xref: /xnu-10002.61.3/bsd/kern/kern_core.c (revision 0f4c859e951fba394238ab619495c4e1d54d0f34)
1 /*
2  * Copyright (c) 2000-2021 Apple Computer, Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 /* Copyright (c) 1991 NeXT Computer, Inc.  All rights reserved.
29  *
30  *	File:	bsd/kern/kern_core.c
31  *
32  *	This file contains machine independent code for performing core dumps.
33  *
34  */
35 #if CONFIG_COREDUMP
36 
37 #include <mach/vm_param.h>
38 #include <mach/thread_status.h>
39 #include <sys/content_protection.h>
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/signalvar.h>
43 #include <sys/resourcevar.h>
44 #include <sys/namei.h>
45 #include <sys/vnode_internal.h>
46 #include <sys/proc_internal.h>
47 #include <sys/kauth.h>
48 #include <sys/timeb.h>
49 #include <sys/times.h>
50 #include <sys/acct.h>
51 #include <sys/file_internal.h>
52 #include <sys/uio.h>
53 #include <sys/kernel.h>
54 #include <sys/stat.h>
55 
56 #include <mach-o/loader.h>
57 #include <mach/vm_region.h>
58 #include <mach/vm_statistics.h>
59 
60 #include <IOKit/IOBSD.h>
61 
62 #include <vm/vm_kern.h>
63 #include <vm/vm_protos.h> /* last */
64 #include <vm/vm_map.h>          /* current_map() */
65 #include <vm/pmap.h>            /* pmap_user_va_bits() */
66 #include <mach/mach_vm.h>       /* mach_vm_region_recurse() */
67 #include <mach/task.h>          /* task_suspend() */
68 #include <kern/task.h>          /* get_task_numacts() */
69 
70 #include <security/audit/audit.h>
71 
72 #if CONFIG_MACF
73 #include <security/mac_framework.h>
74 #endif /* CONFIG_MACF */
75 
76 #include <kdp/core_notes.h>
77 
78 #define COREDUMP_CUSTOM_LOCATION_ENTITLEMENT "com.apple.private.custom-coredump-location"
79 
80 typedef struct {
81 	int     flavor;                 /* the number for this flavor */
82 	mach_msg_type_number_t  count;  /* count of ints in this flavor */
83 } mythread_state_flavor_t;
84 
85 #if defined (__i386__) || defined (__x86_64__)
86 mythread_state_flavor_t thread_flavor_array[] = {
87 	{x86_THREAD_STATE, x86_THREAD_STATE_COUNT},
88 	{x86_FLOAT_STATE, x86_FLOAT_STATE_COUNT},
89 	{x86_EXCEPTION_STATE, x86_EXCEPTION_STATE_COUNT},
90 };
91 int mynum_flavors = 3;
92 #elif defined (__arm64__)
93 mythread_state_flavor_t thread_flavor_array[] = {
94 	{ARM_THREAD_STATE64, ARM_THREAD_STATE64_COUNT},
95 	/* ARM64_TODO: VFP */
96 	{ARM_EXCEPTION_STATE64, ARM_EXCEPTION_STATE64_COUNT}
97 };
98 int mynum_flavors = 2;
99 #else
100 #error architecture not supported
101 #endif
102 
103 
104 typedef struct {
105 	vm_offset_t header;
106 	size_t hoffset;
107 	mythread_state_flavor_t *flavors;
108 	size_t tstate_size;
109 	size_t flavor_count;
110 } tir_t;
111 
112 extern int freespace_mb(vnode_t vp);
113 extern void task_lock(task_t);
114 extern void task_unlock(task_t);
115 
116 /* XXX not in a Mach header anywhere */
117 kern_return_t thread_getstatus(thread_t act, int flavor,
118     thread_state_t tstate, mach_msg_type_number_t *count);
119 void task_act_iterate_wth_args_locked(task_t, void (*)(thread_t, void *), void *);
120 
121 #ifdef SECURE_KERNEL
122 __XNU_PRIVATE_EXTERN int do_coredump = 0;       /* default: don't dump cores */
123 #else
124 __XNU_PRIVATE_EXTERN int do_coredump = 1;       /* default: dump cores */
125 #endif /* SECURE_KERNEL */
126 __XNU_PRIVATE_EXTERN int sugid_coredump = 0; /* default: but not SGUID binaries */
127 
128 
129 /* cpu_type returns only the most generic indication of the current CPU. */
130 /* in a core we want to know the kind of process. */
131 
132 cpu_type_t
process_cpu_type(proc_t core_proc)133 process_cpu_type(proc_t core_proc)
134 {
135 	cpu_type_t what_we_think;
136 #if defined (__i386__) || defined (__x86_64__)
137 	if (IS_64BIT_PROCESS(core_proc)) {
138 		what_we_think = CPU_TYPE_X86_64;
139 	} else {
140 		what_we_think = CPU_TYPE_I386;
141 	}
142 #elif defined(__arm64__)
143 	if (IS_64BIT_PROCESS(core_proc)) {
144 		what_we_think = CPU_TYPE_ARM64;
145 	} else {
146 		what_we_think = CPU_TYPE_ARM;
147 	}
148 #endif
149 
150 	return what_we_think;
151 }
152 
153 cpu_type_t
process_cpu_subtype(proc_t core_proc)154 process_cpu_subtype(proc_t core_proc)
155 {
156 	cpu_type_t what_we_think;
157 #if defined (__i386__) || defined (__x86_64__)
158 	if (IS_64BIT_PROCESS(core_proc)) {
159 		what_we_think = CPU_SUBTYPE_X86_64_ALL;
160 	} else {
161 		what_we_think = CPU_SUBTYPE_I386_ALL;
162 	}
163 #elif defined(__arm64__)
164 	if (IS_64BIT_PROCESS(core_proc)) {
165 		what_we_think = CPU_SUBTYPE_ARM64_ALL;
166 	} else {
167 		what_we_think = CPU_SUBTYPE_ARM_ALL;
168 	}
169 #endif
170 	return what_we_think;
171 }
172 
173 static void
collectth_state(thread_t th_act,void * tirp)174 collectth_state(thread_t th_act, void *tirp)
175 {
176 	vm_offset_t     header;
177 	size_t  hoffset, i;
178 	mythread_state_flavor_t *flavors;
179 	struct thread_command   *tc;
180 	tir_t *t = (tir_t *)tirp;
181 
182 	/*
183 	 *	Fill in thread command structure.
184 	 */
185 	header = t->header;
186 	hoffset = t->hoffset;
187 	flavors = t->flavors;
188 
189 	tc = (struct thread_command *) (header + hoffset);
190 	tc->cmd = LC_THREAD;
191 	tc->cmdsize = (uint32_t)(sizeof(struct thread_command)
192 	    + t->tstate_size);
193 	hoffset += sizeof(struct thread_command);
194 	/*
195 	 * Follow with a struct thread_state_flavor and
196 	 * the appropriate thread state struct for each
197 	 * thread state flavor.
198 	 */
199 	for (i = 0; i < t->flavor_count; i++) {
200 		*(mythread_state_flavor_t *)(header + hoffset) =
201 		    flavors[i];
202 		hoffset += sizeof(mythread_state_flavor_t);
203 		thread_getstatus(th_act, flavors[i].flavor,
204 		    (thread_state_t)(header + hoffset),
205 		    &flavors[i].count);
206 		hoffset += flavors[i].count * sizeof(int);
207 	}
208 
209 	t->hoffset = hoffset;
210 }
211 
212 #if DEVELOPMENT || DEBUG
213 #define COREDUMPLOG(fmt, args...) printf("coredump (%s, pid %d): " fmt "\n", core_proc->p_comm, proc_getpid(core_proc), ## args)
214 #else
215 #define COREDUMPLOG(fmt, args...)
216 #endif
217 
218 /*
219  * LC_NOTE support for userspace coredumps.
220  */
221 
222 typedef int (write_note_cb_t)(struct vnode *vp, off_t foffset);
223 
224 static int
note_addrable_bits(struct vnode * vp,off_t foffset)225 note_addrable_bits(struct vnode *vp, off_t foffset)
226 {
227 	task_t t = current_task();
228 	vfs_context_t ctx = vfs_context_current();
229 	kauth_cred_t cred = vfs_context_ucred(ctx);
230 
231 	addrable_bits_note_t note = {
232 		.version = ADDRABLE_BITS_VER,
233 		.addressing_bits = pmap_user_va_bits(get_task_pmap(t)),
234 		.unused = 0
235 	};
236 
237 	return vn_rdwr_64(UIO_WRITE, vp, (vm_offset_t)&note, sizeof(note), foffset, UIO_SYSSPACE,
238 	           IO_NODELOCKED | IO_UNIT, cred, 0, current_proc());
239 }
240 
241 /*
242  * note handling
243  */
244 
245 struct core_note {
246 	size_t          cn_size;
247 	const char      *cn_owner;
248 	write_note_cb_t *cn_write_cb;
249 } const core_notes[] = {
250 	{
251 		.cn_size = sizeof(addrable_bits_note_t),
252 		.cn_owner = ADDRABLE_BITS_DATA_OWNER,
253 		.cn_write_cb = note_addrable_bits,
254 	}
255 };
256 
257 const size_t notes_count = sizeof(core_notes) / sizeof(struct core_note);
258 
259 /*
260  * LC_NOTE commands are allocated as a part of Mach-O header and are written to
261  * disk at the end of coredump. LC_NOTE's payload has to be written in callbacks here.
262  */
263 static int
dump_notes(proc_t __unused core_proc,vm_offset_t header,size_t hoffset,struct vnode * vp,off_t foffset)264 dump_notes(proc_t __unused core_proc, vm_offset_t header, size_t hoffset, struct vnode *vp, off_t foffset)
265 {
266 	for (size_t i = 0; i < notes_count; i++) {
267 		int error = 0;
268 
269 		if (core_notes[i].cn_write_cb == NULL) {
270 			continue;
271 		}
272 
273 		/* Generate LC_NOTE command. */
274 		struct note_command *nc = (struct note_command *)(header + hoffset);
275 
276 		nc->cmd = LC_NOTE;
277 		nc->cmdsize = sizeof(struct note_command);
278 		nc->offset = foffset;
279 		nc->size = core_notes[i].cn_size;
280 		strlcpy(nc->data_owner, core_notes[i].cn_owner, sizeof(nc->data_owner));
281 
282 		hoffset += sizeof(struct note_command);
283 
284 		/* Add note's payload. */
285 		error = core_notes[i].cn_write_cb(vp, foffset);
286 		if (error != KERN_SUCCESS) {
287 			COREDUMPLOG("failed to write LC_NOTE %s: error %d", core_notes[i].cn_owner, error);
288 			return error;
289 		}
290 
291 		foffset += core_notes[i].cn_size;
292 	}
293 
294 	return 0;
295 }
296 
297 /*
298  * coredump
299  *
300  * Description:	Create a core image on the file "core" for the process
301  *		indicated
302  *
303  * Parameters:	core_proc			Process to dump core [*]
304  *				reserve_mb			If non-zero, leave filesystem with
305  *									at least this much free space.
306  *				coredump_flags	Extra options (ignore rlimit, run fsync)
307  *
308  * Returns:	0				Success
309  *		!0				Failure errno
310  *
311  * IMPORTANT:	This function can only be called on the current process, due
312  *		to assumptions below; see variable declaration section for
313  *		details.
314  */
315 #define MAX_TSTATE_FLAVORS      10
316 int
coredump(proc_t core_proc,uint32_t reserve_mb,int coredump_flags)317 coredump(proc_t core_proc, uint32_t reserve_mb, int coredump_flags)
318 {
319 /* Begin assumptions that limit us to only the current process */
320 	vfs_context_t ctx = vfs_context_current();
321 	vm_map_t        map = current_map();
322 	task_t          task = current_task();
323 /* End assumptions */
324 	kauth_cred_t cred = vfs_context_ucred(ctx);
325 	int error = 0;
326 	struct vnode_attr *vap = NULL;
327 	size_t          thread_count, segment_count;
328 	size_t          command_size, header_size, tstate_size;
329 	size_t          hoffset;
330 	off_t           foffset;
331 	mach_vm_offset_t vmoffset;
332 	vm_offset_t     header;
333 	mach_vm_size_t  vmsize;
334 	vm_prot_t       prot;
335 	vm_prot_t       maxprot;
336 	int             error1 = 0;
337 	char            stack_name[MAXCOMLEN + 6];
338 	char            *alloced_name = NULL;
339 	char            *name = NULL;
340 	mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
341 	vm_size_t       mapsize;
342 	size_t          i;
343 	uint32_t nesting_depth = 0;
344 	kern_return_t   kret;
345 	struct vm_region_submap_info_64 vbr;
346 	mach_msg_type_number_t vbrcount = 0;
347 	tir_t tir1;
348 	struct vnode * vp;
349 	struct mach_header      *mh = NULL;     /* protected by is_64 */
350 	struct mach_header_64   *mh64 = NULL;   /* protected by is_64 */
351 	int             is_64 = 0;
352 	size_t          mach_header_sz = sizeof(struct mach_header);
353 	size_t          segment_command_sz = sizeof(struct segment_command);
354 	size_t          notes_size = 0;
355 	const char     *format = NULL;
356 	char           *custom_location_entitlement = NULL;
357 	size_t          custom_location_entitlement_len = 0;
358 	char           *alloced_format = NULL;
359 	size_t          alloced_format_len = 0;
360 	bool            include_iokit_memory = task_is_driver(task);
361 	bool            coredump_attempted = false;
362 	bool            task_locked = false;
363 
364 	if (current_proc() != core_proc) {
365 		panic("coredump() called against proc that is not current_proc: %p", core_proc);
366 	}
367 
368 	if (do_coredump == 0 ||         /* Not dumping at all */
369 	    ((sugid_coredump == 0) &&   /* Not dumping SUID/SGID binaries */
370 	    ((kauth_cred_getsvuid(cred) != kauth_cred_getruid(cred)) ||
371 	    (kauth_cred_getsvgid(cred) != kauth_cred_getrgid(cred))))) {
372 		error = EFAULT;
373 		goto out2;
374 	}
375 
376 #if CONFIG_MACF
377 	error = mac_proc_check_dump_core(core_proc);
378 	if (error != 0) {
379 		goto out2;
380 	}
381 #endif
382 
383 	if (IS_64BIT_PROCESS(core_proc)) {
384 		is_64 = 1;
385 		mach_header_sz = sizeof(struct mach_header_64);
386 		segment_command_sz = sizeof(struct segment_command_64);
387 	}
388 
389 	mapsize = get_vmmap_size(map);
390 
391 	custom_location_entitlement = IOCurrentTaskGetEntitlement(COREDUMP_CUSTOM_LOCATION_ENTITLEMENT);
392 	if (custom_location_entitlement != NULL) {
393 		custom_location_entitlement_len = strlen(custom_location_entitlement);
394 		const char * dirname;
395 		if (proc_is_driver(core_proc)) {
396 			dirname = defaultdrivercorefiledir;
397 		} else {
398 			dirname = defaultcorefiledir;
399 		}
400 		size_t dirname_len = strlen(dirname);
401 		size_t printed_len;
402 
403 		/* new format is dirname + "/" + string from entitlement */
404 		alloced_format_len = dirname_len + 1 + custom_location_entitlement_len;
405 		alloced_format = kalloc_data(alloced_format_len + 1, Z_ZERO | Z_WAITOK | Z_NOFAIL);
406 		printed_len = snprintf(alloced_format, alloced_format_len + 1, "%s/%s", dirname, custom_location_entitlement);
407 		assert(printed_len == alloced_format_len);
408 
409 		format = alloced_format;
410 		coredump_flags |= COREDUMP_IGNORE_ULIMIT;
411 	} else {
412 		if (proc_is_driver(core_proc)) {
413 			format = drivercorefilename;
414 		} else {
415 			format = corefilename;
416 		}
417 	}
418 
419 	if (((coredump_flags & COREDUMP_IGNORE_ULIMIT) == 0) &&
420 	    (mapsize >= proc_limitgetcur(core_proc, RLIMIT_CORE))) {
421 		error = EFAULT;
422 		goto out2;
423 	}
424 
425 	/* log coredump failures from here */
426 	coredump_attempted = true;
427 
428 	task_lock(task);
429 	task_locked = true;
430 	(void) task_suspend_internal_locked(task);
431 
432 	alloced_name = zalloc_flags(ZV_NAMEI, Z_NOWAIT | Z_ZERO);
433 
434 	/* create name according to sysctl'able format string */
435 	/* if name creation fails, fall back to historical behaviour... */
436 	if (alloced_name == NULL ||
437 	    proc_core_name(format, core_proc->p_comm, kauth_cred_getuid(cred),
438 	    proc_getpid(core_proc), alloced_name, MAXPATHLEN)) {
439 		snprintf(stack_name, sizeof(stack_name),
440 		    "/cores/core.%d", proc_getpid(core_proc));
441 		name = stack_name;
442 	} else {
443 		name = alloced_name;
444 	}
445 
446 	COREDUMPLOG("writing core to %s", name);
447 	if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, ctx))) {
448 		COREDUMPLOG("failed to open core dump file %s: error %d", name, error);
449 		goto out2;
450 	}
451 
452 	vap = kalloc_type(struct vnode_attr, Z_WAITOK | Z_ZERO);
453 	VATTR_INIT(vap);
454 	VATTR_WANTED(vap, va_nlink);
455 	/* Don't dump to non-regular files or files with links. */
456 	if (vp->v_type != VREG ||
457 	    vnode_getattr(vp, vap, ctx) || vap->va_nlink != 1) {
458 		COREDUMPLOG("failed to write core to non-regular file");
459 		error = EFAULT;
460 		goto out;
461 	}
462 
463 	VATTR_INIT(vap);        /* better to do it here than waste more stack in vnode_setsize */
464 	VATTR_SET(vap, va_data_size, 0);
465 	if (core_proc == initproc) {
466 		VATTR_SET(vap, va_dataprotect_class, PROTECTION_CLASS_D);
467 	}
468 	vnode_setattr(vp, vap, ctx);
469 	core_proc->p_acflag |= ACORE;
470 
471 	COREDUMPLOG("map size: %lu", mapsize);
472 	if ((reserve_mb > 0) &&
473 	    ((freespace_mb(vp) - (mapsize >> 20)) < reserve_mb)) {
474 		COREDUMPLOG("insufficient free space (free=%d MB, needed=%lu MB, reserve=%d MB)", freespace_mb(vp), (mapsize >> 20), reserve_mb);
475 		error = ENOSPC;
476 		goto out;
477 	}
478 
479 	thread_count = get_task_numacts(task);
480 	segment_count = get_vmmap_entries(map); /* XXX */
481 	tir1.flavor_count = sizeof(thread_flavor_array) / sizeof(mythread_state_flavor_t);
482 	bcopy(thread_flavor_array, flavors, sizeof(thread_flavor_array));
483 	tstate_size = 0;
484 	for (i = 0; i < tir1.flavor_count; i++) {
485 		tstate_size += sizeof(mythread_state_flavor_t) +
486 		    (flavors[i].count * sizeof(int));
487 	}
488 
489 	{
490 		size_t lhs;
491 		size_t rhs;
492 
493 		/* lhs = segment_count * segment_command_sz */
494 		if (os_mul_overflow(segment_count, segment_command_sz, &lhs)) {
495 			COREDUMPLOG("error: segment size overflow: segment_count=%lu, segment_command_sz=%lu", segment_count, segment_command_sz);
496 			error = ENOMEM;
497 			goto out;
498 		}
499 
500 		/* rhs = (tstate_size + sizeof(struct thread_command)) * thread_count */
501 		if (os_add_and_mul_overflow(tstate_size, sizeof(struct thread_command), thread_count, &rhs)) {
502 			COREDUMPLOG("error: thread state size overflow: tstate_size=%lu, thread_count=%lu", tstate_size, thread_count);
503 			error = ENOMEM;
504 			goto out;
505 		}
506 
507 		/* command_size = lhs + rhs */
508 		if (os_add_overflow(lhs, rhs, &command_size)) {
509 			COREDUMPLOG("error: command size overflow: lhs=%lu, rhs=%lu", lhs, rhs);
510 			error = ENOMEM;
511 			goto out;
512 		}
513 
514 		/* Add notes payload. */
515 		if (os_mul_overflow(notes_count, sizeof(struct note_command), &notes_size)) {
516 			COREDUMPLOG("error: note command size overflow: note=%lu", i);
517 			error = ENOMEM;
518 			goto out;
519 		}
520 
521 		if (os_add_overflow(command_size, notes_size, &command_size)) {
522 			COREDUMPLOG("error: notes overflow: notes_size=%lu", notes_size);
523 			error = ENOMEM;
524 			goto out;
525 		}
526 	}
527 
528 	if (os_add_overflow(command_size, mach_header_sz, &header_size)) {
529 		COREDUMPLOG("error: header size overflow: command_size=%lu, mach_header_sz=%lu", command_size, mach_header_sz);
530 		error = ENOMEM;
531 		goto out;
532 	}
533 
534 	if (kmem_alloc(kernel_map, &header, (vm_size_t)header_size,
535 	    KMA_DATA | KMA_ZERO, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
536 		COREDUMPLOG("error: failed to allocate memory for header (size=%lu)", header_size);
537 		error = ENOMEM;
538 		goto out;
539 	}
540 
541 	/*
542 	 *	Set up Mach-O header.
543 	 */
544 	if (is_64) {
545 		mh64 = (struct mach_header_64 *)header;
546 		mh64->magic = MH_MAGIC_64;
547 		mh64->cputype = process_cpu_type(core_proc);
548 		mh64->cpusubtype = process_cpu_subtype(core_proc);
549 		mh64->filetype = MH_CORE;
550 		mh64->ncmds = (uint32_t)(segment_count + notes_count + thread_count);
551 		mh64->sizeofcmds = (uint32_t)command_size;
552 	} else {
553 		mh = (struct mach_header *)header;
554 		mh->magic = MH_MAGIC;
555 		mh->cputype = process_cpu_type(core_proc);
556 		mh->cpusubtype = process_cpu_subtype(core_proc);
557 		mh->filetype = MH_CORE;
558 		mh->ncmds = (uint32_t)(segment_count + notes_count + thread_count);
559 		mh->sizeofcmds = (uint32_t)command_size;
560 	}
561 
562 	hoffset = mach_header_sz;       /* offset into header */
563 	foffset = round_page(header_size);      /* offset into file */
564 	vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
565 	COREDUMPLOG("mach header size: %zu", header_size);
566 
567 	/*
568 	 * We use to check for an error, here, now we try and get
569 	 * as much as we can
570 	 */
571 	COREDUMPLOG("dumping %zu segments", segment_count);
572 	while (segment_count > 0) {
573 		struct segment_command          *sc;
574 		struct segment_command_64       *sc64;
575 
576 		/*
577 		 *	Get region information for next region.
578 		 */
579 
580 		while (1) {
581 			vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
582 			if ((kret = mach_vm_region_recurse(map,
583 			    &vmoffset, &vmsize, &nesting_depth,
584 			    (vm_region_recurse_info_t)&vbr,
585 			    &vbrcount)) != KERN_SUCCESS) {
586 				break;
587 			}
588 			/*
589 			 * If we get a valid mapping back, but we're dumping
590 			 * a 32 bit process,  and it's over the allowable
591 			 * address space of a 32 bit process, it's the same
592 			 * as if mach_vm_region_recurse() failed.
593 			 */
594 			if (!(is_64) &&
595 			    (vmoffset + vmsize > VM_MAX_ADDRESS)) {
596 				kret = KERN_INVALID_ADDRESS;
597 				COREDUMPLOG("exceeded allowable region for 32-bit process");
598 				break;
599 			}
600 			if (vbr.is_submap) {
601 				nesting_depth++;
602 				continue;
603 			} else {
604 				break;
605 			}
606 		}
607 		if (kret != KERN_SUCCESS) {
608 			COREDUMPLOG("ending segment dump, kret=%d", kret);
609 			break;
610 		}
611 
612 		prot = vbr.protection;
613 		maxprot = vbr.max_protection;
614 
615 		if ((prot | maxprot) == VM_PROT_NONE) {
616 			/*
617 			 * Elide unreadable (likely reserved) segments
618 			 */
619 			COREDUMPLOG("eliding unreadable segment %llx->%llx", vmoffset, vmoffset + vmsize);
620 			vmoffset += vmsize;
621 			continue;
622 		}
623 
624 		/*
625 		 * Try as hard as possible to get read access to the data.
626 		 */
627 		if ((prot & VM_PROT_READ) == 0) {
628 			mach_vm_protect(map, vmoffset, vmsize, FALSE,
629 			    prot | VM_PROT_READ);
630 		}
631 
632 		/*
633 		 * But only try and perform the write if we can read it.
634 		 */
635 		int64_t fsize = ((maxprot & VM_PROT_READ) == VM_PROT_READ
636 		    && (include_iokit_memory || vbr.user_tag != VM_MEMORY_IOKIT)
637 		    && coredumpok(map, vmoffset)) ? vmsize : 0;
638 
639 		if (fsize) {
640 			int64_t resid = 0;
641 			const enum uio_seg sflg = IS_64BIT_PROCESS(core_proc) ?
642 			    UIO_USERSPACE64 : UIO_USERSPACE32;
643 
644 			error = vn_rdwr_64(UIO_WRITE, vp, vmoffset, fsize,
645 			    foffset, sflg, IO_NODELOCKED | IO_UNIT,
646 			    cred, &resid, core_proc);
647 
648 			if (error) {
649 				/*
650 				 * Mark segment as empty
651 				 */
652 				fsize = 0;
653 				COREDUMPLOG("failed to write segment %llx->%llx: error %d", vmoffset, vmoffset + vmsize, error);
654 			} else if (resid) {
655 				/*
656 				 * Partial write. Extend the file size so
657 				 * that the segment command contains a valid
658 				 * range of offsets, possibly creating a hole.
659 				 */
660 				VATTR_INIT(vap);
661 				VATTR_SET(vap, va_data_size, foffset + fsize);
662 				vnode_setattr(vp, vap, ctx);
663 				COREDUMPLOG("partially wrote segment %llx->%llx, resid %lld", vmoffset, vmoffset + vmsize, resid);
664 			}
665 		} else {
666 			COREDUMPLOG("skipping unreadable segment %llx->%llx", vmoffset, vmoffset + vmsize);
667 		}
668 
669 		/*
670 		 *	Fill in segment command structure.
671 		 */
672 
673 		if (is_64) {
674 			sc64 = (struct segment_command_64 *)(header + hoffset);
675 			sc64->cmd = LC_SEGMENT_64;
676 			sc64->cmdsize = sizeof(struct segment_command_64);
677 			/* segment name is zeroed by kmem_alloc */
678 			sc64->segname[0] = 0;
679 			sc64->vmaddr = vmoffset;
680 			sc64->vmsize = vmsize;
681 			sc64->fileoff = foffset;
682 			sc64->filesize = fsize;
683 			sc64->maxprot = maxprot;
684 			sc64->initprot = prot;
685 			sc64->nsects = 0;
686 			sc64->flags = 0;
687 		} else {
688 			sc = (struct segment_command *) (header + hoffset);
689 			sc->cmd = LC_SEGMENT;
690 			sc->cmdsize = sizeof(struct segment_command);
691 			/* segment name is zeroed by kmem_alloc */
692 			sc->segname[0] = 0;
693 			sc->vmaddr = CAST_DOWN_EXPLICIT(uint32_t, vmoffset);
694 			sc->vmsize = CAST_DOWN_EXPLICIT(uint32_t, vmsize);
695 			sc->fileoff = CAST_DOWN_EXPLICIT(uint32_t, foffset); /* will never truncate */
696 			sc->filesize = CAST_DOWN_EXPLICIT(uint32_t, fsize); /* will never truncate */
697 			sc->maxprot = maxprot;
698 			sc->initprot = prot;
699 			sc->nsects = 0;
700 			sc->flags = 0;
701 		}
702 
703 		hoffset += segment_command_sz;
704 		foffset += fsize;
705 		vmoffset += vmsize;
706 		segment_count--;
707 	}
708 	COREDUMPLOG("max file offset: %lld", foffset);
709 
710 	/*
711 	 * If there are remaining segments which have not been written
712 	 * out because break in the loop above, then they were not counted
713 	 * because they exceed the real address space of the executable
714 	 * type: remove them from the header's count.  This is OK, since
715 	 * we are allowed to have a sparse area following the segments.
716 	 */
717 	if (is_64) {
718 		mh64->ncmds -= segment_count;
719 		mh64->sizeofcmds -= segment_count * segment_command_sz;
720 	} else {
721 		mh->ncmds -= segment_count;
722 		mh->sizeofcmds -= segment_count * segment_command_sz;
723 	}
724 
725 	/* Add LC_NOTES */
726 	COREDUMPLOG("dumping %zu notes", notes_count);
727 	if (dump_notes(core_proc, header, hoffset, vp, foffset) != 0) {
728 		error = EFAULT;
729 		goto out;
730 	}
731 
732 	tir1.header = header;
733 	tir1.hoffset = hoffset + notes_size;
734 	tir1.flavors = flavors;
735 	tir1.tstate_size = tstate_size;
736 	COREDUMPLOG("dumping %zu threads", thread_count);
737 	task_act_iterate_wth_args_locked(task, collectth_state, &tir1);
738 
739 	/*
740 	 *	Write out the Mach header at the beginning of the
741 	 *	file.  OK to use a 32 bit write for this.
742 	 */
743 	error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, (int)MIN(header_size, INT_MAX), (off_t)0,
744 	    UIO_SYSSPACE, IO_NODELOCKED | IO_UNIT, cred, (int *) 0, core_proc);
745 	if (error != KERN_SUCCESS) {
746 		COREDUMPLOG("failed to write mach header: error %d", error);
747 	}
748 	kmem_free(kernel_map, header, header_size);
749 
750 	if ((coredump_flags & COREDUMP_FULLFSYNC) && error == 0) {
751 		error = VNOP_IOCTL(vp, F_FULLFSYNC, (caddr_t)NULL, 0, ctx);
752 		if (error != KERN_SUCCESS) {
753 			COREDUMPLOG("failed to FULLFSYNC core: error %d", error);
754 		}
755 	}
756 out:
757 	if (vap) {
758 		kfree_type(struct vnode_attr, vap);
759 	}
760 	error1 = vnode_close(vp, FWRITE, ctx);
761 	if (error1 != KERN_SUCCESS) {
762 		COREDUMPLOG("failed to close core file: error %d", error1);
763 	}
764 out2:
765 #if CONFIG_AUDIT
766 	audit_proc_coredump(core_proc, name, error);
767 #endif
768 	if (alloced_name != NULL) {
769 		zfree(ZV_NAMEI, alloced_name);
770 	}
771 	if (alloced_format != NULL) {
772 		kfree_data(alloced_format, alloced_format_len + 1);
773 	}
774 	if (custom_location_entitlement != NULL) {
775 		kfree_data(custom_location_entitlement, custom_location_entitlement_len + 1);
776 	}
777 	if (error == 0) {
778 		error = error1;
779 	}
780 
781 	if (coredump_attempted) {
782 		if (error != 0) {
783 			COREDUMPLOG("core dump failed: error %d\n", error);
784 		} else {
785 			COREDUMPLOG("core dump succeeded");
786 		}
787 	}
788 
789 	if (task_locked) {
790 		task_unlock(task);
791 	}
792 
793 	return error;
794 }
795 
796 #else /* CONFIG_COREDUMP */
797 
798 /* When core dumps aren't needed, no need to compile this file at all */
799 
800 #error assertion failed: this section is not compiled
801 
802 #endif /* CONFIG_COREDUMP */
803