
1053 
1054 		while (ncmds--) {
1055 			/* ensure enough space for a minimal load command */
1056 			if (offset + sizeof(struct load_command) > cmds_size) {
1057 				ret = LOAD_BADMACHO;
1058 				break;
1059 			}
1060 
1061 			/*
1062 			 *	Get a pointer to the command.
1063 			 */
1064 			lcp = (struct load_command *)((uintptr_t)addr + offset);
1065 			oldoffset = offset;
1066 
1067 			/*
1068 			 * Perform prevalidation of the struct load_command
1069 			 * before we attempt to use its contents.  Invalid
1070 			 * values are ones which result in an overflow, or
1071 			 * which can not possibly be valid commands, or which
1072 			 * straddle or exist past the reserved section at the
1073 			 * start of the image.
1074 			 */
1075 			if (os_add_overflow(offset, lcp->cmdsize, &offset) ||
1076 			    lcp->cmdsize < sizeof(struct load_command) ||
1077 			    offset > cmds_size) {
1078 				ret = LOAD_BADMACHO;
1079 				break;
1080 			}
1081 
1082 			/*
1083 			 * Act on struct load_command's for which kernel
1084 			 * intervention is required.
1085 			 * Note that each load command implementation is expected to validate
1086 			 * that lcp->cmdsize is large enough to fit its specific struct type
1087 			 * before dereferencing fields not covered by struct load_command.
1088 			 */
1089 			switch (lcp->cmd) {
1090 			case LC_SEGMENT: {
1091 				struct segment_command *scp = (struct segment_command *) lcp;
1092 				if (scp->cmdsize < sizeof(*scp)) {
1093 					ret = LOAD_BADMACHO;
1094 					break;
1095 				}
1096 				if (pass == 0) {
1097 					if (is_dyld && scp->vmaddr == 0 && scp->fileoff == 0) {
1098 						dyld_no_load_addr = TRUE;
1099 						if (!slide_realign) {
1100 							/* got what we need, bail early on pass 0 */
1101 							continue;
1102 						}
1103 					}
1104 
1105 #if __arm64__
1106 					assert(!abi64);
1107 
1108 					if (scp->initprot == 0 && scp->maxprot == 0 && scp->vmaddr == 0) {
1109 						/* PAGEZERO */
1110 						if (os_add3_overflow(scp->vmaddr, scp->vmsize, slide, &pagezero_end) || pagezero_end > UINT32_MAX) {
1111 							ret = LOAD_BADMACHO;
1112 							break;
1113 						}
1114 					}
1115 					if (scp->initprot & VM_PROT_EXECUTE) {
1116 						/* TEXT */
1117 						if (os_add3_overflow(scp->vmaddr, scp->vmsize, slide, &executable_end) || executable_end > UINT32_MAX) {
1118 							ret = LOAD_BADMACHO;
1119 							break;
1120 						}
1121 					}
1122 					if (scp->initprot & VM_PROT_WRITE) {
1123 						/* DATA */
1124 						if (os_add_overflow(scp->vmaddr, slide, &writable_start) || writable_start > UINT32_MAX) {
1125 							ret = LOAD_BADMACHO;
1126 							break;
1127 						}
1128 					}
1129 #endif /* __arm64__ */
1130 					break;
1131 				}
1132 
1133 				if (pass == 1 && !strncmp(scp->segname, "__XHDR", sizeof(scp->segname))) {
1134 					found_xhdr = TRUE;
1135 				}
1136 
1137 				if (pass != 2) {
1138 					break;
1139 				}
1140 
1141 				if (abi64) {
1142 					/*
1143 					 * Having an LC_SEGMENT command for the
1144 					 * wrong ABI is invalid <rdar://problem/11021230>
1145 					 */
1146 					ret = LOAD_BADMACHO;
1147 					break;
1148 				}
1149 
1150 				ret = load_segment(lcp,
1151 				    header->filetype,
1152 				    control,
1153 				    file_offset,
1154 				    macho_size,
1155 				    vp,
1156 				    map,
1157 				    slide,
1158 				    result,
1159 				    imgp);
1160 				if (ret == LOAD_SUCCESS && scp->fileoff == 0 && scp->filesize > 0) {
1161 					/* Enforce a single segment mapping offset zero, with R+X
1162 					 * protection. */
1163 					if (found_header_segment ||
1164 					    ((scp->initprot & (VM_PROT_READ | VM_PROT_EXECUTE)) != (VM_PROT_READ | VM_PROT_EXECUTE))) {
1165 						ret = LOAD_BADMACHO;
1166 						break;
1167 					}
1168 					found_header_segment = TRUE;
1169 				}
1170 
1171 				break;
1172 			}
1173 			case LC_SEGMENT_64: {
1174 				struct segment_command_64 *scp64 = (struct segment_command_64 *) lcp;
1175 				if (scp64->cmdsize < sizeof(*scp64)) {
1176 					ret = LOAD_BADMACHO;
1177 					break;
1178 				}
1179 				if (pass == 0) {
1180 					if (is_dyld && scp64->vmaddr == 0 && scp64->fileoff == 0) {
1181 						dyld_no_load_addr = TRUE;
1182 					}
1183 					/* got what we need, bail early on pass 0 */
1184 					continue;
1185 				}
1186 
1187 				if (pass == 1 && !strncmp(scp64->segname, "__XHDR", sizeof(scp64->segname))) {
1188 					found_xhdr = TRUE;
1189 				}
1190 
1191 				if (pass != 2) {
1192 					break;
1193 				}
1194 
1195 				if (!abi64) {
1196 					/*
1197 					 * Having an LC_SEGMENT_64 command for the
1198 					 * wrong ABI is invalid <rdar://problem/11021230>
1199 					 */
1200 					ret = LOAD_BADMACHO;
1201 					break;
1202 				}
1203 
1204 				ret = load_segment(lcp,
1205 				    header->filetype,
1206 				    control,
1207 				    file_offset,
1208 				    macho_size,
1209 				    vp,
1210 				    map,
1211 				    slide,
1212 				    result,
1213 				    imgp);
1214 
1215 				if (ret == LOAD_SUCCESS && scp64->fileoff == 0 && scp64->filesize > 0) {
1216 					/* Enforce a single segment mapping offset zero, with R+X
1217 					 * protection. */
1218 					if (found_header_segment ||
1219 					    ((scp64->initprot & (VM_PROT_READ | VM_PROT_EXECUTE)) != (VM_PROT_READ | VM_PROT_EXECUTE))) {
1220 						ret = LOAD_BADMACHO;
1221 						break;
1222 					}
1223 					found_header_segment = TRUE;
1224 				}
1225 
1226 				break;
1227 			}
1228 			case LC_UNIXTHREAD: {
1229 				boolean_t is_x86_64_compat_binary = FALSE;
1230 				if (pass != 1) {
1231 					break;
1232 				}
1233 #if CONFIG_ROSETTA
1234 				if (depth == 2 && (imgp->ip_flags & IMGPF_ROSETTA)) {
1235 					// Ignore dyld, Rosetta will parse it's load commands to get the
1236 					// entry point.
1237 					result->validentry = 1;
1238 					break;
1239 				}
1240 #endif
1241 				ret = load_unixthread(
1242 					(struct thread_command *) lcp,
1243 					thread,
1244 					slide,
1245 					is_x86_64_compat_binary,
1246 					result);
1247 				break;
1248 			}
1249 			case LC_MAIN:
1250 				if (pass != 1) {
1251 					break;
1252 				}
1253 				if (depth != 1) {
1254 					break;
1255 				}
1256 				ret = load_main(
1257 					(struct entry_point_command *) lcp,
1258 					thread,
1259 					slide,
1260 					result);
1261 				break;
1262 			case LC_LOAD_DYLINKER:
1263 				if (pass != 3) {
1264 					break;
1265 				}
1266 				if ((depth == 1) && (dlp == 0)) {
1267 					dlp = (struct dylinker_command *)lcp;
1268 				} else {
1269 					ret = LOAD_FAILURE;
1270 				}
1271 				break;
1272 			case LC_UUID:
1273 				if (pass == 1 && depth == 1) {
1274 					ret = load_uuid((struct uuid_command *) lcp,
1275 					    (char *)addr + cmds_size,
1276 					    result);
1277 				}
1278 				break;
1279 			case LC_CODE_SIGNATURE:
1280 				/* CODE SIGNING */
1281 				if (pass != 1) {
1282 					break;
1283 				}
1284 
1285 				/* pager -> uip ->
1286 				 *  load signatures & store in uip
1287 				 *  set VM object "signed_pages"
1288 				 */
1289 				ret = load_code_signature(
1290 					(struct linkedit_data_command *) lcp,
1291 					vp,
1292 					file_offset,
1293 					macho_size,
1294 					header->cputype,
1295 					header->cpusubtype,
1296 					result,
1297 					imgp);
1298 				if (ret != LOAD_SUCCESS) {
1299 					printf("proc %d: load code signature error %d "
1300 					    "for file \"%s\"\n",
1301 					    proc_getpid(p), ret, vp->v_name);
1302 					/*
1303 					 * Allow injections to be ignored on devices w/o enforcement enabled
1304 					 */
1305 					if (!cs_process_global_enforcement()) {
1306 						ret = LOAD_SUCCESS; /* ignore error */
1307 					}
1308 				} else {
1309 					got_code_signatures = TRUE;
1310 				}
1311 
1312 				if (got_code_signatures) {
1313 					unsigned tainted = CS_VALIDATE_TAINTED;
1314 					boolean_t valid = FALSE;
1315 					vm_size_t off = 0;
1316 
1317 
1318 					if (cs_debug > 10) {
1319 						printf("validating initial pages of %s\n", vp->v_name);
1320 					}
1321 
1322 					while (off < alloc_size && ret == LOAD_SUCCESS) {
1323 						tainted = CS_VALIDATE_TAINTED;
1324 
1325 						valid = cs_validate_range(vp,
1326 						    NULL,
1327 						    file_offset + off,
1328 						    (const void *)((uintptr_t)addr + off),
1329 						    MIN(PAGE_SIZE, cmds_size),
1330 						    &tainted);
1331 						if (!valid || (tainted & CS_VALIDATE_TAINTED)) {
1332 							if (cs_debug) {
1333 								printf("CODE SIGNING: %s[%d]: invalid initial page at offset %lld validated:%d tainted:%d csflags:0x%x\n",
1334 								    vp->v_name, proc_getpid(p), (long long)(file_offset + off), valid, tainted, result->csflags);
1335 							}
1336 							if (cs_process_global_enforcement() ||
1337 							    (result->csflags & (CS_HARD | CS_KILL | CS_ENFORCEMENT))) {
1338 								ret = LOAD_FAILURE;
1339 							}
1340 							result->csflags &= ~CS_VALID;
1341 						}
1342 						off += PAGE_SIZE;
1343 					}
1344 				}
1345 
1346 				break;
1347 #if CONFIG_CODE_DECRYPTION
1348 			case LC_ENCRYPTION_INFO:
1349 			case LC_ENCRYPTION_INFO_64:
1350 				if (pass != 3) {
1351 					break;
1352 				}
1353 				ret = set_code_unprotect(
1354 					(struct encryption_info_command *) lcp,
1355 					addr, map, slide, vp, file_offset,
1356 					header->cputype, header->cpusubtype);
1357 				if (ret != LOAD_SUCCESS) {
1358 					os_reason_t load_failure_reason = OS_REASON_NULL;
1359 					printf("proc %d: set_code_unprotect() error %d "
1360 					    "for file \"%s\"\n",
1361 					    proc_getpid(p), ret, vp->v_name);
1362 					/*
1363 					 * Don't let the app run if it's
1364 					 * encrypted but we failed to set up the
1365 					 * decrypter. If the keys are missing it will
1366 					 * return LOAD_DECRYPTFAIL.
1367 					 */
1368 					if (ret == LOAD_DECRYPTFAIL) {
1369 						/* failed to load due to missing FP keys */
1370 						proc_lock(p);
1371 						p->p_lflag |= P_LTERM_DECRYPTFAIL;
1372 						proc_unlock(p);
1373 
1374 						KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1375 						    proc_getpid(p), OS_REASON_EXEC, EXEC_EXIT_REASON_FAIRPLAY_DECRYPT, 0, 0);
1376 						load_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_FAIRPLAY_DECRYPT);
1377 					} else {
1378 						KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1379 						    proc_getpid(p), OS_REASON_EXEC, EXEC_EXIT_REASON_DECRYPT, 0, 0);
1380 						load_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_DECRYPT);
1381 					}
1382 
1383 					/*
1384 					 * Don't signal the process if it was forked and in a partially constructed
1385 					 * state as part of a spawn -- it will just be torn down when the exec fails.
1386 					 */
1387 					if (!spawn) {
1388 						assert(load_failure_reason != OS_REASON_NULL);
1389 						{
1390 							psignal_with_reason(current_proc(), SIGKILL, load_failure_reason);
1391 							load_failure_reason = OS_REASON_NULL;
1392 						}
1393 					} else {
1394 						os_reason_free(load_failure_reason);
1395 						load_failure_reason = OS_REASON_NULL;
1396 					}
1397 				}
1398 				break;
1399 #endif
1400 			case LC_VERSION_MIN_IPHONEOS:
1401 			case LC_VERSION_MIN_MACOSX:
1402 			case LC_VERSION_MIN_WATCHOS:
1403 			case LC_VERSION_MIN_TVOS: {
1404 				struct version_min_command *vmc;
1405 
1406 				if (depth != 1 || pass != 0) {
1407 					break;
1408 				}
1409 				vmc = (struct version_min_command *) lcp;
1410 				ret = load_version(vmc, &found_version_cmd, imgp, result);
1411 #if XNU_TARGET_OS_OSX
1412 				if (ret == LOAD_SUCCESS) {
1413 					if (result->ip_platform == PLATFORM_IOS) {
1414 						vm_map_mark_alien(map);
1415 					} else {
1416 						assert(!vm_map_is_alien(map));
1417 					}
1418 				}
1419 #endif /* XNU_TARGET_OS_OSX */
1420 				break;
1421 			}
1422 			case LC_BUILD_VERSION: {
1423 				if (depth != 1 || pass != 0) {
1424 					break;
1425 				}
1426 				struct build_version_command* bvc = (struct build_version_command*)lcp;
1427 				if (bvc->cmdsize < sizeof(*bvc)) {
1428 					ret = LOAD_BADMACHO;
1429 					break;
1430 				}
1431 				if (found_version_cmd == TRUE) {
1432 					ret = LOAD_BADMACHO;
1433 					break;
1434 				}
1435 				result->ip_platform = bvc->platform;
1436 				result->lr_sdk = bvc->sdk;
1437 				result->lr_min_sdk = bvc->minos;
1438 				found_version_cmd = TRUE;
1439 #if XNU_TARGET_OS_OSX
1440 				if (result->ip_platform == PLATFORM_IOS) {
1441 					vm_map_mark_alien(map);
1442 				} else {
1443 					assert(!vm_map_is_alien(map));
1444 				}
1445 #endif /* XNU_TARGET_OS_OSX */
1446 				break;
1447 			}
1448 			default:
1449 				/* Other commands are ignored by the kernel */
1450 				ret = LOAD_SUCCESS;
1451 				break;
1452 			}
1453 			if (ret != LOAD_SUCCESS) {
1454 				break;
1455 			}
1456 		}
1457 		if (ret != LOAD_SUCCESS) {
1458 			break;
1459 		}
1460 	}
1461 
1462 	if (ret == LOAD_SUCCESS) {
1463 		if (!got_code_signatures && cs_process_global_enforcement()) {
1464 			ret = LOAD_FAILURE;
1465 		}
1466 
1467 		/* Make sure if we need dyld, we got it */
1468 		if (result->needs_dynlinker && !dlp) {
1469 			ret = LOAD_FAILURE;
1470 		}
1471 
1472 		if ((ret == LOAD_SUCCESS) && (dlp != 0)) {
1473 			/*
1474 			 * load the dylinker, and slide it by the independent DYLD ASLR
1475 			 * offset regardless of the PIE-ness of the main binary.
1476 			 */
1477 			ret = load_dylinker(dlp, header->cputype, map, thread, depth,
1478 			    dyld_aslr_offset, result, imgp);
1479 		}
1480 
1481 #if CONFIG_ROSETTA
1482 		if ((ret == LOAD_SUCCESS) && (depth == 1) && (imgp->ip_flags & IMGPF_ROSETTA)) {
1483 			ret = load_rosetta(map, thread, result, imgp);
1484 			if (ret == LOAD_SUCCESS) {
1485 				if (result->user_stack_alloc_size != 0) {
1486 					// If a stack allocation is required then add a 4gb gap after the main
1487 					// binary/dyld for the worst case static translation size.
1488 					mach_vm_size_t reserved_aot_size = 0x100000000;
1489 					vm_map_offset_t mask = vm_map_page_mask(map);
1490 
1491 					mach_vm_address_t vm_end;
1492 					if (dlp != 0) {
1493 						vm_end = vm_map_round_page(result->dynlinker_max_vm_addr, mask);
1494 					} else {
1495 						vm_end = vm_map_round_page(result->max_vm_addr, mask);
1496 					}
1497 
1498 					mach_vm_size_t user_stack_size = vm_map_round_page(result->user_stack_alloc_size, mask);
1499 					result->user_stack = vm_map_round_page(vm_end + user_stack_size + reserved_aot_size + slide, mask);
1500 				}
1501 			}
1502 		}
1503 #endif
1504 
1505 		if ((ret == LOAD_SUCCESS) && (depth == 1)) {
1506 			if (result->thread_count == 0) {
1507 				ret = LOAD_FAILURE;
1508 			}
1509 #if CONFIG_ENFORCE_SIGNED_CODE
1510 			if (!(result->csflags & CS_NO_UNTRUSTED_HELPERS)) {
1511 				ret = LOAD_FAILURE;
1512 			}
1513 #endif
1514 		}
1515 	}
1516 
1517 	if (ret == LOAD_BADMACHO && found_xhdr) {
1518 		ret = LOAD_BADMACHO_UPX;
1519 	}
1520 
1521 	kfree_data(addr, alloc_size);
1522 
1523 	return ret;
1524 }
1525 
1526 load_return_t
validate_potential_simulator_binary(cpu_type_t exectype __unused,struct image_params * imgp __unused,off_t file_offset __unused,off_t macho_size __unused)1527 validate_potential_simulator_binary(
1528 	cpu_type_t               exectype __unused,
1529 	struct image_params      *imgp __unused,
1530 	off_t                    file_offset __unused,
1531 	off_t                    macho_size __unused)
1532 {
1533 #if __x86_64__
1534 	/* Allow 32 bit exec only for simulator binaries */
1535 	if (bootarg_no32exec && imgp != NULL && exectype == CPU_TYPE_X86) {
1536 		if (imgp->ip_simulator_binary == IMGPF_SB_DEFAULT) {
1537 			boolean_t simulator_binary = check_if_simulator_binary(imgp, file_offset, macho_size);
1538 			imgp->ip_simulator_binary = simulator_binary ? IMGPF_SB_TRUE : IMGPF_SB_FALSE;
1539 		}
1540 
1541 		if (imgp->ip_simulator_binary != IMGPF_SB_TRUE) {
1542 			return LOAD_BADARCH;
1543 		}
1544 	}
1545 #endif
1546 	return LOAD_SUCCESS;
1547 }
1548 
1549 #if __x86_64__
1550 static boolean_t
check_if_simulator_binary(struct image_params * imgp,off_t file_offset,off_t macho_size)1551 check_if_simulator_binary(
1552 	struct image_params     *imgp,
1553 	off_t                   file_offset,
1554 	off_t                   macho_size)
1555 {
1556 	struct mach_header      *header;
1557 	char                    *ip_vdata = NULL;
1558 	kauth_cred_t            cred = NULL;
1559 	uint32_t                ncmds;
1560 	struct load_command     *lcp;
1561 	boolean_t               simulator_binary = FALSE;
1562 	void *                  addr = NULL;
1563 	vm_size_t               alloc_size, cmds_size;
1564 	size_t                  offset;
1565 	proc_t                  p = current_proc();             /* XXXX */
1566 	int                     error;
1567 	int                     resid = 0;
1568 	size_t                  mach_header_sz = sizeof(struct mach_header);
1569 
1570 
1571 	cred =  kauth_cred_proc_ref(p);
1572 
1573 	/* Allocate page to copyin mach header */
1574 	ip_vdata = kalloc_data(PAGE_SIZE, Z_WAITOK | Z_ZERO);
1575 	if (ip_vdata == NULL) {
1576 		goto bad;
1577 	}
1578 
1579 	/* Read the Mach-O header */
1580 	error = vn_rdwr(UIO_READ, imgp->ip_vp, ip_vdata,
1581 	    PAGE_SIZE, file_offset,
1582 	    UIO_SYSSPACE, (IO_UNIT | IO_NODELOCKED),
1583 	    cred, &resid, p);
1584 	if (error) {
1585 		goto bad;
1586 	}
1587 
1588 	header = (struct mach_header *)ip_vdata;
1589 
1590 	if (header->magic == MH_MAGIC_64 ||
1591 	    header->magic == MH_CIGAM_64) {
1592 		mach_header_sz = sizeof(struct mach_header_64);
1593 	}
1594 
1595 	/* ensure header + sizeofcmds falls within the file */
1596 	if (os_add_overflow(mach_header_sz, header->sizeofcmds, &cmds_size) ||
1597 	    (off_t)cmds_size > macho_size ||
1598 	    round_page_overflow(cmds_size, &alloc_size) ||
1599 	    alloc_size > INT_MAX) {
1600 		goto bad;
1601 	}
1602 
1603 	/*
1604 	 * Map the load commands into kernel memory.
1605 	 */
1606 	addr = kalloc_data(alloc_size, Z_WAITOK);
1607 	if (addr == NULL) {
1608 		goto bad;
1609 	}
1610 
1611 	error = vn_rdwr(UIO_READ, imgp->ip_vp, addr, (int)alloc_size, file_offset,
1612 	    UIO_SYSSPACE, IO_NODELOCKED, cred, &resid, p);
1613 	if (error) {
1614 		goto bad;
1615 	}
1616 
1617 	if (resid) {
1618 		/* We must be able to read in as much as the mach_header indicated */
1619 		goto bad;
1620 	}
1621 
1622 	/*
1623 	 * Loop through each of the load_commands indicated by the
1624 	 * Mach-O header; if an absurd value is provided, we just
1625 	 * run off the end of the reserved section by incrementing
1626 	 * the offset too far, so we are implicitly fail-safe.
1627 	 */
1628 	offset = mach_header_sz;
1629 	ncmds = header->ncmds;
1630 
1631 	while (ncmds--) {
1632 		/* ensure enough space for a minimal load command */
1633 		if (offset + sizeof(struct load_command) > cmds_size) {
1634 			break;
1635 		}
1636 
1637 		/*
1638 		 *	Get a pointer to the command.
1639 		 */
1640 		lcp = (struct load_command *)((uintptr_t)addr + offset);
1641 
1642 		/*
1643 		 * Perform prevalidation of the struct load_command
1644 		 * before we attempt to use its contents.  Invalid
1645 		 * values are ones which result in an overflow, or
1646 		 * which can not possibly be valid commands, or which
1647 		 * straddle or exist past the reserved section at the
1648 		 * start of the image.
1649 		 */
1650 		if (os_add_overflow(offset, lcp->cmdsize, &offset) ||
1651 		    lcp->cmdsize < sizeof(struct load_command) ||
1652 		    offset > cmds_size) {
1653 			break;
1654 		}
1655 
1656 		/* Check if its a simulator binary. */
1657 		switch (lcp->cmd) {
1658 		case LC_VERSION_MIN_WATCHOS:
1659 			simulator_binary = TRUE;
1660 			break;
1661 
1662 		case LC_BUILD_VERSION: {
1663 			struct build_version_command *bvc;
1664 
1665 			bvc = (struct build_version_command *) lcp;
1666 			if (bvc->cmdsize < sizeof(*bvc)) {
1667 				/* unsafe to use this command struct if cmdsize
1668 				* validated above is too small for it to fit */
1669 				break;
1670 			}
1671 			if (bvc->platform == PLATFORM_IOSSIMULATOR ||
1672 			    bvc->platform == PLATFORM_WATCHOSSIMULATOR) {
1673 				simulator_binary = TRUE;
1674 			}
1675 
1676 			break;
1677 		}
1678 
1679 		case LC_VERSION_MIN_IPHONEOS: {
1680 			simulator_binary = TRUE;
1681 			break;
1682 		}
1683 
1684 		default:
1685 			/* ignore other load commands */
1686 			break;
1687 		}
1688 
1689 		if (simulator_binary == TRUE) {
1690 			break;
1691 		}
1692 	}
1693 
1694 bad:
1695 	if (ip_vdata) {
1696 		kfree_data(ip_vdata, PAGE_SIZE);
1697 	}
1698 
1699 	if (cred) {
1700 		kauth_cred_unref(&cred);
1701 	}
1702 
1703 	if (addr) {
1704 		kfree_data(addr, alloc_size);
1705 	}
1706 
1707 	return simulator_binary;
1708 }
1709 #endif /* __x86_64__ */
1710 
1711 #if CONFIG_CODE_DECRYPTION
1712 
1713 #define APPLE_UNPROTECTED_HEADER_SIZE   (3 * 4096)
1714 
1715 static load_return_t
unprotect_dsmos_segment(uint64_t file_off,uint64_t file_size,struct vnode * vp,off_t macho_offset,vm_map_t map,vm_map_offset_t map_addr,vm_map_size_t map_size)1716 unprotect_dsmos_segment(
1717 	uint64_t        file_off,
1718 	uint64_t        file_size,
1719 	struct vnode    *vp,
1720 	off_t           macho_offset,
1721 	vm_map_t        map,
1722 	vm_map_offset_t map_addr,
1723 	vm_map_size_t   map_size)
1724 {
1725 	kern_return_t   kr;
1726 	uint64_t        slice_off;
1727 
1728 	/*
1729 	 * The first APPLE_UNPROTECTED_HEADER_SIZE bytes (from offset 0 of
1730 	 * this part of a Universal binary) are not protected...
1731 	 * The rest needs to be "transformed".
1732 	 */
1733 	slice_off = file_off - macho_offset;
1734 	if (slice_off <= APPLE_UNPROTECTED_HEADER_SIZE &&
1735 	    slice_off + file_size <= APPLE_UNPROTECTED_HEADER_SIZE) {
1736 		/* it's all unprotected, nothing to do... */
1737 		kr = KERN_SUCCESS;
1738 	} else {
1739 		if (slice_off <= APPLE_UNPROTECTED_HEADER_SIZE) {
1740 			/*
1741 			 * We start mapping in the unprotected area.
1742 			 * Skip the unprotected part...
1743 			 */
1744 			uint64_t delta_file;
1745 			vm_map_offset_t delta_map;
1746 
1747 			delta_file = (uint64_t)APPLE_UNPROTECTED_HEADER_SIZE;
1748 			delta_file -= slice_off;
1749 			if (os_convert_overflow(delta_file, &delta_map)) {
1750 				return LOAD_BADMACHO;
1751 			}
1752 			if (os_add_overflow(map_addr, delta_map, &map_addr)) {
1753 				return LOAD_BADMACHO;
1754 			}
1755 			if (os_sub_overflow(map_size, delta_map, &map_size)) {
1756 				return LOAD_BADMACHO;
1757 			}
1758 		}
1759 		/* ... transform the rest of the mapping. */
1760 		struct pager_crypt_info crypt_info;
1761 		crypt_info.page_decrypt = dsmos_page_transform;
1762 		crypt_info.crypt_ops = NULL;
1763 		crypt_info.crypt_end = NULL;
1764 #pragma unused(vp, macho_offset)
1765 		crypt_info.crypt_ops = (void *)0x2e69cf40;
1766 		vm_map_offset_t crypto_backing_offset;
1767 		crypto_backing_offset = -1; /* i.e. use map entry's offset */
1768 #if VM_MAP_DEBUG_APPLE_PROTECT
1769 		if (vm_map_debug_apple_protect) {
1770 			struct proc *p;
1771 			p = current_proc();
1772 			printf("APPLE_PROTECT: %d[%s] map %p "
1773 			    "[0x%llx:0x%llx] %s(%s)\n",
1774 			    proc_getpid(p), p->p_comm, map,
1775 			    (uint64_t) map_addr,
1776 			    (uint64_t) (map_addr + map_size),
1777 			    __FUNCTION__, vp->v_name);
1778 		}
1779 #endif /* VM_MAP_DEBUG_APPLE_PROTECT */
1780 
1781 		/* The DSMOS pager can only be used by apple signed code */
1782 		struct cs_blob * blob = csvnode_get_blob(vp, file_off);
1783 		if (blob == NULL || !blob->csb_platform_binary || blob->csb_platform_path) {
1784 			return LOAD_FAILURE;
1785 		}
1786 
1787 		kr = vm_map_apple_protected(map,
1788 		    map_addr,
1789 		    map_addr + map_size,
1790 		    crypto_backing_offset,
1791 		    &crypt_info,
1792 		    CRYPTID_APP_ENCRYPTION);
1793 	}
1794 
1795 	if (kr != KERN_SUCCESS) {
1796 		return LOAD_FAILURE;
1797 	}
1798 	return LOAD_SUCCESS;
1799 }
1800 #else   /* CONFIG_CODE_DECRYPTION */
1801 static load_return_t
unprotect_dsmos_segment(__unused uint64_t file_off,__unused uint64_t file_size,__unused struct vnode * vp,__unused off_t macho_offset,__unused vm_map_t map,__unused vm_map_offset_t map_addr,__unused vm_map_size_t map_size)1802 unprotect_dsmos_segment(
1803 	__unused        uint64_t        file_off,
1804 	__unused        uint64_t        file_size,
1805 	__unused        struct vnode    *vp,
1806 	__unused        off_t           macho_offset,
1807 	__unused        vm_map_t        map,
1808 	__unused        vm_map_offset_t map_addr,
1809 	__unused        vm_map_size_t   map_size)
1810 {
1811 	return LOAD_SUCCESS;
1812 }
1813 #endif  /* CONFIG_CODE_DECRYPTION */
1814 
1815 
1816 /*
1817  * map_segment:
1818  *	Maps a Mach-O segment.
1819  */
1820 static kern_return_t
map_segment(vm_map_t map,vm_map_offset_t vm_start,vm_map_offset_t vm_end,vm_map_kernel_flags_t vmk_flags,memory_object_control_t control,vm_map_offset_t file_start,vm_map_offset_t file_end,vm_prot_t initprot,vm_prot_t maxprot,load_result_t * result)1821 map_segment(
1822 	vm_map_t                map,
1823 	vm_map_offset_t         vm_start,
1824 	vm_map_offset_t         vm_end,
1825 	vm_map_kernel_flags_t   vmk_flags,
1826 	memory_object_control_t control,
1827 	vm_map_offset_t         file_start,
1828 	vm_map_offset_t         file_end,
1829 	vm_prot_t               initprot,
1830 	vm_prot_t               maxprot,
1831 	load_result_t           *result)
1832 {
1833 	kern_return_t   ret;
1834 	vm_map_offset_t effective_page_mask;
1835 
1836 	if (vm_end < vm_start ||
1837 	    file_end < file_start) {
1838 		return LOAD_BADMACHO;
1839 	}
1840 	if (vm_end == vm_start ||
1841 	    file_end == file_start) {
1842 		/* nothing to map... */
1843 		return LOAD_SUCCESS;
1844 	}
1845 
1846 	effective_page_mask = vm_map_page_mask(map);
1847 
1848 	if (vm_map_page_aligned(vm_start, effective_page_mask) &&
1849 	    vm_map_page_aligned(vm_end, effective_page_mask) &&
1850 	    vm_map_page_aligned(file_start, effective_page_mask) &&
1851 	    vm_map_page_aligned(file_end, effective_page_mask)) {
1852 		/* all page-aligned and map-aligned: proceed */
1853 	} else {
1854 		/*
1855 		 * There's no more fourk_pager to handle mis-alignments;
1856 		 * all binaries should be page-aligned and map-aligned
1857 		 */
1858 		return LOAD_BADMACHO;
1859 	}
1860 
1861 #if !defined(XNU_TARGET_OS_OSX)
1862 	(void) result;
1863 #else /* !defined(XNU_TARGET_OS_OSX) */
1864 	/*
1865 	 * This process doesn't have its new csflags (from
1866 	 * the image being loaded) yet, so tell VM to override the
1867 	 * current process's CS_ENFORCEMENT for this mapping.
1868 	 */
1869 	if (result->csflags & CS_ENFORCEMENT) {
1870 		vmk_flags.vmkf_cs_enforcement = TRUE;
1871 	} else {
1872 		vmk_flags.vmkf_cs_enforcement = FALSE;
1873 	}
1874 	vmk_flags.vmkf_cs_enforcement_override = TRUE;
1875 #endif /* !defined(XNU_TARGET_OS_OSX) */
1876 
1877 	if (result->is_rosetta && (initprot & VM_PROT_EXECUTE) == VM_PROT_EXECUTE) {
1878 		vmk_flags.vmkf_translated_allow_execute = TRUE;
1879 	}
1880 
1881 	if (control != MEMORY_OBJECT_CONTROL_NULL) {
1882 		/* no copy-on-read for mapped binaries */
1883 		vmk_flags.vmkf_no_copy_on_read = 1;
1884 		ret = vm_map_enter_mem_object_control(
1885 			map,
1886 			&vm_start,
1887 			file_end - file_start,
1888 			(mach_vm_offset_t)0,
1889 			vmk_flags,
1890 			control,
1891 			file_start,
1892 			TRUE, /* copy */
1893 			initprot, maxprot,
1894 			VM_INHERIT_DEFAULT);
1895 	} else {
1896 		ret = mach_vm_map_kernel(
1897 			map,
1898 			&vm_start,
1899 			file_end - file_start,
1900 			(mach_vm_offset_t)0,
1901 			vmk_flags,
1902 			IPC_PORT_NULL,
1903 			0, /* offset */
1904 			TRUE, /* copy */
1905 			initprot, maxprot,
1906 			VM_INHERIT_DEFAULT);
1907 	}
1908 	if (ret != KERN_SUCCESS) {
1909 		return LOAD_NOSPACE;
1910 	}
1911 	return LOAD_SUCCESS;
1912 }
1913 
1914 static
1915 load_return_t
load_segment(struct load_command * lcp,uint32_t filetype,void * control,off_t pager_offset,off_t macho_size,struct vnode * vp,vm_map_t map,int64_t slide,load_result_t * result,struct image_params * imgp)1916 load_segment(
1917 	struct load_command     *lcp,
1918 	uint32_t                filetype,
1919 	void *                  control,
1920 	off_t                   pager_offset,
1921 	off_t                   macho_size,
1922 	struct vnode            *vp,
1923 	vm_map_t                map,
1924 	int64_t                 slide,
1925 	load_result_t           *result,
1926 	struct image_params     *imgp)
1927 {
1928 	struct segment_command_64 segment_command, *scp;
1929 	kern_return_t           ret;
1930 	vm_map_size_t           delta_size;
1931 	vm_prot_t               initprot;
1932 	vm_prot_t               maxprot;
1933 	size_t                  segment_command_size, total_section_size,
1934 	    single_section_size;
1935 	uint64_t                file_offset, file_size;
1936 	vm_map_offset_t         vm_offset;
1937 	size_t                  vm_size;
1938 	vm_map_offset_t         vm_start, vm_end, vm_end_aligned;
1939 	vm_map_offset_t         file_start, file_end;
1940 	vm_map_kernel_flags_t   vmk_flags;
1941 	kern_return_t           kr;
1942 	boolean_t               verbose;
1943 	vm_map_size_t           effective_page_size;
1944 	vm_map_offset_t         effective_page_mask;
1945 #if __arm64__
1946 	boolean_t               fourk_align;
1947 #endif /* __arm64__ */
1948 
1949 	(void)imgp;
1950 
1951 	effective_page_size = vm_map_page_size(map);
1952 	effective_page_mask = vm_map_page_mask(map);
1953 	vmk_flags = VM_MAP_KERNEL_FLAGS_FIXED();
1954 
1955 	verbose = FALSE;
1956 	if (LC_SEGMENT_64 == lcp->cmd) {
1957 		segment_command_size = sizeof(struct segment_command_64);
1958 		single_section_size  = sizeof(struct section_64);
1959 #if __arm64__
1960 		/* 64-bit binary: should already be 16K-aligned */
1961 		fourk_align = FALSE;
1962 
1963 		if (vm_map_page_shift(map) == FOURK_PAGE_SHIFT &&
1964 		    PAGE_SHIFT != FOURK_PAGE_SHIFT) {
1965 			fourk_align = TRUE;
1966 			verbose = TRUE;
1967 		}
1968 #endif /* __arm64__ */
1969 	} else {
1970 		segment_command_size = sizeof(struct segment_command);
1971 		single_section_size  = sizeof(struct section);
1972 #if __arm64__
1973 		/* 32-bit binary or arm64_32 binary: should already be page-aligned */
1974 		fourk_align = FALSE;
1975 #endif /* __arm64__ */
1976 	}
1977 	if (lcp->cmdsize < segment_command_size) {
1978 		DEBUG4K_ERROR("LOAD_BADMACHO cmdsize %d < %zu\n", lcp->cmdsize, segment_command_size);
1979 		return LOAD_BADMACHO;
1980 	}
1981 	total_section_size = lcp->cmdsize - segment_command_size;
1982 
1983 	if (LC_SEGMENT_64 == lcp->cmd) {
1984 		scp = (struct segment_command_64 *)lcp;
1985 	} else {
1986 		scp = &segment_command;
1987 		widen_segment_command((struct segment_command *)lcp, scp);
1988 	}
1989 
1990 	if (verbose) {
1991 		MACHO_PRINTF(("+++ load_segment %s "
1992 		    "vm[0x%llx:0x%llx] file[0x%llx:0x%llx] "
1993 		    "prot %d/%d flags 0x%x\n",
1994 		    scp->segname,
1995 		    (uint64_t)(slide + scp->vmaddr),
1996 		    (uint64_t)(slide + scp->vmaddr + scp->vmsize),
1997 		    pager_offset + scp->fileoff,
1998 		    pager_offset + scp->fileoff + scp->filesize,
1999 		    scp->initprot,
2000 		    scp->maxprot,
2001 		    scp->flags));
2002 	}
2003 
2004 	/*
2005 	 * Make sure what we get from the file is really ours (as specified
2006 	 * by macho_size).
2007 	 */
2008 	if (scp->fileoff + scp->filesize < scp->fileoff ||
2009 	    scp->fileoff + scp->filesize > (uint64_t)macho_size) {
2010 		DEBUG4K_ERROR("LOAD_BADMACHO fileoff 0x%llx filesize 0x%llx macho_size 0x%llx\n", scp->fileoff, scp->filesize, (uint64_t)macho_size);
2011 		return LOAD_BADMACHO;
2012 	}
2013 	/*
2014 	 * Ensure that the number of sections specified would fit
2015 	 * within the load command size.
2016 	 */
2017 	if (total_section_size / single_section_size < scp->nsects) {
2018 		DEBUG4K_ERROR("LOAD_BADMACHO 0x%zx 0x%zx %d\n", total_section_size, single_section_size, scp->nsects);
2019 		return LOAD_BADMACHO;
2020 	}
2021 	/*
2022 	 * Make sure the segment is page-aligned in the file.
2023 	 */
2024 	if (os_add_overflow(pager_offset, scp->fileoff, &file_offset)) {
2025 		DEBUG4K_ERROR("LOAD_BADMACHO file_offset: 0x%llx + 0x%llx\n", pager_offset, scp->fileoff);
2026 		return LOAD_BADMACHO;
2027 	}
2028 	file_size = scp->filesize;
2029 #if __arm64__
2030 	if (fourk_align) {
2031 		if ((file_offset & FOURK_PAGE_MASK) != 0) {
2032 			/*
2033 			 * we can't mmap() it if it's not at least 4KB-aligned
2034 			 * in the file
2035 			 */
2036 			DEBUG4K_ERROR("LOAD_BADMACHO file_offset 0x%llx\n", file_offset);
2037 			return LOAD_BADMACHO;
2038 		}
2039 	} else
2040 #endif /* __arm64__ */
2041 	if ((file_offset & PAGE_MASK_64) != 0 ||
2042 	    /* we can't mmap() it if it's not page-aligned in the file */
2043 	    (file_offset & vm_map_page_mask(map)) != 0) {
2044 		/*
2045 		 * The 1st test would have failed if the system's page size
2046 		 * was what this process believe is the page size, so let's
2047 		 * fail here too for the sake of consistency.
2048 		 */
2049 		DEBUG4K_ERROR("LOAD_BADMACHO file_offset 0x%llx\n", file_offset);
2050 		return LOAD_BADMACHO;
2051 	}
2052 
2053 	/*
2054 	 * If we have a code signature attached for this slice
2055 	 * require that the segments are within the signed part
2056 	 * of the file.
2057 	 */
2058 	if (result->cs_end_offset &&
2059 	    result->cs_end_offset < (off_t)scp->fileoff &&
2060 	    result->cs_end_offset - scp->fileoff < scp->filesize) {
2061 		if (cs_debug) {
2062 			printf("section outside code signature\n");
2063 		}
2064 		DEBUG4K_ERROR("LOAD_BADMACHO end_offset 0x%llx fileoff 0x%llx filesize 0x%llx\n", result->cs_end_offset, scp->fileoff, scp->filesize);
2065 		return LOAD_BADMACHO;
2066 	}
2067 
2068 	if (os_add_overflow(scp->vmaddr, slide, &vm_offset)) {
2069 		if (cs_debug) {
2070 			printf("vmaddr too large\n");
2071 		}
2072 		DEBUG4K_ERROR("LOAD_BADMACHO vmaddr 0x%llx slide 0x%llx vm_offset 0x%llx\n", scp->vmaddr, slide, (uint64_t)vm_offset);
2073 		return LOAD_BADMACHO;
2074 	}
2075 
2076 	if (scp->vmsize > SIZE_MAX) {
2077 		DEBUG4K_ERROR("LOAD_BADMACHO vmsize 0x%llx\n", scp->vmsize);
2078 		return LOAD_BADMACHO;
2079 	}
2080 
2081 	vm_size = (size_t)scp->vmsize;
2082 
2083 	if (vm_size == 0) {
2084 		return LOAD_SUCCESS;
2085 	}
2086 	if (scp->vmaddr == 0 &&
2087 	    file_size == 0 &&
2088 	    vm_size != 0 &&
2089 	    (scp->initprot & VM_PROT_ALL) == VM_PROT_NONE &&
2090 	    (scp->maxprot & VM_PROT_ALL) == VM_PROT_NONE) {
2091 		if (map == VM_MAP_NULL) {
2092 			return LOAD_SUCCESS;
2093 		}
2094 
2095 		/*
2096 		 * For PIE, extend page zero rather than moving it.  Extending
2097 		 * page zero keeps early allocations from falling predictably
2098 		 * between the end of page zero and the beginning of the first
2099 		 * slid segment.
2100 		 */
2101 		/*
2102 		 * This is a "page zero" segment:  it starts at address 0,
2103 		 * is not mapped from the binary file and is not accessible.
2104 		 * User-space should never be able to access that memory, so
2105 		 * make it completely off limits by raising the VM map's
2106 		 * minimum offset.
2107 		 */
2108 		vm_end = (vm_map_offset_t)(vm_offset + vm_size);
2109 		if (vm_end < vm_offset) {
2110 			DEBUG4K_ERROR("LOAD_BADMACHO vm_end 0x%llx vm_offset 0x%llx vm_size 0x%llx\n", (uint64_t)vm_end, (uint64_t)vm_offset, (uint64_t)vm_size);
2111 			return LOAD_BADMACHO;
2112 		}
2113 
2114 		if (verbose) {
2115 			MACHO_PRINTF(("++++++ load_segment: "
2116 			    "page_zero up to 0x%llx\n",
2117 			    (uint64_t) vm_end));
2118 		}
2119 #if __arm64__
2120 		if (fourk_align) {
2121 			/* raise min_offset as much as page-alignment allows */
2122 			vm_end_aligned = vm_map_trunc_page(vm_end,
2123 			    effective_page_mask);
2124 		} else
2125 #endif /* __arm64__ */
2126 		{
2127 			vm_end = vm_map_round_page(vm_end,
2128 			    PAGE_MASK_64);
2129 			vm_end_aligned = vm_end;
2130 		}
2131 		ret = vm_map_raise_min_offset(map,
2132 		    vm_end_aligned);
2133 		if (ret != KERN_SUCCESS) {
2134 			DEBUG4K_ERROR("LOAD_FAILURE ret 0x%x\n", ret);
2135 			return LOAD_FAILURE;
2136 		}
2137 		return LOAD_SUCCESS;
2138 	} else {
2139 #if !defined(XNU_TARGET_OS_OSX)
2140 		/* not PAGEZERO: should not be mapped at address 0 */
2141 		if (filetype != MH_DYLINKER && (imgp->ip_flags & IMGPF_ROSETTA) == 0 && scp->vmaddr == 0) {
2142 			DEBUG4K_ERROR("LOAD_BADMACHO filetype %d vmaddr 0x%llx\n", filetype, scp->vmaddr);
2143 			return LOAD_BADMACHO;
2144 		}
2145 #endif /* !defined(XNU_TARGET_OS_OSX) */
2146 	}
2147 
2148 #if __arm64__
2149 	if (fourk_align) {
2150 		/* 4K-align */
2151 		file_start = vm_map_trunc_page(file_offset,
2152 		    FOURK_PAGE_MASK);
2153 		file_end = vm_map_round_page(file_offset + file_size,
2154 		    FOURK_PAGE_MASK);
2155 		vm_start = vm_map_trunc_page(vm_offset,
2156 		    FOURK_PAGE_MASK);
2157 		vm_end = vm_map_round_page(vm_offset + vm_size,
2158 		    FOURK_PAGE_MASK);
2159 
2160 		if (file_offset - file_start > FOURK_PAGE_MASK ||
2161 		    file_end - file_offset - file_size > FOURK_PAGE_MASK) {
2162 			DEBUG4K_ERROR("LOAD_BADMACHO file_start / file_size wrap "
2163 			    "[0x%llx:0x%llx] -> [0x%llx:0x%llx]\n",
2164 			    file_offset,
2165 			    file_offset + file_size,
2166 			    (uint64_t) file_start,
2167 			    (uint64_t) file_end);
2168 			return LOAD_BADMACHO;
2169 		}
2170 
2171 		if (!strncmp(scp->segname, "__LINKEDIT", 11) &&
2172 		    page_aligned(file_start) &&
2173 		    vm_map_page_aligned(file_start, vm_map_page_mask(map)) &&
2174 		    page_aligned(vm_start) &&
2175 		    vm_map_page_aligned(vm_start, vm_map_page_mask(map))) {
2176 			/* XXX last segment: ignore mis-aligned tail */
2177 			file_end = vm_map_round_page(file_end,
2178 			    effective_page_mask);
2179 			vm_end = vm_map_round_page(vm_end,
2180 			    effective_page_mask);
2181 		}
2182 	} else
2183 #endif /* __arm64__ */
2184 	{
2185 		file_start = vm_map_trunc_page(file_offset,
2186 		    effective_page_mask);
2187 		file_end = vm_map_round_page(file_offset + file_size,
2188 		    effective_page_mask);
2189 		vm_start = vm_map_trunc_page(vm_offset,
2190 		    effective_page_mask);
2191 		vm_end = vm_map_round_page(vm_offset + vm_size,
2192 		    effective_page_mask);
2193 
2194 		if (file_offset - file_start > effective_page_mask ||
2195 		    file_end - file_offset - file_size > effective_page_mask) {
2196 			DEBUG4K_ERROR("LOAD_BADMACHO file_start / file_size wrap "
2197 			    "[0x%llx:0x%llx] -> [0x%llx:0x%llx]\n",
2198 			    file_offset,
2199 			    file_offset + file_size,
2200 			    (uint64_t) file_start,
2201 			    (uint64_t) file_end);
2202 			return LOAD_BADMACHO;
2203 		}
2204 	}
2205 
2206 	if (vm_start < result->min_vm_addr) {
2207 		result->min_vm_addr = vm_start;
2208 	}
2209 	if (vm_end > result->max_vm_addr) {
2210 		result->max_vm_addr = vm_end;
2211 	}
2212 
2213 	if (map == VM_MAP_NULL) {
2214 		return LOAD_SUCCESS;
2215 	}
2216 
2217 	if (scp->flags & SG_READ_ONLY) {
2218 		/*
2219 		 * Record the VM start/end of a segment which should
2220 		 * be RO after fixups. Only __DATA_CONST should
2221 		 * have this flag.
2222 		 */
2223 		if (result->ro_vm_start != MACH_VM_MIN_ADDRESS ||
2224 		    result->ro_vm_end != MACH_VM_MIN_ADDRESS) {
2225 			DEBUG4K_ERROR("LOAD_BADMACHO segment flags [%x] "
2226 			    "multiple segments with SG_READ_ONLY flag\n",
2227 			    scp->flags);
2228 			return LOAD_BADMACHO;
2229 		}
2230 
2231 		result->ro_vm_start = vm_start;
2232 		result->ro_vm_end = vm_end;
2233 	}
2234 
2235 	if (vm_size > 0) {
2236 #if !__x86_64__
2237 		if (!strncmp(scp->segname, "__LINKEDIT", 11)) {
2238 			vmk_flags.vmf_permanent = true;
2239 		}
2240 #endif /* !__x86_64__ */
2241 		initprot = (scp->initprot) & VM_PROT_ALL;
2242 		maxprot = (scp->maxprot) & VM_PROT_ALL;
2243 		/*
2244 		 *	Map a copy of the file into the address space.
2245 		 */
2246 		if (verbose) {
2247 			MACHO_PRINTF(("++++++ load_segment: "
2248 			    "mapping at vm [0x%llx:0x%llx] of "
2249 			    "file [0x%llx:0x%llx]\n",
2250 			    (uint64_t) vm_start,
2251 			    (uint64_t) vm_end,
2252 			    (uint64_t) file_start,
2253 			    (uint64_t) file_end));
2254 		}
2255 		ret = map_segment(map,
2256 		    vm_start,
2257 		    vm_end,
2258 		    vmk_flags,
2259 		    control,
2260 		    file_start,
2261 		    file_end,
2262 		    initprot,
2263 		    maxprot,
2264 		    result);
2265 		if (ret) {
2266 			DEBUG4K_ERROR("LOAD_NOSPACE start 0x%llx end 0x%llx ret 0x%x\n", (uint64_t)vm_start, (uint64_t)vm_end, ret);
2267 			return LOAD_NOSPACE;
2268 		}
2269 
2270 #if FIXME
2271 		/*
2272 		 *	If the file didn't end on a page boundary,
2273 		 *	we need to zero the leftover.
2274 		 */
2275 		delta_size = map_size - scp->filesize;
2276 		if (delta_size > 0) {
2277 			void *tmp = kalloc_data(delta_size, Z_WAITOK | Z_ZERO);
2278 			int rc;
2279 
2280 			if (tmp == NULL) {
2281 				DEBUG4K_ERROR("LOAD_RESOURCE delta_size 0x%llx ret 0x%x\n", delta_size, ret);
2282 				return LOAD_RESOURCE;
2283 			}
2284 
2285 			rc = copyout(tmp, map_addr + scp->filesize, delta_size);
2286 			kfree_data(tmp, delta_size);
2287 
2288 			if (rc) {
2289 				DEBUG4K_ERROR("LOAD_FAILURE copyout 0x%llx 0x%llx\n", map_addr + scp->filesize, delta_size);
2290 				return LOAD_FAILURE;
2291 			}
2292 		}
2293 #endif /* FIXME */
2294 	}
2295 
2296 	/*
2297 	 *	If the virtual size of the segment is greater
2298 	 *	than the size from the file, we need to allocate
2299 	 *	zero fill memory for the rest.
2300 	 */
2301 	if ((vm_end - vm_start) > (file_end - file_start)) {
2302 		delta_size = (vm_end - vm_start) - (file_end - file_start);
2303 	} else {
2304 		delta_size = 0;
2305 	}
2306 	if (delta_size > 0) {
2307 		vm_map_offset_t tmp_start;
2308 		vm_map_offset_t tmp_end;
2309 
2310 		if (os_add_overflow(vm_start, file_end - file_start, &tmp_start)) {
2311 			DEBUG4K_ERROR("LOAD_NOSPACE tmp_start: 0x%llx + 0x%llx\n", (uint64_t)vm_start, (uint64_t)(file_end - file_start));
2312 			return LOAD_NOSPACE;
2313 		}
2314 
2315 		if (os_add_overflow(tmp_start, delta_size, &tmp_end)) {
2316 			DEBUG4K_ERROR("LOAD_NOSPACE tmp_end: 0x%llx + 0x%llx\n", (uint64_t)tmp_start, (uint64_t)delta_size);
2317 			return LOAD_NOSPACE;
2318 		}
2319 
2320 		if (verbose) {
2321 			MACHO_PRINTF(("++++++ load_segment: "
2322 			    "delta mapping vm [0x%llx:0x%llx]\n",
2323 			    (uint64_t) tmp_start,
2324 			    (uint64_t) tmp_end));
2325 		}
2326 		kr = map_segment(map,
2327 		    tmp_start,
2328 		    tmp_end,
2329 		    vmk_flags,
2330 		    MEMORY_OBJECT_CONTROL_NULL,
2331 		    0,
2332 		    delta_size,
2333 		    scp->initprot,
2334 		    scp->maxprot,
2335 		    result);
2336 		if (kr != KERN_SUCCESS) {
2337 			DEBUG4K_ERROR("LOAD_NOSPACE 0x%llx 0x%llx kr 0x%x\n", (unsigned long long)tmp_start, (uint64_t)delta_size, kr);
2338 			return LOAD_NOSPACE;
2339 		}
2340 	}
2341 
2342 	if ((scp->fileoff == 0) && (scp->filesize != 0)) {
2343 		result->mach_header = vm_offset;
2344 	}
2345 
2346 	if (scp->flags & SG_PROTECTED_VERSION_1) {
2347 		ret = unprotect_dsmos_segment(file_start,
2348 		    file_end - file_start,
2349 		    vp,
2350 		    pager_offset,
2351 		    map,
2352 		    vm_start,
2353 		    vm_end - vm_start);
2354 		if (ret != LOAD_SUCCESS) {
2355 			DEBUG4K_ERROR("unprotect 0x%llx 0x%llx ret %d \n", (uint64_t)vm_start, (uint64_t)vm_end, ret);
2356 			return ret;
2357 		}
2358 	} else {
2359 		ret = LOAD_SUCCESS;
2360 	}
2361 
2362 	if (LOAD_SUCCESS == ret &&
2363 	    filetype == MH_DYLINKER &&
2364 	    result->all_image_info_addr == MACH_VM_MIN_ADDRESS) {
2365 		note_all_image_info_section(scp,
2366 		    LC_SEGMENT_64 == lcp->cmd,
2367 		    single_section_size,
2368 		    ((const char *)lcp +
2369 		    segment_command_size),
2370 		    slide,
2371 		    result);
2372 	}
2373 
2374 	if (result->entry_point != MACH_VM_MIN_ADDRESS) {
2375 		if ((result->entry_point >= vm_offset) && (result->entry_point < (vm_offset + vm_size))) {
2376 			if ((scp->initprot & (VM_PROT_READ | VM_PROT_EXECUTE)) == (VM_PROT_READ | VM_PROT_EXECUTE)) {
2377 				result->validentry = 1;
2378 			} else {
2379 				/* right range but wrong protections, unset if previously validated */
2380 				result->validentry = 0;
2381 			}
2382 		}
2383 	}
2384 
2385 	if (ret != LOAD_SUCCESS && verbose) {
2386 		DEBUG4K_ERROR("ret %d\n", ret);
2387 	}
2388 	return ret;
2389 }
2390 
2391 static
2392 load_return_t
load_uuid(struct uuid_command * uulp,char * command_end,load_result_t * result)2393 load_uuid(
2394 	struct uuid_command     *uulp,
2395 	char                    *command_end,
2396 	load_result_t           *result
2397 	)
2398 {
2399 	/*
2400 	 * We need to check the following for this command:
2401 	 * - The command size should be atleast the size of struct uuid_command
2402 	 * - The UUID part of the command should be completely within the mach-o header
2403 	 */
2404 
2405 	if ((uulp->cmdsize < sizeof(struct uuid_command)) ||
2406 	    (((char *)uulp + sizeof(struct uuid_command)) > command_end)) {
2407 		return LOAD_BADMACHO;
2408 	}
2409 
2410 	memcpy(&result->uuid[0], &uulp->uuid[0], sizeof(result->uuid));
2411 	return LOAD_SUCCESS;
2412 }
2413 
2414 static
2415 load_return_t
load_version(struct version_min_command * vmc,boolean_t * found_version_cmd,struct image_params * imgp __unused,load_result_t * result)2416 load_version(
2417 	struct version_min_command     *vmc,
2418 	boolean_t               *found_version_cmd,
2419 	struct image_params             *imgp __unused,
2420 	load_result_t           *result
2421 	)
2422 {
2423 	uint32_t platform = 0;
2424 	uint32_t sdk;
2425 	uint32_t min_sdk;
2426 
2427 	if (vmc->cmdsize < sizeof(*vmc)) {
2428 		return LOAD_BADMACHO;
2429 	}
2430 	if (*found_version_cmd == TRUE) {
2431 		return LOAD_BADMACHO;
2432 	}
2433 	*found_version_cmd = TRUE;
2434 	sdk = vmc->sdk;
2435 	min_sdk = vmc->version;
2436 	switch (vmc->cmd) {
2437 	case LC_VERSION_MIN_MACOSX:
2438 		platform = PLATFORM_MACOS;
2439 		break;
2440 #if __x86_64__ /* __x86_64__ */
2441 	case LC_VERSION_MIN_IPHONEOS:
2442 		platform = PLATFORM_IOSSIMULATOR;
2443 		break;
2444 	case LC_VERSION_MIN_WATCHOS:
2445 		platform = PLATFORM_WATCHOSSIMULATOR;
2446 		break;
2447 	case LC_VERSION_MIN_TVOS:
2448 		platform = PLATFORM_TVOSSIMULATOR;
2449 		break;
2450 #else
2451 	case LC_VERSION_MIN_IPHONEOS: {
2452 #if __arm64__
2453 		if (vmc->sdk < (12 << 16)) {
2454 			/* app built with a pre-iOS12 SDK: apply legacy footprint mitigation */
2455 			result->legacy_footprint = TRUE;
2456 		}
2457 #endif /* __arm64__ */
2458 		platform = PLATFORM_IOS;
2459 		break;
2460 	}
2461 	case LC_VERSION_MIN_WATCHOS:
2462 		platform = PLATFORM_WATCHOS;
2463 		break;
2464 	case LC_VERSION_MIN_TVOS:
2465 		platform = PLATFORM_TVOS;
2466 		break;
2467 #endif /* __x86_64__ */
2468 	/* All LC_VERSION_MIN_* load commands are legacy and we will not be adding any more */
2469 	default:
2470 		sdk = (uint32_t)-1;
2471 		min_sdk = (uint32_t)-1;
2472 		__builtin_unreachable();
2473 	}
2474 	result->ip_platform = platform;
2475 	result->lr_min_sdk = min_sdk;
2476 	result->lr_sdk = sdk;
2477 	return LOAD_SUCCESS;
2478 }
2479 
2480 static
2481 load_return_t
load_main(struct entry_point_command * epc,thread_t thread,int64_t slide,load_result_t * result)2482 load_main(
2483 	struct entry_point_command      *epc,
2484 	thread_t                thread,
2485 	int64_t                         slide,
2486 	load_result_t           *result
2487 	)
2488 {
2489 	mach_vm_offset_t addr;
2490 	kern_return_t   ret;
2491 
2492 	if (epc->cmdsize < sizeof(*epc)) {
2493 		return LOAD_BADMACHO;
2494 	}
2495 	if (result->thread_count != 0) {
2496 		return LOAD_FAILURE;
2497 	}
2498 
2499 	if (thread == THREAD_NULL) {
2500 		return LOAD_SUCCESS;
2501 	}
2502 
2503 	/*
2504 	 * LC_MAIN specifies stack size but not location.
2505 	 * Add guard page to allocation size (MAXSSIZ includes guard page).
2506 	 */
2507 	if (epc->stacksize) {
2508 		if (os_add_overflow(epc->stacksize, 4 * PAGE_SIZE, &result->user_stack_size)) {
2509 			/*
2510 			 * We are going to immediately throw away this result, but we want
2511 			 * to make sure we aren't loading a dangerously close to
2512 			 * overflowing value, since this will have a guard page added to it
2513 			 * and be rounded to page boundaries
2514 			 */
2515 			return LOAD_BADMACHO;
2516 		}
2517 		result->user_stack_size = epc->stacksize;
2518 		if (os_add_overflow(epc->stacksize, PAGE_SIZE, &result->user_stack_alloc_size)) {
2519 			return LOAD_BADMACHO;
2520 		}
2521 		result->custom_stack = TRUE;
2522 	} else {
2523 		result->user_stack_alloc_size = MAXSSIZ;
2524 	}
2525 
2526 	/* use default location for stack */
2527 	ret = thread_userstackdefault(&addr, result->is_64bit_addr);
2528 	if (ret != KERN_SUCCESS) {
2529 		return LOAD_FAILURE;
2530 	}
2531 
2532 	/* The stack slides down from the default location */
2533 	result->user_stack = (user_addr_t)mach_vm_trunc_page((user_addr_t)addr - slide);
2534 
2535 	if (result->using_lcmain || result->entry_point != MACH_VM_MIN_ADDRESS) {
2536 		/* Already processed LC_MAIN or LC_UNIXTHREAD */
2537 		return LOAD_FAILURE;
2538 	}
2539 
2540 	/* kernel does *not* use entryoff from LC_MAIN.	 Dyld uses it. */
2541 	result->needs_dynlinker = TRUE;
2542 	result->using_lcmain = TRUE;
2543 
2544 	ret = thread_state_initialize( thread );
2545 	if (ret != KERN_SUCCESS) {
2546 		return LOAD_FAILURE;
2547 	}
2548 
2549 	result->unixproc = TRUE;
2550 	result->thread_count++;
2551 
2552 	return LOAD_SUCCESS;
2553 }
2554 
2555 static
2556 load_return_t
setup_driver_main(thread_t thread,int64_t slide,load_result_t * result)2557 setup_driver_main(
2558 	thread_t                thread,
2559 	int64_t                         slide,
2560 	load_result_t           *result
2561 	)
2562 {
2563 	mach_vm_offset_t addr;
2564 	kern_return_t   ret;
2565 
2566 	/* Driver binaries have no LC_MAIN, use defaults */
2567 
2568 	if (thread == THREAD_NULL) {
2569 		return LOAD_SUCCESS;
2570 	}
2571 
2572 	result->user_stack_alloc_size = MAXSSIZ;
2573 
2574 	/* use default location for stack */
2575 	ret = thread_userstackdefault(&addr, result->is_64bit_addr);
2576 	if (ret != KERN_SUCCESS) {
2577 		return LOAD_FAILURE;
2578 	}
2579 
2580 	/* The stack slides down from the default location */
2581 	result->user_stack = (user_addr_t)addr;
2582 	result->user_stack -= slide;
2583 
2584 	if (result->using_lcmain || result->entry_point != MACH_VM_MIN_ADDRESS) {
2585 		/* Already processed LC_MAIN or LC_UNIXTHREAD */
2586 		return LOAD_FAILURE;
2587 	}
2588 
2589 	result->needs_dynlinker = TRUE;
2590 
2591 	ret = thread_state_initialize( thread );
2592 	if (ret != KERN_SUCCESS) {
2593 		return LOAD_FAILURE;
2594 	}
2595 
2596 	result->unixproc = TRUE;
2597 	result->thread_count++;
2598 
2599 	return LOAD_SUCCESS;
2600 }
2601 
2602 static
2603 load_return_t
load_unixthread(struct thread_command * tcp,thread_t thread,int64_t slide,boolean_t is_x86_64_compat_binary,load_result_t * result)2604 load_unixthread(
2605 	struct thread_command   *tcp,
2606 	thread_t                thread,
2607 	int64_t                         slide,
2608 	boolean_t               is_x86_64_compat_binary,
2609 	load_result_t           *result
2610 	)
2611 {
2612 	load_return_t   ret;
2613 	int customstack = 0;
2614 	mach_vm_offset_t addr;
2615 	if (tcp->cmdsize < sizeof(*tcp)) {
2616 		return LOAD_BADMACHO;
2617 	}
2618 	if (result->thread_count != 0) {
2619 		return LOAD_FAILURE;
2620 	}
2621 
2622 	if (thread == THREAD_NULL) {
2623 		return LOAD_SUCCESS;
2624 	}
2625 
2626 	ret = load_threadstack(thread,
2627 	    (uint32_t *)(((vm_offset_t)tcp) +
2628 	    sizeof(struct thread_command)),
2629 	    tcp->cmdsize - sizeof(struct thread_command),
2630 	    &addr, &customstack, is_x86_64_compat_binary, result);
2631 	if (ret != LOAD_SUCCESS) {
2632 		return ret;
2633 	}
2634 
2635 	/* LC_UNIXTHREAD optionally specifies stack size and location */
2636 
2637 	if (customstack) {
2638 		result->custom_stack = TRUE;
2639 	} else {
2640 		result->user_stack_alloc_size = MAXSSIZ;
2641 	}
2642 
2643 	/* The stack slides down from the default location */
2644 	result->user_stack = (user_addr_t)mach_vm_trunc_page((user_addr_t)addr - slide);
2645 
2646 	{
2647 		ret = load_threadentry(thread,
2648 		    (uint32_t *)(((vm_offset_t)tcp) +
2649 		    sizeof(struct thread_command)),
2650 		    tcp->cmdsize - sizeof(struct thread_command),
2651 		    &addr);
2652 		if (ret != LOAD_SUCCESS) {
2653 			return ret;
2654 		}
2655 
2656 		if (result->using_lcmain || result->entry_point != MACH_VM_MIN_ADDRESS) {
2657 			/* Already processed LC_MAIN or LC_UNIXTHREAD */
2658 			return LOAD_FAILURE;
2659 		}
2660 
2661 		result->entry_point = (user_addr_t)addr;
2662 		result->entry_point += slide;
2663 
2664 		ret = load_threadstate(thread,
2665 		    (uint32_t *)(((vm_offset_t)tcp) + sizeof(struct thread_command)),
2666 		    tcp->cmdsize - sizeof(struct thread_command),
2667 		    result);
2668 		if (ret != LOAD_SUCCESS) {
2669 			return ret;
2670 		}
2671 	}
2672 
2673 	result->unixproc = TRUE;
2674 	result->thread_count++;
2675 
2676 	return LOAD_SUCCESS;
2677 }
2678 
2679 static
2680 load_return_t
load_threadstate(thread_t thread,uint32_t * ts,uint32_t total_size,load_result_t * result)2681 load_threadstate(
2682 	thread_t        thread,
2683 	uint32_t        *ts,
2684 	uint32_t        total_size,
2685 	load_result_t   *result
2686 	)
2687 {
2688 	uint32_t        size;
2689 	int             flavor;
2690 	uint32_t        thread_size;
2691 	uint32_t        *local_ts = NULL;
2692 	uint32_t        local_ts_size = 0;
2693 	int             ret;
2694 
2695 	(void)thread;
2696 
2697 	if (total_size > 0) {
2698 		local_ts_size = total_size;
2699 		local_ts = (uint32_t *)kalloc_data(local_ts_size, Z_WAITOK);
2700 		if (local_ts == NULL) {
2701 			return LOAD_FAILURE;
2702 		}
2703 		memcpy(local_ts, ts, local_ts_size);
2704 		ts = local_ts;
2705 	}
2706 
2707 	/*
2708 	 * Validate the new thread state; iterate through the state flavors in
2709 	 * the Mach-O file.
2710 	 * XXX: we should validate the machine state here, to avoid failing at
2711 	 * activation time where we can't bail out cleanly.
2712 	 */
2713 	while (total_size > 0) {
2714 		if (total_size < 2 * sizeof(uint32_t)) {
2715 			return LOAD_BADMACHO;
2716 		}
2717 
2718 		flavor = *ts++;
2719 		size = *ts++;
2720 
2721 		if (os_add_and_mul_overflow(size, 2, sizeof(uint32_t), &thread_size) ||
2722 		    os_sub_overflow(total_size, thread_size, &total_size)) {
2723 			ret = LOAD_BADMACHO;
2724 			goto bad;
2725 		}
2726 
2727 		ts += size;     /* ts is a (uint32_t *) */
2728 	}
2729 
2730 	result->threadstate = local_ts;
2731 	result->threadstate_sz = local_ts_size;
2732 	return LOAD_SUCCESS;
2733 
2734 bad:
2735 	if (local_ts) {
2736 		kfree_data(local_ts, local_ts_size);
2737 	}
2738 	return ret;
2739 }
2740 
2741 
2742 static
2743 load_return_t
load_threadstack(thread_t thread,uint32_t * ts,uint32_t total_size,mach_vm_offset_t * user_stack,int * customstack,__unused boolean_t is_x86_64_compat_binary,load_result_t * result)2744 load_threadstack(
2745 	thread_t                thread,
2746 	uint32_t                *ts,
2747 	uint32_t                total_size,
2748 	mach_vm_offset_t        *user_stack,
2749 	int                     *customstack,
2750 	__unused boolean_t      is_x86_64_compat_binary,
2751 	load_result_t           *result
2752 	)
2753 {
2754 	kern_return_t   ret;
2755 	uint32_t        size;
2756 	int             flavor;
2757 	uint32_t        stack_size;
2758 
2759 	if (total_size == 0) {
2760 		return LOAD_BADMACHO;
2761 	}
2762 
2763 	while (total_size > 0) {
2764 		if (total_size < 2 * sizeof(uint32_t)) {
2765 			return LOAD_BADMACHO;
2766 		}
2767 
2768 		flavor = *ts++;
2769 		size = *ts++;
2770 		if (UINT32_MAX - 2 < size ||
2771 		    UINT32_MAX / sizeof(uint32_t) < size + 2) {
2772 			return LOAD_BADMACHO;
2773 		}
2774 		stack_size = (size + 2) * sizeof(uint32_t);
2775 		if (stack_size > total_size) {
2776 			return LOAD_BADMACHO;
2777 		}
2778 		total_size -= stack_size;
2779 
2780 		/*
2781 		 * Third argument is a kernel space pointer; it gets cast
2782 		 * to the appropriate type in thread_userstack() based on
2783 		 * the value of flavor.
2784 		 */
2785 		{
2786 			ret = thread_userstack(thread, flavor, (thread_state_t)ts, size, user_stack, customstack, result->is_64bit_data);
2787 			if (ret != KERN_SUCCESS) {
2788 				return LOAD_FAILURE;
2789 			}
2790 		}
2791 
2792 		ts += size;     /* ts is a (uint32_t *) */
2793 	}
2794 	return LOAD_SUCCESS;
2795 }
2796 
2797 static
2798 load_return_t
load_threadentry(thread_t thread,uint32_t * ts,uint32_t total_size,mach_vm_offset_t * entry_point)2799 load_threadentry(
2800 	thread_t        thread,
2801 	uint32_t        *ts,
2802 	uint32_t        total_size,
2803 	mach_vm_offset_t        *entry_point
2804 	)
2805 {
2806 	kern_return_t   ret;
2807 	uint32_t        size;
2808 	int             flavor;
2809 	uint32_t        entry_size;
2810 
2811 	/*
2812 	 *	Set the thread state.
2813 	 */
2814 	*entry_point = MACH_VM_MIN_ADDRESS;
2815 	while (total_size > 0) {
2816 		if (total_size < 2 * sizeof(uint32_t)) {
2817 			return LOAD_BADMACHO;
2818 		}
2819 
2820 		flavor = *ts++;
2821 		size = *ts++;
2822 		if (UINT32_MAX - 2 < size ||
2823 		    UINT32_MAX / sizeof(uint32_t) < size + 2) {
2824 			return LOAD_BADMACHO;
2825 		}
2826 		entry_size = (size + 2) * sizeof(uint32_t);
2827 		if (entry_size > total_size) {
2828 			return LOAD_BADMACHO;
2829 		}
2830 		total_size -= entry_size;
2831 		/*
2832 		 * Third argument is a kernel space pointer; it gets cast
2833 		 * to the appropriate type in thread_entrypoint() based on
2834 		 * the value of flavor.
2835 		 */
2836 		ret = thread_entrypoint(thread, flavor, (thread_state_t)ts, size, entry_point);
2837 		if (ret != KERN_SUCCESS) {
2838 			return LOAD_FAILURE;
2839 		}
2840 		ts += size;     /* ts is a (uint32_t *) */
2841 	}
2842 	return LOAD_SUCCESS;
2843 }
2844 
2845 struct macho_data {
2846 	struct nameidata        __nid;
2847 	union macho_vnode_header {
2848 		struct mach_header      mach_header;
2849 		struct fat_header       fat_header;
2850 		char    __pad[512];
2851 	} __header;
2852 };
2853 
2854 #define DEFAULT_DYLD_PATH "/usr/lib/dyld"
2855 
2856 #if (DEVELOPMENT || DEBUG)
2857 extern char dyld_alt_path[];
2858 extern int use_alt_dyld;
2859 
2860 extern char dyld_suffix[];
2861 extern int use_dyld_suffix;
2862 
2863 typedef struct _dyld_suffix_map_entry {
2864 	const char *suffix;
2865 	const char *path;
2866 } dyld_suffix_map_entry_t;
2867 
2868 static const dyld_suffix_map_entry_t _dyld_suffix_map[] = {
2869 	[0] = {
2870 		.suffix = "",
2871 		.path = DEFAULT_DYLD_PATH,
2872 	}, {
2873 		.suffix = "release",
2874 		.path = DEFAULT_DYLD_PATH,
2875 	}, {
2876 		.suffix = "bringup",
2877 		.path = "/usr/appleinternal/lib/dyld.bringup",
2878 	},
2879 };
2880 #endif
2881 
2882 static load_return_t
load_dylinker(struct dylinker_command * lcp,cpu_type_t cputype,vm_map_t map,thread_t thread,int depth,int64_t slide,load_result_t * result,struct image_params * imgp)2883 load_dylinker(
2884 	struct dylinker_command *lcp,
2885 	cpu_type_t              cputype,
2886 	vm_map_t                map,
2887 	thread_t        thread,
2888 	int                     depth,
2889 	int64_t                 slide,
2890 	load_result_t           *result,
2891 	struct image_params     *imgp
2892 	)
2893 {
2894 	const char              *name;
2895 	struct vnode            *vp = NULLVP;   /* set by get_macho_vnode() */
2896 	struct mach_header      *header;
2897 	off_t                   file_offset = 0; /* set by get_macho_vnode() */
2898 	off_t                   macho_size = 0; /* set by get_macho_vnode() */
2899 	load_result_t           *myresult;
2900 	kern_return_t           ret;
2901 	struct macho_data       *macho_data;
2902 	struct {
2903 		struct mach_header      __header;
2904 		load_result_t           __myresult;
2905 		struct macho_data       __macho_data;
2906 	} *dyld_data;
2907 
2908 	if (lcp->cmdsize < sizeof(*lcp) || lcp->name.offset >= lcp->cmdsize) {
2909 		return LOAD_BADMACHO;
2910 	}
2911 
2912 	name = (const char *)lcp + lcp->name.offset;
2913 
2914 	/* Check for a proper null terminated string. */
2915 	size_t maxsz = lcp->cmdsize - lcp->name.offset;
2916 	size_t namelen = strnlen(name, maxsz);
2917 	if (namelen >= maxsz) {
2918 		return LOAD_BADMACHO;
2919 	}
2920 
2921 #if (DEVELOPMENT || DEBUG)
2922 
2923 	/*
2924 	 * rdar://23680808
2925 	 * If an alternate dyld has been specified via boot args, check
2926 	 * to see if PROC_UUID_ALT_DYLD_POLICY has been set on this
2927 	 * executable and redirect the kernel to load that linker.
2928 	 */
2929 
2930 	if (use_alt_dyld) {
2931 		int policy_error;
2932 		uint32_t policy_flags = 0;
2933 		int32_t policy_gencount = 0;
2934 
2935 		policy_error = proc_uuid_policy_lookup(result->uuid, &policy_flags, &policy_gencount);
2936 		if (policy_error == 0) {
2937 			if (policy_flags & PROC_UUID_ALT_DYLD_POLICY) {
2938 				name = dyld_alt_path;
2939 			}
2940 		}
2941 	} else if (use_dyld_suffix) {
2942 		size_t i = 0;
2943 
2944 #define countof(x) (sizeof(x) / sizeof(x[0]))
2945 		for (i = 0; i < countof(_dyld_suffix_map); i++) {
2946 			const dyld_suffix_map_entry_t *entry = &_dyld_suffix_map[i];
2947 
2948 			if (strcmp(entry->suffix, dyld_suffix) == 0) {
2949 				name = entry->path;
2950 				break;
2951 			}
2952 		}
2953 	}
2954 #endif
2955 
2956 #if !(DEVELOPMENT || DEBUG)
2957 	if (0 != strcmp(name, DEFAULT_DYLD_PATH)) {
2958 		return LOAD_BADMACHO;
2959 	}
2960 #endif
2961 
2962 	/* Allocate wad-of-data from heap to reduce excessively deep stacks */
2963 
2964 	dyld_data = kalloc_type(typeof(*dyld_data), Z_WAITOK);
2965 	header = &dyld_data->__header;
2966 	myresult = &dyld_data->__myresult;
2967 	macho_data = &dyld_data->__macho_data;
2968 
2969 	{
2970 		cputype = (cputype & CPU_ARCH_MASK) | (cpu_type() & ~CPU_ARCH_MASK);
2971 	}
2972 
2973 	ret = get_macho_vnode(name, cputype, header,
2974 	    &file_offset, &macho_size, macho_data, &vp, imgp);
2975 	if (ret) {
2976 		goto novp_out;
2977 	}
2978 
2979 	*myresult = load_result_null;
2980 	myresult->is_64bit_addr = result->is_64bit_addr;
2981 	myresult->is_64bit_data = result->is_64bit_data;
2982 
2983 	ret = parse_machfile(vp, map, thread, header, file_offset,
2984 	    macho_size, depth, slide, 0, myresult, result, imgp);
2985 
2986 	if (ret == LOAD_SUCCESS) {
2987 		if (result->threadstate) {
2988 			/* don't use the app's threadstate if we have a dyld */
2989 			kfree_data(result->threadstate, result->threadstate_sz);
2990 		}
2991 		result->threadstate = myresult->threadstate;
2992 		result->threadstate_sz = myresult->threadstate_sz;
2993 
2994 		result->dynlinker = TRUE;
2995 		result->entry_point = myresult->entry_point;
2996 		result->validentry = myresult->validentry;
2997 		result->all_image_info_addr = myresult->all_image_info_addr;
2998 		result->all_image_info_size = myresult->all_image_info_size;
2999 		if (!myresult->platform_binary) {
3000 			result->csflags &= ~CS_NO_UNTRUSTED_HELPERS;
3001 		}
3002 
3003 #if CONFIG_ROSETTA
3004 		if (imgp->ip_flags & IMGPF_ROSETTA) {
3005 			extern const struct fileops vnops;
3006 			// Save the file descriptor and mach header address for dyld. These will
3007 			// be passed on the stack for the Rosetta runtime's use.
3008 			struct fileproc *fp;
3009 			int dyld_fd;
3010 			proc_t p = vfs_context_proc(imgp->ip_vfs_context);
3011 			int error = falloc_exec(p, imgp->ip_vfs_context, &fp, &dyld_fd);
3012 			if (error == 0) {
3013 				error = VNOP_OPEN(vp, FREAD, imgp->ip_vfs_context);
3014 				if (error == 0) {
3015 					fp->fp_glob->fg_flag = FREAD;
3016 					fp->fp_glob->fg_ops = &vnops;
3017 					fp_set_data(fp, vp);
3018 
3019 					proc_fdlock(p);
3020 					procfdtbl_releasefd(p, dyld_fd, NULL);
3021 					fp_drop(p, dyld_fd, fp, 1);
3022 					proc_fdunlock(p);
3023 
3024 					vnode_ref(vp);
3025 
3026 					result->dynlinker_fd = dyld_fd;
3027 					result->dynlinker_fp = fp;
3028 					result->dynlinker_mach_header = myresult->mach_header;
3029 					result->dynlinker_max_vm_addr = myresult->max_vm_addr;
3030 					result->dynlinker_ro_vm_start = myresult->ro_vm_start;
3031 					result->dynlinker_ro_vm_end = myresult->ro_vm_end;
3032 				} else {
3033 					fp_free(p, dyld_fd, fp);
3034 					ret = LOAD_IOERROR;
3035 				}
3036 			} else {
3037 				ret = LOAD_IOERROR;
3038 			}
3039 		}
3040 #endif
3041 	}
3042 
3043 	struct vnode_attr *va;
3044 	va = kalloc_type(struct vnode_attr, Z_WAITOK | Z_ZERO);
3045 	VATTR_INIT(va);
3046 	VATTR_WANTED(va, va_fsid64);
3047 	VATTR_WANTED(va, va_fsid);
3048 	VATTR_WANTED(va, va_fileid);
3049 	int error = vnode_getattr(vp, va, imgp->ip_vfs_context);
3050 	if (error == 0) {
3051 		imgp->ip_dyld_fsid = vnode_get_va_fsid(va);
3052 		imgp->ip_dyld_fsobjid = va->va_fileid;
3053 	}
3054 
3055 	vnode_put(vp);
3056 	kfree_type(struct vnode_attr, va);
3057 novp_out:
3058 	kfree_type(typeof(*dyld_data), dyld_data);
3059 	return ret;
3060 }
3061 
3062 #if CONFIG_ROSETTA
3063 static const char* rosetta_runtime_path = "/usr/libexec/rosetta/runtime";
3064 
3065 #if (DEVELOPMENT || DEBUG)
3066 static const char* rosetta_runtime_path_alt_x86 = "/usr/local/libexec/rosetta/runtime_internal";
3067 static const char* rosetta_runtime_path_alt_arm = "/usr/local/libexec/rosetta/runtime_arm_internal";
3068 #endif
3069 
3070 static load_return_t
load_rosetta(vm_map_t map,thread_t thread,load_result_t * result,struct image_params * imgp)3071 load_rosetta(
3072 	vm_map_t                        map,
3073 	thread_t                        thread,
3074 	load_result_t           *result,
3075 	struct image_params     *imgp)
3076 {
3077 	struct vnode            *vp = NULLVP;   /* set by get_macho_vnode() */
3078 	struct mach_header      *header;
3079 	off_t                   file_offset = 0; /* set by get_macho_vnode() */
3080 	off_t                   macho_size = 0; /* set by get_macho_vnode() */
3081 	load_result_t           *myresult;
3082 	kern_return_t           ret;
3083 	struct macho_data       *macho_data;
3084 	const char              *rosetta_file_path;
3085 	struct {
3086 		struct mach_header      __header;
3087 		load_result_t           __myresult;
3088 		struct macho_data       __macho_data;
3089 	} *rosetta_data;
3090 	mach_vm_address_t rosetta_load_addr;
3091 	mach_vm_size_t    rosetta_size;
3092 	mach_vm_address_t shared_cache_base = SHARED_REGION_BASE_ARM64;
3093 	int64_t           slide = 0;
3094 
3095 	/* Allocate wad-of-data from heap to reduce excessively deep stacks */
3096 	rosetta_data = kalloc_type(typeof(*rosetta_data), Z_WAITOK | Z_NOFAIL);
3097 	header = &rosetta_data->__header;
3098 	myresult = &rosetta_data->__myresult;
3099 	macho_data = &rosetta_data->__macho_data;
3100 
3101 	rosetta_file_path = rosetta_runtime_path;
3102 
3103 #if (DEVELOPMENT || DEBUG)
3104 	bool use_alt_rosetta = false;
3105 	if (imgp->ip_flags & IMGPF_ALT_ROSETTA) {
3106 		use_alt_rosetta = true;
3107 	} else {
3108 		int policy_error;
3109 		uint32_t policy_flags = 0;
3110 		int32_t policy_gencount = 0;
3111 		policy_error = proc_uuid_policy_lookup(result->uuid, &policy_flags, &policy_gencount);
3112 		if (policy_error == 0 && (policy_flags & PROC_UUID_ALT_ROSETTA_POLICY) != 0) {
3113 			use_alt_rosetta = true;
3114 		}
3115 	}
3116 
3117 	if (use_alt_rosetta) {
3118 		if (imgp->ip_origcputype == CPU_TYPE_X86_64) {
3119 			rosetta_file_path = rosetta_runtime_path_alt_x86;
3120 		} else if (imgp->ip_origcputype == CPU_TYPE_ARM64) {
3121 			rosetta_file_path = rosetta_runtime_path_alt_arm;
3122 		} else {
3123 			ret = LOAD_BADARCH;
3124 			goto novp_out;
3125 		}
3126 	}
3127 #endif
3128 
3129 	ret = get_macho_vnode(rosetta_file_path, CPU_TYPE_ARM64, header,
3130 	    &file_offset, &macho_size, macho_data, &vp, imgp);
3131 	if (ret) {
3132 		goto novp_out;
3133 	}
3134 
3135 	*myresult = load_result_null;
3136 	myresult->is_64bit_addr = TRUE;
3137 	myresult->is_64bit_data = TRUE;
3138 
3139 	ret = parse_machfile(vp, NULL, NULL, header, file_offset, macho_size,
3140 	    2, 0, 0, myresult, NULL, imgp);
3141 	if (ret != LOAD_SUCCESS) {
3142 		goto out;
3143 	}
3144 
3145 	if (!(imgp->ip_flags & IMGPF_DISABLE_ASLR)) {
3146 		slide = random();
3147 		slide = (slide % (vm_map_get_max_loader_aslr_slide_pages(map) - 1)) + 1;
3148 		slide <<= vm_map_page_shift(map);
3149 	}
3150 
3151 	if (imgp->ip_origcputype == CPU_TYPE_X86_64) {
3152 		shared_cache_base = SHARED_REGION_BASE_X86_64;
3153 	}
3154 
3155 	rosetta_size = round_page(myresult->max_vm_addr - myresult->min_vm_addr);
3156 	rosetta_load_addr = shared_cache_base - rosetta_size - slide;
3157 
3158 	*myresult = load_result_null;
3159 	myresult->is_64bit_addr = TRUE;
3160 	myresult->is_64bit_data = TRUE;
3161 	myresult->is_rosetta = TRUE;
3162 
3163 	ret = parse_machfile(vp, map, thread, header, file_offset, macho_size,
3164 	    2, rosetta_load_addr, 0, myresult, result, imgp);
3165 	if (ret == LOAD_SUCCESS) {
3166 		if (result) {
3167 			if (result->threadstate) {
3168 				/* don't use the app's/dyld's threadstate */
3169 				kfree_data(result->threadstate, result->threadstate_sz);
3170 			}
3171 			assert(myresult->threadstate != NULL);
3172 
3173 			result->is_rosetta = TRUE;
3174 
3175 			result->threadstate = myresult->threadstate;
3176 			result->threadstate_sz = myresult->threadstate_sz;
3177 
3178 			result->entry_point = myresult->entry_point;
3179 			result->validentry = myresult->validentry;
3180 			if (!myresult->platform_binary) {
3181 				result->csflags &= ~CS_NO_UNTRUSTED_HELPERS;
3182 			}
3183 
3184 			if ((header->cpusubtype & ~CPU_SUBTYPE_MASK) != CPU_SUBTYPE_ARM64E) {
3185 				imgp->ip_flags |= IMGPF_NOJOP;
3186 			}
3187 		}
3188 	}
3189 
3190 out:
3191 	vnode_put(vp);
3192 novp_out:
3193 	kfree_type(typeof(*rosetta_data), rosetta_data);
3194 	return ret;
3195 }
3196 #endif
3197 
3198 static void
set_signature_error(struct vnode * vp,struct image_params * imgp,const char * fatal_failure_desc,const size_t fatal_failure_desc_len)3199 set_signature_error(
3200 	struct vnode* vp,
3201 	struct image_params * imgp,
3202 	const char* fatal_failure_desc,
3203 	const size_t fatal_failure_desc_len)
3204 {
3205 	char *vn_path = NULL;
3206 	vm_size_t vn_pathlen = MAXPATHLEN;
3207 	char const *path = NULL;
3208 
3209 	vn_path = zalloc(ZV_NAMEI);
3210 	if (vn_getpath(vp, vn_path, (int*)&vn_pathlen) == 0) {
3211 		path = vn_path;
3212 	} else {
3213 		path = "(get vnode path failed)";
3214 	}
3215 	os_reason_t reason = os_reason_create(OS_REASON_CODESIGNING,
3216 	    CODESIGNING_EXIT_REASON_TASKGATED_INVALID_SIG);
3217 
3218 	if (reason == OS_REASON_NULL) {
3219 		printf("load_code_signature: %s: failure to allocate exit reason for validation failure: %s\n",
3220 		    path, fatal_failure_desc);
3221 		goto out;
3222 	}
3223 
3224 	imgp->ip_cs_error = reason;
3225 	reason->osr_flags = (OS_REASON_FLAG_GENERATE_CRASH_REPORT |
3226 	    OS_REASON_FLAG_CONSISTENT_FAILURE);
3227 
3228 	mach_vm_address_t data_addr = 0;
3229 
3230 	int reason_error = 0;
3231 	int kcdata_error = 0;
3232 
3233 	if ((reason_error = os_reason_alloc_buffer_noblock(reason, kcdata_estimate_required_buffer_size
3234 	    (1, (uint32_t)fatal_failure_desc_len))) == 0 &&
3235 	    (kcdata_error = kcdata_get_memory_addr(&reason->osr_kcd_descriptor,
3236 	    EXIT_REASON_USER_DESC, (uint32_t)fatal_failure_desc_len,
3237 	    &data_addr)) == KERN_SUCCESS) {
3238 		kern_return_t mc_error = kcdata_memcpy(&reason->osr_kcd_descriptor, (mach_vm_address_t)data_addr,
3239 		    fatal_failure_desc, (uint32_t)fatal_failure_desc_len);
3240 
3241 		if (mc_error != KERN_SUCCESS) {
3242 			printf("load_code_signature: %s: failed to copy reason string "
3243 			    "(kcdata_memcpy error: %d, length: %ld)\n",
3244 			    path, mc_error, fatal_failure_desc_len);
3245 		}
3246 	} else {
3247 		printf("load_code_signature: %s: failed to allocate space for reason string "
3248 		    "(os_reason_alloc_buffer error: %d, kcdata error: %d, length: %ld)\n",
3249 		    path, reason_error, kcdata_error, fatal_failure_desc_len);
3250 	}
3251 out:
3252 	if (vn_path) {
3253 		zfree(ZV_NAMEI, vn_path);
3254 	}
3255 }
3256 
3257 static load_return_t
load_code_signature(struct linkedit_data_command * lcp,struct vnode * vp,off_t macho_offset,off_t macho_size,cpu_type_t cputype,cpu_subtype_t cpusubtype,load_result_t * result,struct image_params * imgp)3258 load_code_signature(
3259 	struct linkedit_data_command    *lcp,
3260 	struct vnode                    *vp,
3261 	off_t                           macho_offset,
3262 	off_t                           macho_size,
3263 	cpu_type_t                      cputype,
3264 	cpu_subtype_t                   cpusubtype,
3265 	load_result_t                   *result,
3266 	struct image_params             *imgp)
3267 {
3268 	int             ret;
3269 	kern_return_t   kr;
3270 	vm_offset_t     addr;
3271 	int             resid;
3272 	struct cs_blob  *blob;
3273 	int             error;
3274 	vm_size_t       blob_size;
3275 	uint32_t        sum;
3276 	boolean_t               anyCPU;
3277 
3278 	addr = 0;
3279 	blob = NULL;
3280 
3281 	cpusubtype &= ~CPU_SUBTYPE_MASK;
3282 
3283 	blob = ubc_cs_blob_get(vp, cputype, cpusubtype, macho_offset);
3284 
3285 	if (blob != NULL) {
3286 		/* we already have a blob for this vnode and cpu(sub)type */
3287 		anyCPU = blob->csb_cpu_type == -1;
3288 		if ((blob->csb_cpu_type != cputype &&
3289 		    blob->csb_cpu_subtype != cpusubtype && !anyCPU) ||
3290 		    (blob->csb_base_offset != macho_offset) ||
3291 		    ((blob->csb_flags & CS_VALID) == 0)) {
3292 			/* the blob has changed for this vnode: fail ! */
3293 			ret = LOAD_BADMACHO;
3294 			const char* fatal_failure_desc = "embedded signature doesn't match attached signature";
3295 			const size_t fatal_failure_desc_len = strlen(fatal_failure_desc) + 1;
3296 
3297 			printf("load_code_signature: %s\n", fatal_failure_desc);
3298 			set_signature_error(vp, imgp, fatal_failure_desc, fatal_failure_desc_len);
3299 			goto out;
3300 		}
3301 
3302 		/* It matches the blob we want here, let's verify the version */
3303 		if (!anyCPU && ubc_cs_generation_check(vp) == 0) {
3304 			/* No need to revalidate, we're good! */
3305 			ret = LOAD_SUCCESS;
3306 			goto out;
3307 		}
3308 
3309 		/* That blob may be stale, let's revalidate. */
3310 		error = ubc_cs_blob_revalidate(vp, blob, imgp, 0, result->ip_platform);
3311 		if (error == 0) {
3312 			/* Revalidation succeeded, we're good! */
3313 			/* If we were revaliding a CS blob with any CPU arch we adjust it */
3314 			if (anyCPU) {
3315 				vnode_lock_spin(vp);
3316 				struct cs_cpu_info cpu_info = {
3317 					.csb_cpu_type = cputype,
3318 					.csb_cpu_subtype = cpusubtype
3319 				};
3320 				zalloc_ro_update_field(ZONE_ID_CS_BLOB, blob, csb_cpu_info, &cpu_info);
3321 				vnode_unlock(vp);
3322 			}
3323 			ret = LOAD_SUCCESS;
3324 			goto out;
3325 		}
3326 
3327 		if (error != EAGAIN) {
3328 			printf("load_code_signature: revalidation failed: %d\n", error);
3329 			ret = LOAD_FAILURE;
3330 			goto out;
3331 		}
3332 
3333 		assert(error == EAGAIN);
3334 
3335 		/*
3336 		 * Revalidation was not possible for this blob. We just continue as if there was no blob,
3337 		 * rereading the signature, and ubc_cs_blob_add will do the right thing.
3338 		 */
3339 		blob = NULL;
3340 	}
3341 
3342 	if (lcp->cmdsize != sizeof(struct linkedit_data_command)) {
3343 		ret = LOAD_BADMACHO;
3344 		goto out;
3345 	}
3346 
3347 	sum = 0;
3348 	if (os_add_overflow(lcp->dataoff, lcp->datasize, &sum) || sum > macho_size) {
3349 		ret = LOAD_BADMACHO;
3350 		goto out;
3351 	}
3352 
3353 	blob_size = lcp->datasize;
3354 	kr = ubc_cs_blob_allocate(&addr, &blob_size);
3355 	if (kr != KERN_SUCCESS) {
3356 		ret = LOAD_NOSPACE;
3357 		goto out;
3358 	}
3359 
3360 	resid = 0;
3361 	error = vn_rdwr(UIO_READ,
3362 	    vp,
3363 	    (caddr_t) addr,
3364 	    lcp->datasize,
3365 	    macho_offset + lcp->dataoff,
3366 	    UIO_SYSSPACE,
3367 	    0,
3368 	    kauth_cred_get(),
3369 	    &resid,
3370 	    current_proc());
3371 	if (error || resid != 0) {
3372 		ret = LOAD_IOERROR;
3373 		goto out;
3374 	}
3375 
3376 	if (ubc_cs_blob_add(vp,
3377 	    result->ip_platform,
3378 	    cputype,
3379 	    cpusubtype,
3380 	    macho_offset,
3381 	    &addr,
3382 	    lcp->datasize,
3383 	    imgp,
3384 	    0,
3385 	    &blob,
3386 	    CS_BLOB_ADD_ALLOW_MAIN_BINARY)) {
3387 		if (addr) {
3388 			ubc_cs_blob_deallocate(addr, blob_size);
3389 			addr = 0;
3390 		}
3391 		ret = LOAD_FAILURE;
3392 		goto out;
3393 	} else {
3394 		/* ubc_cs_blob_add() has consumed "addr" */
3395 		addr = 0;
3396 	}
3397 
3398 #if CHECK_CS_VALIDATION_BITMAP
3399 	ubc_cs_validation_bitmap_allocate( vp );
3400 #endif
3401 
3402 	ret = LOAD_SUCCESS;
3403 out:
3404 	if (ret == LOAD_SUCCESS) {
3405 		if (blob == NULL) {
3406 			panic("success, but no blob!");
3407 		}
3408 
3409 		result->csflags |= blob->csb_flags;
3410 		result->platform_binary = blob->csb_platform_binary;
3411 		result->cs_end_offset = blob->csb_end_offset;
3412 	}
3413 	if (addr != 0) {
3414 		ubc_cs_blob_deallocate(addr, blob_size);
3415 		addr = 0;
3416 	}
3417 
3418 	return ret;
3419 }
3420 
3421 
3422 #if CONFIG_CODE_DECRYPTION
3423 
3424 static load_return_t
set_code_unprotect(struct encryption_info_command * eip,caddr_t addr,vm_map_t map,int64_t slide,struct vnode * vp,off_t macho_offset,cpu_type_t cputype,cpu_subtype_t cpusubtype)3425 set_code_unprotect(
3426 	struct encryption_info_command *eip,
3427 	caddr_t addr,
3428 	vm_map_t map,
3429 	int64_t slide,
3430 	struct vnode *vp,
3431 	off_t macho_offset,
3432 	cpu_type_t cputype,
3433 	cpu_subtype_t cpusubtype)
3434 {
3435 	int error, len;
3436 	pager_crypt_info_t crypt_info;
3437 	const char * cryptname = 0;
3438 	char *vpath;
3439 
3440 	size_t offset;
3441 	struct segment_command_64 *seg64;
3442 	struct segment_command *seg32;
3443 	vm_map_offset_t map_offset, map_size;
3444 	vm_object_offset_t crypto_backing_offset;
3445 	kern_return_t kr;
3446 
3447 	if (eip->cmdsize < sizeof(*eip)) {
3448 		return LOAD_BADMACHO;
3449 	}
3450 
3451 	switch (eip->cryptid) {
3452 	case 0:
3453 		/* not encrypted, just an empty load command */
3454 		return LOAD_SUCCESS;
3455 	case 1:
3456 		cryptname = "com.apple.unfree";
3457 		break;
3458 	case 0x10:
3459 		/* some random cryptid that you could manually put into
3460 		 * your binary if you want NULL */
3461 		cryptname = "com.apple.null";
3462 		break;
3463 	default:
3464 		return LOAD_BADMACHO;
3465 	}
3466 
3467 	if (map == VM_MAP_NULL) {
3468 		return LOAD_SUCCESS;
3469 	}
3470 	if (NULL == text_crypter_create) {
3471 		return LOAD_FAILURE;
3472 	}
3473 
3474 	vpath = zalloc(ZV_NAMEI);
3475 
3476 	len = MAXPATHLEN;
3477 	error = vn_getpath(vp, vpath, &len);
3478 	if (error) {
3479 		zfree(ZV_NAMEI, vpath);
3480 		return LOAD_FAILURE;
3481 	}
3482 
3483 	if (eip->cryptsize == 0) {
3484 		printf("%s:%d '%s': cryptoff 0x%llx cryptsize 0x%llx cryptid 0x%x ignored\n", __FUNCTION__, __LINE__, vpath, (uint64_t)eip->cryptoff, (uint64_t)eip->cryptsize, eip->cryptid);
3485 		zfree(ZV_NAMEI, vpath);
3486 		return LOAD_SUCCESS;
3487 	}
3488 
3489 	/* set up decrypter first */
3490 	crypt_file_data_t crypt_data = {
3491 		.filename = vpath,
3492 		.cputype = cputype,
3493 		.cpusubtype = cpusubtype,
3494 		.origin = CRYPT_ORIGIN_APP_LAUNCH,
3495 	};
3496 	kr = text_crypter_create(&crypt_info, cryptname, (void*)&crypt_data);
3497 #if VM_MAP_DEBUG_APPLE_PROTECT
3498 	if (vm_map_debug_apple_protect) {
3499 		struct proc *p;
3500 		p  = current_proc();
3501 		printf("APPLE_PROTECT: %d[%s] map %p %s(%s) -> 0x%x\n",
3502 		    proc_getpid(p), p->p_comm, map, __FUNCTION__, vpath, kr);
3503 	}
3504 #endif /* VM_MAP_DEBUG_APPLE_PROTECT */
3505 	zfree(ZV_NAMEI, vpath);
3506 
3507 	if (kr) {
3508 		printf("set_code_unprotect: unable to create decrypter %s, kr=%d\n",
3509 		    cryptname, kr);
3510 		if (kr == kIOReturnNotPrivileged) {
3511 			/* text encryption returned decryption failure */
3512 			return LOAD_DECRYPTFAIL;
3513 		} else {
3514 			return LOAD_RESOURCE;
3515 		}
3516 	}
3517 
3518 	/* this is terrible, but we have to rescan the load commands to find the
3519 	 * virtual address of this encrypted stuff. This code is gonna look like
3520 	 * the dyld source one day... */
3521 	struct mach_header *header = (struct mach_header *)addr;
3522 	size_t mach_header_sz = sizeof(struct mach_header);
3523 	if (header->magic == MH_MAGIC_64 ||
3524 	    header->magic == MH_CIGAM_64) {
3525 		mach_header_sz = sizeof(struct mach_header_64);
3526 	}
3527 	offset = mach_header_sz;
3528 	uint32_t ncmds = header->ncmds;
3529 	while (ncmds--) {
3530 		/*
3531 		 *	Get a pointer to the command.
3532 		 */
3533 		struct load_command *lcp = (struct load_command *)(addr + offset);
3534 		offset += lcp->cmdsize;
3535 
3536 		switch (lcp->cmd) {
3537 		case LC_SEGMENT_64:
3538 			seg64 = (struct segment_command_64 *)lcp;
3539 			if ((seg64->fileoff <= eip->cryptoff) &&
3540 			    (seg64->fileoff + seg64->filesize >=
3541 			    eip->cryptoff + eip->cryptsize)) {
3542 				map_offset = (vm_map_offset_t)(seg64->vmaddr + eip->cryptoff - seg64->fileoff + slide);
3543 				map_size = eip->cryptsize;
3544 				crypto_backing_offset = macho_offset + eip->cryptoff;
3545 				goto remap_now;
3546 			}
3547 			break;
3548 		case LC_SEGMENT:
3549 			seg32 = (struct segment_command *)lcp;
3550 			if ((seg32->fileoff <= eip->cryptoff) &&
3551 			    (seg32->fileoff + seg32->filesize >=
3552 			    eip->cryptoff + eip->cryptsize)) {
3553 				map_offset = (vm_map_offset_t)(seg32->vmaddr + eip->cryptoff - seg32->fileoff + slide);
3554 				map_size = eip->cryptsize;
3555 				crypto_backing_offset = macho_offset + eip->cryptoff;
3556 				goto remap_now;
3557 			}
3558 			break;
3559 		}
3560 	}
3561 
3562 	/* if we get here, did not find anything */
3563 	return LOAD_BADMACHO;
3564 
3565 remap_now:
3566 	/* now remap using the decrypter */
3567 	MACHO_PRINTF(("+++ set_code_unprotect: vm[0x%llx:0x%llx]\n",
3568 	    (uint64_t) map_offset,
3569 	    (uint64_t) (map_offset + map_size)));
3570 	kr = vm_map_apple_protected(map,
3571 	    map_offset,
3572 	    map_offset + map_size,
3573 	    crypto_backing_offset,
3574 	    &crypt_info,
3575 	    CRYPTID_APP_ENCRYPTION);
3576 	if (kr) {
3577 		printf("set_code_unprotect(): mapping failed with %x\n", kr);
3578 		return LOAD_PROTECT;
3579 	}
3580 
3581 	return LOAD_SUCCESS;
3582 }
3583 
3584 #endif
3585 
3586 /*
3587  * This routine exists to support the load_dylinker().
3588  *
3589  * This routine has its own, separate, understanding of the FAT file format,
3590  * which is terrifically unfortunate.
3591  */
3592 static
3593 load_return_t
get_macho_vnode(const char * path,cpu_type_t cputype,struct mach_header * mach_header,off_t * file_offset,off_t * macho_size,struct macho_data * data,struct vnode ** vpp,struct image_params * imgp)3594 get_macho_vnode(
3595 	const char              *path,
3596 	cpu_type_t              cputype,
3597 	struct mach_header      *mach_header,
3598 	off_t                   *file_offset,
3599 	off_t                   *macho_size,
3600 	struct macho_data       *data,
3601 	struct vnode            **vpp,
3602 	struct image_params     *imgp
3603 	)
3604 {
3605 	struct vnode            *vp;
3606 	vfs_context_t           ctx = vfs_context_current();
3607 	proc_t                  p = vfs_context_proc(ctx);
3608 	kauth_cred_t            kerncred;
3609 	struct nameidata        *ndp = &data->__nid;
3610 	boolean_t               is_fat;
3611 	struct fat_arch         fat_arch;
3612 	int                     error;
3613 	int resid;
3614 	union macho_vnode_header *header = &data->__header;
3615 	off_t fsize = (off_t)0;
3616 
3617 	/*
3618 	 * Capture the kernel credential for use in the actual read of the
3619 	 * file, since the user doing the execution may have execute rights
3620 	 * but not read rights, but to exec something, we have to either map
3621 	 * or read it into the new process address space, which requires
3622 	 * read rights.  This is to deal with lack of common credential
3623 	 * serialization code which would treat NOCRED as "serialize 'root'".
3624 	 */
3625 	kerncred = vfs_context_ucred(vfs_context_kernel());
3626 
3627 	/* init the namei data to point the file user's program name */
3628 	NDINIT(ndp, LOOKUP, OP_OPEN, FOLLOW | LOCKLEAF, UIO_SYSSPACE, CAST_USER_ADDR_T(path), ctx);
3629 
3630 	if ((error = namei(ndp)) != 0) {
3631 		if (error == ENOENT) {
3632 			error = LOAD_ENOENT;
3633 		} else {
3634 			error = LOAD_FAILURE;
3635 		}
3636 		return error;
3637 	}
3638 	nameidone(ndp);
3639 	vp = ndp->ni_vp;
3640 
3641 	/* check for regular file */
3642 	if (vp->v_type != VREG) {
3643 		error = LOAD_PROTECT;
3644 		goto bad1;
3645 	}
3646 
3647 	/* get size */
3648 	if ((error = vnode_size(vp, &fsize, ctx)) != 0) {
3649 		error = LOAD_FAILURE;
3650 		goto bad1;
3651 	}
3652 
3653 	/* Check mount point */
3654 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
3655 		error = LOAD_PROTECT;
3656 		goto bad1;
3657 	}
3658 
3659 	/* check access */
3660 	if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_EXECUTE | KAUTH_VNODE_READ_DATA, ctx)) != 0) {
3661 		error = LOAD_PROTECT;
3662 		goto bad1;
3663 	}
3664 
3665 	/* try to open it */
3666 	if ((error = VNOP_OPEN(vp, FREAD, ctx)) != 0) {
3667 		error = LOAD_PROTECT;
3668 		goto bad1;
3669 	}
3670 
3671 	if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)header, sizeof(*header), 0,
3672 	    UIO_SYSSPACE, IO_NODELOCKED, kerncred, &resid, p)) != 0) {
3673 		error = LOAD_IOERROR;
3674 		goto bad2;
3675 	}
3676 
3677 	if (resid) {
3678 		error = LOAD_BADMACHO;
3679 		goto bad2;
3680 	}
3681 
3682 	if (header->mach_header.magic == MH_MAGIC ||
3683 	    header->mach_header.magic == MH_MAGIC_64) {
3684 		is_fat = FALSE;
3685 	} else if (OSSwapBigToHostInt32(header->fat_header.magic) == FAT_MAGIC) {
3686 		is_fat = TRUE;
3687 	} else {
3688 		error = LOAD_BADMACHO;
3689 		goto bad2;
3690 	}
3691 
3692 	if (is_fat) {
3693 		error = fatfile_validate_fatarches((vm_offset_t)(&header->fat_header),
3694 		    sizeof(*header), fsize);
3695 		if (error != LOAD_SUCCESS) {
3696 			goto bad2;
3697 		}
3698 
3699 		/* Look up our architecture in the fat file. */
3700 		error = fatfile_getbestarch_for_cputype(cputype, CPU_SUBTYPE_ANY,
3701 		    (vm_offset_t)(&header->fat_header), sizeof(*header), imgp, &fat_arch);
3702 		if (error != LOAD_SUCCESS) {
3703 			goto bad2;
3704 		}
3705 
3706 		/* Read the Mach-O header out of it */
3707 		error = vn_rdwr(UIO_READ, vp, (caddr_t)&header->mach_header,
3708 		    sizeof(header->mach_header), fat_arch.offset,
3709 		    UIO_SYSSPACE, IO_NODELOCKED, kerncred, &resid, p);
3710 		if (error) {
3711 			error = LOAD_IOERROR;
3712 			goto bad2;
3713 		}
3714 
3715 		if (resid) {
3716 			error = LOAD_BADMACHO;
3717 			goto bad2;
3718 		}
3719 
3720 		/* Is this really a Mach-O? */
3721 		if (header->mach_header.magic != MH_MAGIC &&
3722 		    header->mach_header.magic != MH_MAGIC_64) {
3723 			error = LOAD_BADMACHO;
3724 			goto bad2;
3725 		}
3726 
3727 		*file_offset = fat_arch.offset;
3728 		*macho_size = fat_arch.size;
3729 	} else {
3730 		/*
3731 		 * Force get_macho_vnode() to fail if the architecture bits
3732 		 * do not match the expected architecture bits.  This in
3733 		 * turn causes load_dylinker() to fail for the same reason,
3734 		 * so it ensures the dynamic linker and the binary are in
3735 		 * lock-step.  This is potentially bad, if we ever add to
3736 		 * the CPU_ARCH_* bits any bits that are desirable but not
3737 		 * required, since the dynamic linker might work, but we will
3738 		 * refuse to load it because of this check.
3739 		 */
3740 		if ((cpu_type_t)header->mach_header.cputype != cputype) {
3741 			error = LOAD_BADARCH;
3742 			goto bad2;
3743 		}
3744 
3745 		*file_offset = 0;
3746 		*macho_size = fsize;
3747 	}
3748 
3749 	*mach_header = header->mach_header;
3750 	*vpp = vp;
3751 
3752 	ubc_setsize(vp, fsize);
3753 	return error;
3754 
3755 bad2:
3756 	(void) VNOP_CLOSE(vp, FREAD, ctx);
3757 bad1:
3758 	vnode_put(vp);
3759 	return error;
3760 }
3761