xref: /xnu-10063.121.3/bsd/kern/kern_fork.c (revision 2c2f96dc2b9a4408a43d3150ae9c105355ca3daa)
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 /* Copyright (c) 1995, 1997 Apple Computer, Inc. All Rights Reserved */
29 /*
30  * Copyright (c) 1982, 1986, 1989, 1991, 1993
31  *	The Regents of the University of California.  All rights reserved.
32  * (c) UNIX System Laboratories, Inc.
33  * All or some portions of this file are derived from material licensed
34  * to the University of California by American Telephone and Telegraph
35  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
36  * the permission of UNIX System Laboratories, Inc.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)kern_fork.c	8.8 (Berkeley) 2/14/95
67  */
68 /*
69  * NOTICE: This file was modified by McAfee Research in 2004 to introduce
70  * support for mandatory and extensible security protections.  This notice
71  * is included in support of clause 2.2 (b) of the Apple Public License,
72  * Version 2.0.
73  */
74 /*
75  * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
76  * support for mandatory and extensible security protections.  This notice
77  * is included in support of clause 2.2 (b) of the Apple Public License,
78  * Version 2.0.
79  */
80 
81 #include <kern/assert.h>
82 #include <kern/bits.h>
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/filedesc.h>
86 #include <sys/kernel.h>
87 #include <sys/malloc.h>
88 #include <sys/proc_internal.h>
89 #include <sys/kauth.h>
90 #include <sys/user.h>
91 #include <sys/reason.h>
92 #include <sys/resourcevar.h>
93 #include <sys/vnode_internal.h>
94 #include <sys/file_internal.h>
95 #include <sys/acct.h>
96 #include <sys/codesign.h>
97 #include <sys/sysent.h>
98 #include <sys/sysproto.h>
99 #include <sys/ulock.h>
100 #if CONFIG_PERSONAS
101 #include <sys/persona.h>
102 #endif
103 #include <sys/doc_tombstone.h>
104 #if CONFIG_DTRACE
105 /* Do not include dtrace.h, it redefines kmem_[alloc/free] */
106 extern void (*dtrace_proc_waitfor_exec_ptr)(proc_t);
107 extern void dtrace_proc_fork(proc_t, proc_t, int);
108 
109 /*
110  * Since dtrace_proc_waitfor_exec_ptr can be added/removed in dtrace_subr.c,
111  * we will store its value before actually calling it.
112  */
113 static void (*dtrace_proc_waitfor_hook)(proc_t) = NULL;
114 
115 #include <sys/dtrace_ptss.h>
116 #endif
117 
118 #include <security/audit/audit.h>
119 
120 #include <mach/mach_types.h>
121 #include <kern/coalition.h>
122 #include <kern/kern_types.h>
123 #include <kern/kalloc.h>
124 #include <kern/mach_param.h>
125 #include <kern/task.h>
126 #include <kern/thread.h>
127 #include <kern/thread_call.h>
128 #include <kern/zalloc.h>
129 
130 #if CONFIG_MACF
131 #include <security/mac_framework.h>
132 #include <security/mac_mach_internal.h>
133 #endif
134 
135 #include <vm/vm_map.h>
136 #include <vm/vm_protos.h>
137 #include <vm/vm_shared_region.h>
138 
139 #include <sys/shm_internal.h>   /* for shmfork() */
140 #include <mach/task.h>          /* for thread_create() */
141 #include <mach/thread_act.h>    /* for thread_resume() */
142 
143 #include <sys/sdt.h>
144 
145 #if CONFIG_MEMORYSTATUS
146 #include <sys/kern_memorystatus.h>
147 #endif
148 
149 static const uint64_t startup_serial_num_procs = 300;
150 bool startup_serial_logging_active = true;
151 
152 /* XXX routines which should have Mach prototypes, but don't */
153 extern void act_thread_catt(void *ctx);
154 void thread_set_child(thread_t child, int pid);
155 boolean_t thread_is_active(thread_t thread);
156 void *act_thread_csave(void);
157 extern boolean_t task_is_exec_copy(task_t);
158 int nextpidversion = 0;
159 
160 void ipc_task_enable(task_t task);
161 
162 proc_t forkproc(proc_t, cloneproc_flags_t);
163 void forkproc_free(proc_t);
164 thread_t fork_create_child(task_t parent_task,
165     coalition_t *parent_coalitions,
166     proc_t child,
167     int is_64bit_addr,
168     int is_64bit_data,
169     cloneproc_flags_t clone_flags);
170 
171 __private_extern__ const size_t uthread_size = sizeof(struct uthread);
172 static LCK_GRP_DECLARE(rethrottle_lock_grp, "rethrottle");
173 
174 os_refgrp_decl(, p_refgrp, "proc", NULL);
175 
176 extern const size_t task_alignment;
177 const size_t proc_alignment = _Alignof(struct proc);
178 
179 extern size_t task_struct_size;
180 size_t proc_struct_size = sizeof(struct proc);
181 size_t proc_and_task_size;
182 
183 ZONE_DECLARE_ID(ZONE_ID_PROC_TASK, struct proc);
184 SECURITY_READ_ONLY_LATE(zone_t) proc_task_zone;
185 
186 KALLOC_TYPE_DEFINE(proc_stats_zone, struct pstats, KT_DEFAULT);
187 
188 /*
189  * fork1
190  *
191  * Description:	common code used by all new process creation other than the
192  *		bootstrap of the initial process on the system
193  *
194  * Parameters: parent_proc		parent process of the process being
195  *		child_threadp		pointer to location to receive the
196  *					Mach thread_t of the child process
197  *					created
198  *		kind			kind of creation being requested
199  *		coalitions		if spawn, the set of coalitions the
200  *					child process should join, or NULL to
201  *					inherit the parent's. On non-spawns,
202  *					this param is ignored and the child
203  *					always inherits the parent's
204  *					coalitions.
205  *
206  * Notes:	Permissable values for 'kind':
207  *
208  *		PROC_CREATE_FORK	Create a complete process which will
209  *					return actively running in both the
210  *					parent and the child; the child copies
211  *					the parent address space.
212  *		PROC_CREATE_SPAWN	Create a complete process which will
213  *					return actively running in the parent
214  *					only after returning actively running
215  *					in the child; the child address space
216  *					is newly created by an image activator,
217  *					after which the child is run.
218  *
219  *		At first it may seem strange that we return the child thread
220  *		address rather than process structure, since the process is
221  *		the only part guaranteed to be "new"; however, since we do
222  *		not actualy adjust other references between Mach and BSD, this
223  *		is the only method which guarantees us the ability to get
224  *		back to the other information.
225  */
226 int
fork1(proc_t parent_proc,thread_t * child_threadp,int kind,coalition_t * coalitions)227 fork1(proc_t parent_proc, thread_t *child_threadp, int kind, coalition_t *coalitions)
228 {
229 	proc_t child_proc = NULL;       /* set in switch, but compiler... */
230 	thread_t child_thread = NULL;
231 	uid_t uid;
232 	size_t count;
233 	int err = 0;
234 	int spawn = 0;
235 	rlim_t rlimit_nproc_cur;
236 
237 	/*
238 	 * Although process entries are dynamically created, we still keep
239 	 * a global limit on the maximum number we will create.  Don't allow
240 	 * a nonprivileged user to use the last process; don't let root
241 	 * exceed the limit. The variable nprocs is the current number of
242 	 * processes, maxproc is the limit.
243 	 */
244 	uid = kauth_getruid();
245 	proc_list_lock();
246 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
247 #if (DEVELOPMENT || DEBUG) && !defined(XNU_TARGET_OS_OSX)
248 		/*
249 		 * On the development kernel, panic so that the fact that we hit
250 		 * the process limit is obvious, as this may very well wedge the
251 		 * system.
252 		 */
253 		panic("The process table is full; parent pid=%d", proc_getpid(parent_proc));
254 #endif
255 		proc_list_unlock();
256 		tablefull("proc");
257 		return EAGAIN;
258 	}
259 	proc_list_unlock();
260 
261 	/*
262 	 * Increment the count of procs running with this uid. Don't allow
263 	 * a nonprivileged user to exceed their current limit, which is
264 	 * always less than what an rlim_t can hold.
265 	 * (locking protection is provided by list lock held in chgproccnt)
266 	 */
267 	count = chgproccnt(uid, 1);
268 	rlimit_nproc_cur = proc_limitgetcur(parent_proc, RLIMIT_NPROC);
269 	if (uid != 0 &&
270 	    (rlim_t)count > rlimit_nproc_cur) {
271 #if (DEVELOPMENT || DEBUG) && !defined(XNU_TARGET_OS_OSX)
272 		/*
273 		 * On the development kernel, panic so that the fact that we hit
274 		 * the per user process limit is obvious.  This may be less dire
275 		 * than hitting the global process limit, but we cannot rely on
276 		 * that.
277 		 */
278 		panic("The per-user process limit has been hit; parent pid=%d, uid=%d", proc_getpid(parent_proc), uid);
279 #endif
280 		err = EAGAIN;
281 		goto bad;
282 	}
283 
284 #if CONFIG_MACF
285 	/*
286 	 * Determine if MAC policies applied to the process will allow
287 	 * it to fork.  This is an advisory-only check.
288 	 */
289 	err = mac_proc_check_fork(parent_proc);
290 	if (err != 0) {
291 		goto bad;
292 	}
293 #endif
294 
295 	switch (kind) {
296 	case PROC_CREATE_SPAWN:
297 		/*
298 		 * A spawned process differs from a forked process in that
299 		 * the spawned process does not carry around the parents
300 		 * baggage with regard to address space copying, dtrace,
301 		 * and so on.
302 		 */
303 		spawn = 1;
304 
305 		OS_FALLTHROUGH;
306 
307 	case PROC_CREATE_FORK:
308 		/*
309 		 * When we clone the parent process, we are going to inherit
310 		 * its task attributes and memory, since when we fork, we
311 		 * will, in effect, create a duplicate of it, with only minor
312 		 * differences.  Contrarily, spawned processes do not inherit.
313 		 */
314 		if ((child_thread = cloneproc(proc_task(parent_proc),
315 		    spawn ? coalitions : NULL,
316 		    parent_proc,
317 		    spawn ? CLONEPROC_SPAWN : CLONEPROC_FORK)) == NULL) {
318 			/* Failed to create thread */
319 			err = EAGAIN;
320 			goto bad;
321 		}
322 
323 		/* child_proc = child_thread->task->proc; */
324 		child_proc = (proc_t)(get_bsdtask_info(get_threadtask(child_thread)));
325 
326 		if (!spawn) {
327 			/* Copy current thread state into the child thread (only for fork) */
328 			thread_dup(child_thread);
329 		}
330 
331 // XXX BEGIN: wants to move to be common code (and safe)
332 #if CONFIG_MACF
333 		/*
334 		 * allow policies to associate the credential/label that
335 		 * we referenced from the parent ... with the child
336 		 * JMM - this really isn't safe, as we can drop that
337 		 *       association without informing the policy in other
338 		 *       situations (keep long enough to get policies changed)
339 		 */
340 		mac_cred_label_associate_fork(proc_ucred_unsafe(child_proc),
341 		    child_proc);
342 #endif
343 
344 		/*
345 		 * Propogate change of PID - may get new cred if auditing.
346 		 */
347 		set_security_token(child_proc, proc_ucred_unsafe(child_proc));
348 
349 		AUDIT_ARG(pid, proc_getpid(child_proc));
350 
351 // XXX END: wants to move to be common code (and safe)
352 
353 		/*
354 		 * Blow thread state information; this is what gives the child
355 		 * process its "return" value from a fork() call.
356 		 *
357 		 * Note: this should probably move to fork() proper, since it
358 		 * is not relevent to spawn, and the value won't matter
359 		 * until we resume the child there.  If you are in here
360 		 * refactoring code, consider doing this at the same time.
361 		 */
362 		thread_set_child(child_thread, proc_getpid(child_proc));
363 
364 		child_proc->p_acflag = AFORK;   /* forked but not exec'ed */
365 
366 #if CONFIG_DTRACE
367 		dtrace_proc_fork(parent_proc, child_proc, spawn);
368 #endif  /* CONFIG_DTRACE */
369 		if (!spawn) {
370 			/*
371 			 * Of note, we need to initialize the bank context behind
372 			 * the protection of the proc_trans lock to prevent a race with exit.
373 			 */
374 			task_bank_init(get_threadtask(child_thread));
375 		}
376 
377 		break;
378 
379 	default:
380 		panic("fork1 called with unknown kind %d", kind);
381 		break;
382 	}
383 
384 
385 	/* return the thread pointer to the caller */
386 	*child_threadp = child_thread;
387 
388 bad:
389 	/*
390 	 * In the error case, we return a 0 value for the returned pid (but
391 	 * it is ignored in the trampoline due to the error return); this
392 	 * is probably not necessary.
393 	 */
394 	if (err) {
395 		(void)chgproccnt(uid, -1);
396 	}
397 
398 	return err;
399 }
400 
401 
402 
403 
404 /*
405  * fork_create_child
406  *
407  * Description:	Common operations associated with the creation of a child
408  *		process. Return with new task and first thread's control port movable
409  *      and not pinned.
410  *
411  * Parameters:	parent_task		parent task
412  *		parent_coalitions	parent's set of coalitions
413  *		child_proc			child process
414  *		inherit_memory		TRUE, if the parents address space is
415  *							to be inherited by the child
416  *		is_64bit_addr		TRUE, if the child being created will
417  *							be associated with a 64 bit address space
418  *		is_64bit_data		TRUE if the child being created will use a
419  *                                                       64-bit register state
420  *		in_exec				TRUE, if called from execve or posix spawn set exec
421  *							FALSE, if called from fork or vfexec
422  *
423  * Note:	This code is called in the fork() case, from the execve() call
424  *		graph, from the posix_spawn() call graph (which implicitly
425  *		includes a vfork() equivalent call, and in the system
426  *		bootstrap case.
427  *
428  *		It creates a new task and thread (and as a side effect of the
429  *		thread creation, a uthread) in the parent coalition set, which is
430  *		then associated with the process 'child'.  If the parent
431  *		process address space is to be inherited, then a flag
432  *		indicates that the newly created task should inherit this from
433  *		the child task.
434  *
435  *		As a special concession to bootstrapping the initial process
436  *		in the system, it's possible for 'parent_task' to be TASK_NULL;
437  *		in this case, 'inherit_memory' MUST be FALSE.
438  */
439 thread_t
fork_create_child(task_t parent_task,coalition_t * parent_coalitions,proc_t child_proc,int is_64bit_addr,int is_64bit_data,cloneproc_flags_t clone_flags)440 fork_create_child(task_t parent_task,
441     coalition_t *parent_coalitions,
442     proc_t child_proc,
443     int is_64bit_addr,
444     int is_64bit_data,
445     cloneproc_flags_t clone_flags)
446 {
447 	thread_t        child_thread = NULL;
448 	task_t          child_task;
449 	kern_return_t   result;
450 	proc_ro_t       proc_ro;
451 	bool inherit_memory = !!(clone_flags & CLONEPROC_FORK);
452 	bool in_exec = !!(clone_flags & CLONEPROC_EXEC);
453 	/*
454 	 * Exec complete hook should be called for spawn and exec, but not for fork.
455 	 */
456 	uint8_t returnwaitflags = (!inherit_memory ? TRW_LEXEC_COMPLETE : 0) |
457 	    (TRW_LRETURNWAIT | TRW_LRETURNWAITER);
458 
459 	proc_ro = proc_get_ro(child_proc);
460 	if (proc_ro_task(proc_ro) != NULL) {
461 		panic("Proc_ro_task for newly created proc %p is not NULL", child_proc);
462 	}
463 
464 	child_task = proc_get_task_raw(child_proc);
465 
466 	/*
467 	 * Create a new task for the child process, IPC access to the new task will
468 	 * be set up after task has been fully initialized.
469 	 */
470 	result = task_create_internal(parent_task,
471 	    proc_ro,
472 	    parent_coalitions,
473 	    inherit_memory,
474 	    is_64bit_addr,
475 	    is_64bit_data,
476 	    TF_NONE,
477 	    TF_NONE,
478 	    in_exec ? TPF_EXEC_COPY : TPF_NONE,           /* Mark the task exec copy if in execve */
479 	    returnwaitflags,                              /* All created threads will wait in task_wait_to_return */
480 	    child_task);
481 	if (result != KERN_SUCCESS) {
482 		printf("%s: task_create_internal failed.  Code: %d\n",
483 		    __func__, result);
484 		goto bad;
485 	}
486 
487 	/* Set the child proc process to child task */
488 	proc_set_task(child_proc, child_task);
489 
490 	/* Set child task process to child proc */
491 	set_bsdtask_info(child_task, child_proc);
492 
493 	/* Propagate CPU limit timer from parent */
494 	if (timerisset(&child_proc->p_rlim_cpu)) {
495 		task_vtimer_set(child_task, TASK_VTIMER_RLIM);
496 	}
497 
498 	/*
499 	 * Set child process BSD visible scheduler priority if nice value
500 	 * inherited from parent
501 	 */
502 	if (child_proc->p_nice != 0) {
503 		resetpriority(child_proc);
504 	}
505 
506 	/*
507 	 * Create main thread for the child process. Its control port is not immovable/pinned
508 	 * until main_thread_set_immovable_pinned().
509 	 *
510 	 * The new thread is waiting on the event triggered by 'task_clear_return_wait'
511 	 */
512 	result = main_thread_create_waiting(child_task,
513 	    (thread_continue_t)task_wait_to_return,
514 	    task_get_return_wait_event(child_task),
515 	    &child_thread);
516 
517 	if (result != KERN_SUCCESS) {
518 		printf("%s: thread_create failed. Code: %d\n",
519 		    __func__, result);
520 		task_deallocate(child_task);
521 		child_task = NULL;
522 	}
523 
524 	/*
525 	 * Tag thread as being the first thread in its task.
526 	 */
527 	thread_set_tag(child_thread, THREAD_TAG_MAINTHREAD);
528 
529 bad:
530 	thread_yield_internal(1);
531 
532 	return child_thread;
533 }
534 
535 
536 /*
537  * fork
538  *
539  * Description:	fork system call.
540  *
541  * Parameters:	parent			Parent process to fork
542  *		uap (void)		[unused]
543  *		retval			Return value
544  *
545  * Returns:	0			Success
546  *		EAGAIN			Resource unavailable, try again
547  *
548  * Notes:	Attempts to create a new child process which inherits state
549  *		from the parent process.  If successful, the call returns
550  *		having created an initially suspended child process with an
551  *		extra Mach task and thread reference, for which the thread
552  *		is initially suspended.  Until we resume the child process,
553  *		it is not yet running.
554  *
555  *		The return information to the child is contained in the
556  *		thread state structure of the new child, and does not
557  *		become visible to the child through a normal return process,
558  *		since it never made the call into the kernel itself in the
559  *		first place.
560  *
561  *		After resuming the thread, this function returns directly to
562  *		the parent process which invoked the fork() system call.
563  *
564  * Important:	The child thread_resume occurs before the parent returns;
565  *		depending on scheduling latency, this means that it is not
566  *		deterministic as to whether the parent or child is scheduled
567  *		to run first.  It is entirely possible that the child could
568  *		run to completion prior to the parent running.
569  */
570 int
fork(proc_t parent_proc,__unused struct fork_args * uap,int32_t * retval)571 fork(proc_t parent_proc, __unused struct fork_args *uap, int32_t *retval)
572 {
573 	thread_t child_thread;
574 	int err;
575 
576 	retval[1] = 0;          /* flag parent return for user space */
577 
578 	if ((err = fork1(parent_proc, &child_thread, PROC_CREATE_FORK, NULL)) == 0) {
579 		task_t child_task;
580 		proc_t child_proc;
581 
582 		/* Return to the parent */
583 		child_proc = (proc_t)get_bsdthreadtask_info(child_thread);
584 		retval[0] = proc_getpid(child_proc);
585 
586 		child_task = (task_t)get_threadtask(child_thread);
587 		assert(child_task != TASK_NULL);
588 
589 		/* task_control_port_options has been inherited from parent, apply it */
590 		task_set_immovable_pinned(child_task);
591 		main_thread_set_immovable_pinned(child_thread);
592 
593 		/*
594 		 * Since the task ports for this new task are now set to be immovable,
595 		 * we can enable them.
596 		 */
597 		ipc_task_enable(get_threadtask(child_thread));
598 
599 		/*
600 		 * Drop the signal lock on the child which was taken on our
601 		 * behalf by forkproc()/cloneproc() to prevent signals being
602 		 * received by the child in a partially constructed state.
603 		 */
604 		proc_signalend(child_proc, 0);
605 		proc_transend(child_proc, 0);
606 
607 		/* flag the fork has occurred */
608 		proc_knote(parent_proc, NOTE_FORK | proc_getpid(child_proc));
609 		DTRACE_PROC1(create, proc_t, child_proc);
610 
611 #if CONFIG_DTRACE
612 		if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) {
613 			(*dtrace_proc_waitfor_hook)(child_proc);
614 		}
615 #endif
616 
617 		/*
618 		 * If current process died during the fork, the child would contain
619 		 * non consistent vmmap, kill the child and reap it internally.
620 		 */
621 		if (parent_proc->p_lflag & P_LEXIT || !thread_is_active(current_thread())) {
622 			task_terminate_internal(child_task);
623 			proc_list_lock();
624 			child_proc->p_listflag |= P_LIST_DEADPARENT;
625 			proc_list_unlock();
626 		}
627 
628 		/* "Return" to the child */
629 		task_clear_return_wait(get_threadtask(child_thread), TCRW_CLEAR_ALL_WAIT);
630 
631 		/* drop the extra references we got during the creation */
632 		task_deallocate(child_task);
633 		thread_deallocate(child_thread);
634 	}
635 
636 	return err;
637 }
638 
639 
640 /*
641  * cloneproc
642  *
643  * Description: Create a new process from a specified process.
644  *
645  * Parameters:	parent_task		The parent task to be cloned, or
646  *					TASK_NULL is task characteristics
647  *					are not to be inherited
648  *					be cloned, or TASK_NULL if the new
649  *					task is not to inherit the VM
650  *					characteristics of the parent
651  *		parent_proc		The parent process to be cloned
652  *		clone_flags		Clone flags to specify if the cloned
653  *					process should inherit memory,
654  *					marked as memory stat internal,
655  *					or if the cloneproc is called for exec.
656  *
657  * Returns:	!NULL			pointer to new child thread
658  *		NULL			Failure (unspecified)
659  *
660  * Note:	On return newly created child process has signal lock held
661  *		to block delivery of signal to it if called with lock set.
662  *		fork() code needs to explicity remove this lock before
663  *		signals can be delivered
664  *
665  *		In the case of bootstrap, this function can be called from
666  *		bsd_utaskbootstrap() in order to bootstrap the first process;
667  *		the net effect is to provide a uthread structure for the
668  *		kernel process associated with the kernel task.
669  *
670  * XXX:		Tristating using the value parent_task as the major key
671  *		and inherit_memory as the minor key is something we should
672  *		refactor later; we owe the current semantics, ultimately,
673  *		to the semantics of task_create_internal.  For now, we will
674  *		live with this being somewhat awkward.
675  */
676 thread_t
cloneproc(task_t parent_task,coalition_t * parent_coalitions,proc_t parent_proc,cloneproc_flags_t clone_flags)677 cloneproc(task_t parent_task, coalition_t *parent_coalitions, proc_t parent_proc, cloneproc_flags_t clone_flags)
678 {
679 #if !CONFIG_MEMORYSTATUS
680 #pragma unused(cloning_initproc)
681 #endif
682 	task_t child_task;
683 	proc_t child_proc;
684 	thread_t child_thread = NULL;
685 	bool cloning_initproc = !!(clone_flags & CLONEPROC_INITPROC);
686 	bool in_exec = !!(clone_flags & CLONEPROC_EXEC);
687 
688 	if ((child_proc = forkproc(parent_proc, clone_flags)) == NULL) {
689 		/* Failed to allocate new process */
690 		goto bad;
691 	}
692 
693 	/*
694 	 * In the case where the parent_task is TASK_NULL (during the init path)
695 	 * we make the assumption that the register size will be the same as the
696 	 * address space size since there's no way to determine the possible
697 	 * register size until an image is exec'd.
698 	 *
699 	 * The only architecture that has different address space and register sizes
700 	 * (arm64_32) isn't being used within kernel-space, so the above assumption
701 	 * always holds true for the init path.
702 	 */
703 	const int parent_64bit_addr = parent_proc->p_flag & P_LP64;
704 	const int parent_64bit_data = (parent_task == TASK_NULL) ? parent_64bit_addr : task_get_64bit_data(parent_task);
705 
706 	child_thread = fork_create_child(parent_task,
707 	    parent_coalitions,
708 	    child_proc,
709 	    parent_64bit_addr,
710 	    parent_64bit_data,
711 	    clone_flags);
712 
713 	if (child_thread == NULL) {
714 		/*
715 		 * Failed to create thread; now we must deconstruct the new
716 		 * process previously obtained from forkproc().
717 		 */
718 		forkproc_free(child_proc);
719 		goto bad;
720 	}
721 
722 	child_task = get_threadtask(child_thread);
723 	if (parent_64bit_addr) {
724 		OSBitOrAtomic(P_LP64, (UInt32 *)&child_proc->p_flag);
725 		get_bsdthread_info(child_thread)->uu_flag |= UT_LP64;
726 	} else {
727 		OSBitAndAtomic(~((uint32_t)P_LP64), (UInt32 *)&child_proc->p_flag);
728 		get_bsdthread_info(child_thread)->uu_flag &= ~UT_LP64;
729 	}
730 
731 #if CONFIG_MEMORYSTATUS
732 	if (cloning_initproc ||
733 	    (in_exec && (parent_proc->p_memstat_state & P_MEMSTAT_INTERNAL))) {
734 		proc_list_lock();
735 		child_proc->p_memstat_state |= P_MEMSTAT_INTERNAL;
736 		child_proc->p_memstat_effectivepriority = JETSAM_PRIORITY_INTERNAL;
737 		child_proc->p_memstat_requestedpriority = JETSAM_PRIORITY_INTERNAL;
738 		proc_list_unlock();
739 	}
740 	if (in_exec && parent_proc->p_memstat_relaunch_flags != P_MEMSTAT_RELAUNCH_UNKNOWN) {
741 		memorystatus_relaunch_flags_update(child_proc, parent_proc->p_memstat_relaunch_flags);
742 	}
743 #endif
744 
745 	/* make child visible */
746 	pinsertchild(parent_proc, child_proc, in_exec);
747 
748 	/*
749 	 * Make child runnable, set start time.
750 	 */
751 	child_proc->p_stat = SRUN;
752 bad:
753 	return child_thread;
754 }
755 
756 void
proc_set_sigact(proc_t p,int sig,user_addr_t sigact)757 proc_set_sigact(proc_t p, int sig, user_addr_t sigact)
758 {
759 	assert((sig > 0) && (sig < NSIG));
760 
761 	p->p_sigacts.ps_sigact[sig] = sigact;
762 }
763 
764 void
proc_set_trampact(proc_t p,int sig,user_addr_t trampact)765 proc_set_trampact(proc_t p, int sig, user_addr_t trampact)
766 {
767 	assert((sig > 0) && (sig < NSIG));
768 
769 	p->p_sigacts.ps_trampact[sig] = trampact;
770 }
771 
772 void
proc_set_sigact_trampact(proc_t p,int sig,user_addr_t sigact,user_addr_t trampact)773 proc_set_sigact_trampact(proc_t p, int sig, user_addr_t sigact, user_addr_t trampact)
774 {
775 	assert((sig > 0) && (sig < NSIG));
776 
777 	p->p_sigacts.ps_sigact[sig] = sigact;
778 	p->p_sigacts.ps_trampact[sig] = trampact;
779 }
780 
781 void
proc_reset_sigact(proc_t p,sigset_t sigs)782 proc_reset_sigact(proc_t p, sigset_t sigs)
783 {
784 	user_addr_t *sigacts = p->p_sigacts.ps_sigact;
785 	int nc;
786 
787 	while (sigs) {
788 		nc = ffs((unsigned int)sigs);
789 		if (sigacts[nc] != SIG_DFL) {
790 			sigacts[nc] = SIG_DFL;
791 		}
792 		sigs &= ~sigmask(nc);
793 	}
794 }
795 
796 /*
797  * Destroy a process structure that resulted from a call to forkproc(), but
798  * which must be returned to the system because of a subsequent failure
799  * preventing it from becoming active.
800  *
801  * Parameters:	p			The incomplete process from forkproc()
802  *
803  * Returns:	(void)
804  *
805  * Note:	This function should only be used in an error handler following
806  *		a call to forkproc().
807  *
808  *		Operations occur in reverse order of those in forkproc().
809  */
810 void
forkproc_free(proc_t p)811 forkproc_free(proc_t p)
812 {
813 	struct pgrp *pg;
814 
815 #if CONFIG_PERSONAS
816 	persona_proc_drop(p);
817 #endif /* CONFIG_PERSONAS */
818 
819 #if PSYNCH
820 	pth_proc_hashdelete(p);
821 #endif /* PSYNCH */
822 
823 	/* We held signal and a transition locks; drop them */
824 	proc_signalend(p, 0);
825 	proc_transend(p, 0);
826 
827 	/*
828 	 * If we have our own copy of the resource limits structure, we
829 	 * need to free it.  If it's a shared copy, we need to drop our
830 	 * reference on it.
831 	 */
832 	proc_limitdrop(p);
833 
834 #if SYSV_SHM
835 	/* Need to drop references to the shared memory segment(s), if any */
836 	if (p->vm_shm) {
837 		/*
838 		 * Use shmexec(): we have no address space, so no mappings
839 		 *
840 		 * XXX Yes, the routine is badly named.
841 		 */
842 		shmexec(p);
843 	}
844 #endif
845 
846 	/* Need to undo the effects of the fdt_fork(), if any */
847 	fdt_invalidate(p);
848 	fdt_destroy(p);
849 
850 	/*
851 	 * Drop the reference on a text vnode pointer, if any
852 	 * XXX This code is broken in forkproc(); see <rdar://4256419>;
853 	 * XXX if anyone ever uses this field, we will be extremely unhappy.
854 	 */
855 	if (p->p_textvp) {
856 		vnode_rele(p->p_textvp);
857 		p->p_textvp = NULL;
858 	}
859 
860 	/* Update the audit session proc count */
861 	AUDIT_SESSION_PROCEXIT(p);
862 
863 	lck_mtx_destroy(&p->p_mlock, &proc_mlock_grp);
864 	lck_mtx_destroy(&p->p_ucred_mlock, &proc_ucred_mlock_grp);
865 #if CONFIG_AUDIT
866 	lck_mtx_destroy(&p->p_audit_mlock, &proc_ucred_mlock_grp);
867 #endif /* CONFIG_AUDIT */
868 #if CONFIG_DTRACE
869 	lck_mtx_destroy(&p->p_dtrace_sprlock, &proc_lck_grp);
870 #endif
871 	lck_spin_destroy(&p->p_slock, &proc_slock_grp);
872 
873 	proc_list_lock();
874 	/* Decrement the count of processes in the system */
875 	nprocs--;
876 
877 	/* quit the group */
878 	pg = pgrp_leave_locked(p);
879 
880 	/* Take it out of process hash */
881 	assert((os_ref_get_raw_mask(&p->p_refcount) >> P_REF_BITS) == 1);
882 	assert((os_ref_get_raw_mask(&p->p_refcount) & P_REF_NEW) == P_REF_NEW);
883 	os_atomic_xor(&p->p_refcount, P_REF_NEW | P_REF_DEAD, relaxed);
884 
885 	/* Remove from hash if not a shadow proc */
886 	if (!proc_is_shadow(p)) {
887 		phash_remove_locked(p);
888 	}
889 
890 	proc_list_unlock();
891 
892 	pgrp_rele(pg);
893 
894 	thread_call_free(p->p_rcall);
895 
896 	/* Free allocated memory */
897 	zfree(proc_stats_zone, p->p_stats);
898 	p->p_stats = NULL;
899 	if (p->p_subsystem_root_path) {
900 		zfree(ZV_NAMEI, p->p_subsystem_root_path);
901 		p->p_subsystem_root_path = NULL;
902 	}
903 
904 	proc_checkdeadrefs(p);
905 	proc_wait_release(p);
906 }
907 
908 
909 /*
910  * forkproc
911  *
912  * Description:	Create a new process structure, given a parent process
913  *		structure.
914  *
915  * Parameters:	parent_proc		The parent process
916  *
917  * Returns:	!NULL			The new process structure
918  *		NULL			Error (insufficient free memory)
919  *
920  * Note:	When successful, the newly created process structure is
921  *		partially initialized; if a caller needs to deconstruct the
922  *		returned structure, they must call forkproc_free() to do so.
923  */
924 proc_t
forkproc(proc_t parent_proc,cloneproc_flags_t clone_flags)925 forkproc(proc_t parent_proc, cloneproc_flags_t clone_flags)
926 {
927 	static uint64_t nextuniqueid = 0;
928 	static pid_t lastpid = 0;
929 
930 	proc_t child_proc;      /* Our new process */
931 	int error = 0;
932 	struct pgrp *pg;
933 	uthread_t parent_uthread = current_uthread();
934 	rlim_t rlimit_cpu_cur;
935 	pid_t pid;
936 	struct proc_ro_data proc_ro_data = {};
937 	bool in_exec = !!(clone_flags & CLONEPROC_EXEC);
938 	bool in_fork = !!(clone_flags & CLONEPROC_FORK);
939 
940 	child_proc = zalloc_flags(proc_task_zone, Z_WAITOK | Z_ZERO);
941 
942 	child_proc->p_stats = zalloc_flags(proc_stats_zone, Z_WAITOK | Z_ZERO);
943 	child_proc->p_sigacts = parent_proc->p_sigacts;
944 	os_ref_init_mask(&child_proc->p_refcount, P_REF_BITS, &p_refgrp, P_REF_NEW);
945 	os_ref_init_raw(&child_proc->p_waitref, &p_refgrp);
946 	proc_ref_hold_proc_task_struct(child_proc);
947 
948 	/* allocate a callout for use by interval timers */
949 	child_proc->p_rcall = thread_call_allocate((thread_call_func_t)realitexpire, child_proc);
950 
951 
952 	/*
953 	 * Find an unused PID.
954 	 */
955 
956 	fdt_init(child_proc);
957 
958 	proc_list_lock();
959 
960 	if (!in_exec) {
961 		pid = lastpid;
962 		do {
963 			/*
964 			 * If the process ID prototype has wrapped around,
965 			 * restart somewhat above 0, as the low-numbered procs
966 			 * tend to include daemons that don't exit.
967 			 */
968 			if (++pid >= PID_MAX) {
969 				pid = 100;
970 			}
971 			if (pid == lastpid) {
972 				panic("Unable to allocate a new pid");
973 			}
974 
975 			/* if the pid stays in hash both for zombie and runniing state */
976 		} while (phash_find_locked(pid) != PROC_NULL ||
977 		    pghash_exists_locked(pid) ||
978 		    session_find_locked(pid) != SESSION_NULL);
979 
980 		lastpid = pid;
981 		nprocs++;
982 
983 		child_proc->p_pid = pid;
984 		proc_ro_data.p_idversion = OSIncrementAtomic(&nextpidversion);
985 		/* kernel process is handcrafted and not from fork, so start from 1 */
986 		proc_ro_data.p_uniqueid = ++nextuniqueid;
987 
988 		/* Insert in the hash, and inherit our group (and session) */
989 		phash_insert_locked(child_proc);
990 
991 		/* Check if the proc is from App Cryptex */
992 		if (parent_proc->p_ladvflag & P_RSR) {
993 			os_atomic_or(&child_proc->p_ladvflag, P_RSR, relaxed);
994 		}
995 	} else {
996 		/* For exec copy of the proc, copy the pid, pidversion and uniqueid of original proc */
997 		pid = parent_proc->p_pid;
998 		child_proc->p_pid = pid;
999 		proc_ro_data.p_idversion = parent_proc->p_proc_ro->p_idversion;
1000 		proc_ro_data.p_uniqueid = parent_proc->p_proc_ro->p_uniqueid;
1001 
1002 		nprocs++;
1003 		os_atomic_or(&child_proc->p_refcount, P_REF_SHADOW, relaxed);
1004 	}
1005 	pg = pgrp_enter_locked(parent_proc, child_proc);
1006 	proc_list_unlock();
1007 
1008 	if (proc_ro_data.p_uniqueid == startup_serial_num_procs) {
1009 		/*
1010 		 * Turn off startup serial logging now that we have reached
1011 		 * the defined number of startup processes.
1012 		 */
1013 		startup_serial_logging_active = false;
1014 	}
1015 
1016 	/*
1017 	 * We've identified the PID we are going to use;
1018 	 * initialize the new process structure.
1019 	 */
1020 	child_proc->p_stat = SIDL;
1021 
1022 	/*
1023 	 * The zero'ing of the proc was at the allocation time due to need
1024 	 * for insertion to hash.  Copy the section that is to be copied
1025 	 * directly from the parent.
1026 	 */
1027 	child_proc->p_forkcopy = parent_proc->p_forkcopy;
1028 
1029 	proc_ro_data.syscall_filter_mask = proc_syscall_filter_mask(parent_proc);
1030 	proc_ro_data.p_platform_data = proc_get_ro(parent_proc)->p_platform_data;
1031 
1032 	/*
1033 	 * Some flags are inherited from the parent.
1034 	 * Duplicate sub-structures as needed.
1035 	 * Increase reference counts on shared objects.
1036 	 * The p_stats substruct is set in vm_fork.
1037 	 */
1038 #if CONFIG_DELAY_IDLE_SLEEP
1039 	child_proc->p_flag = (parent_proc->p_flag & (P_LP64 | P_TRANSLATED | P_DISABLE_ASLR | P_DELAYIDLESLEEP | P_SUGID | P_AFFINITY));
1040 #else /* CONFIG_DELAY_IDLE_SLEEP */
1041 	child_proc->p_flag = (parent_proc->p_flag & (P_LP64 | P_TRANSLATED | P_DISABLE_ASLR | P_SUGID | P_AFFINITY));
1042 #endif /* CONFIG_DELAY_IDLE_SLEEP */
1043 
1044 	child_proc->p_vfs_iopolicy = (parent_proc->p_vfs_iopolicy & (P_VFS_IOPOLICY_INHERITED_MASK));
1045 
1046 	proc_set_responsible_pid(child_proc, parent_proc->p_responsible_pid);
1047 
1048 	/*
1049 	 * Note that if the current thread has an assumed identity, this
1050 	 * credential will be granted to the new process.
1051 	 * This is OK to do in exec, because it will be over-written during image activation
1052 	 * before the proc is visible.
1053 	 */
1054 	kauth_cred_set(&proc_ro_data.p_ucred.__smr_ptr, kauth_cred_get());
1055 
1056 	lck_mtx_init(&child_proc->p_mlock, &proc_mlock_grp, &proc_lck_attr);
1057 	lck_mtx_init(&child_proc->p_ucred_mlock, &proc_ucred_mlock_grp, &proc_lck_attr);
1058 #if CONFIG_AUDIT
1059 	lck_mtx_init(&child_proc->p_audit_mlock, &proc_ucred_mlock_grp, &proc_lck_attr);
1060 #endif /* CONFIG_AUDIT */
1061 #if CONFIG_DTRACE
1062 	lck_mtx_init(&child_proc->p_dtrace_sprlock, &proc_lck_grp, &proc_lck_attr);
1063 #endif
1064 	lck_spin_init(&child_proc->p_slock, &proc_slock_grp, &proc_lck_attr);
1065 
1066 	klist_init(&child_proc->p_klist);
1067 
1068 	if (child_proc->p_textvp != NULLVP) {
1069 		/* bump references to the text vnode */
1070 		/* Need to hold iocount across the ref call */
1071 		if ((error = vnode_getwithref(child_proc->p_textvp)) == 0) {
1072 			error = vnode_ref(child_proc->p_textvp);
1073 			vnode_put(child_proc->p_textvp);
1074 		}
1075 
1076 		if (error != 0) {
1077 			child_proc->p_textvp = NULLVP;
1078 		}
1079 	}
1080 	uint64_t csflag_inherit_mask = ~CS_KILLED;
1081 	if (!in_fork) {
1082 		/* All non-fork paths should not inherit GTA flag */
1083 		csflag_inherit_mask &= ~CS_GET_TASK_ALLOW;
1084 	}
1085 	proc_ro_data.p_csflags = ((uint32_t)proc_getcsflags(parent_proc) & csflag_inherit_mask);
1086 
1087 	child_proc->p_proc_ro = proc_ro_alloc(child_proc, &proc_ro_data, NULL, NULL);
1088 
1089 	/* update cred on proc */
1090 	proc_update_creds_onproc(child_proc, proc_ucred_unsafe(child_proc));
1091 
1092 	/* update audit session proc count */
1093 	AUDIT_SESSION_PROCNEW(child_proc);
1094 
1095 	/*
1096 	 * Copy the parents per process open file table to the child; if
1097 	 * there is a per-thread current working directory, set the childs
1098 	 * per-process current working directory to that instead of the
1099 	 * parents.
1100 	 */
1101 	if (fdt_fork(&child_proc->p_fd, parent_proc, parent_uthread->uu_cdir, in_exec) != 0) {
1102 		forkproc_free(child_proc);
1103 		child_proc = NULL;
1104 		goto bad;
1105 	}
1106 
1107 #if SYSV_SHM
1108 	if (parent_proc->vm_shm && !in_exec) {
1109 		/* XXX may fail to attach shm to child */
1110 		(void)shmfork(parent_proc, child_proc);
1111 	}
1112 #endif
1113 
1114 	/*
1115 	 * Child inherits the parent's plimit
1116 	 */
1117 	proc_limitfork(parent_proc, child_proc);
1118 
1119 	rlimit_cpu_cur = proc_limitgetcur(child_proc, RLIMIT_CPU);
1120 	if (rlimit_cpu_cur != RLIM_INFINITY) {
1121 		child_proc->p_rlim_cpu.tv_sec = (rlimit_cpu_cur > __INT_MAX__) ? __INT_MAX__ : rlimit_cpu_cur;
1122 	}
1123 
1124 	if (in_exec) {
1125 		/* Keep the original start time for exec'ed proc */
1126 		child_proc->p_stats->ps_start = parent_proc->p_stats->ps_start;
1127 		child_proc->p_start.tv_sec = parent_proc->p_start.tv_sec;
1128 		child_proc->p_start.tv_usec = parent_proc->p_start.tv_usec;
1129 	} else {
1130 		/* Intialize new process stats, including start time */
1131 		/* <rdar://6640543> non-zeroed portion contains garbage AFAICT */
1132 		microtime_with_abstime(&child_proc->p_start, &child_proc->p_stats->ps_start);
1133 	}
1134 
1135 	if (pg->pg_session->s_ttyvp != NULL && parent_proc->p_flag & P_CONTROLT) {
1136 		os_atomic_or(&child_proc->p_flag, P_CONTROLT, relaxed);
1137 	}
1138 
1139 	/*
1140 	 * block all signals to reach the process.
1141 	 * no transition race should be occuring with the child yet,
1142 	 * but indicate that the process is in (the creation) transition.
1143 	 */
1144 	proc_signalstart(child_proc, 0);
1145 	proc_transstart(child_proc, 0, 0);
1146 
1147 	child_proc->p_pcaction = 0;
1148 
1149 	TAILQ_INIT(&child_proc->p_uthlist);
1150 	TAILQ_INIT(&child_proc->p_aio_activeq);
1151 	TAILQ_INIT(&child_proc->p_aio_doneq);
1152 
1153 	/*
1154 	 * Copy work queue information
1155 	 *
1156 	 * Note: This should probably only happen in the case where we are
1157 	 *	creating a child that is a copy of the parent; since this
1158 	 *	routine is called in the non-duplication case of vfork()
1159 	 *	or posix_spawn(), then this information should likely not
1160 	 *	be duplicated.
1161 	 *
1162 	 * <rdar://6640553> Work queue pointers that no longer point to code
1163 	 */
1164 	child_proc->p_wqthread = parent_proc->p_wqthread;
1165 	child_proc->p_threadstart = parent_proc->p_threadstart;
1166 	child_proc->p_pthsize = parent_proc->p_pthsize;
1167 	if ((parent_proc->p_lflag & P_LREGISTER) != 0) {
1168 		child_proc->p_lflag |= P_LREGISTER;
1169 	}
1170 	child_proc->p_dispatchqueue_offset = parent_proc->p_dispatchqueue_offset;
1171 	child_proc->p_dispatchqueue_serialno_offset = parent_proc->p_dispatchqueue_serialno_offset;
1172 	child_proc->p_dispatchqueue_label_offset = parent_proc->p_dispatchqueue_label_offset;
1173 	child_proc->p_return_to_kernel_offset = parent_proc->p_return_to_kernel_offset;
1174 	child_proc->p_mach_thread_self_offset = parent_proc->p_mach_thread_self_offset;
1175 	child_proc->p_pth_tsd_offset = parent_proc->p_pth_tsd_offset;
1176 	child_proc->p_pthread_wq_quantum_offset = parent_proc->p_pthread_wq_quantum_offset;
1177 #if PSYNCH
1178 	pth_proc_hashinit(child_proc);
1179 #endif /* PSYNCH */
1180 
1181 #if CONFIG_PERSONAS
1182 	child_proc->p_persona = NULL;
1183 	if (parent_proc->p_persona) {
1184 		struct persona *persona = proc_persona_get(parent_proc);
1185 
1186 		if (persona) {
1187 			error = persona_proc_adopt(child_proc, persona, NULL);
1188 			if (error != 0) {
1189 				printf("forkproc: persona_proc_inherit failed (persona %d being destroyed?)\n",
1190 				    persona_get_id(persona));
1191 				forkproc_free(child_proc);
1192 				child_proc = NULL;
1193 				goto bad;
1194 			}
1195 		}
1196 	}
1197 #endif
1198 
1199 #if CONFIG_MEMORYSTATUS
1200 	/* Memorystatus init */
1201 	child_proc->p_memstat_state = 0;
1202 	child_proc->p_memstat_effectivepriority = JETSAM_PRIORITY_DEFAULT;
1203 	child_proc->p_memstat_requestedpriority = JETSAM_PRIORITY_DEFAULT;
1204 	child_proc->p_memstat_assertionpriority = 0;
1205 	child_proc->p_memstat_userdata          = 0;
1206 	child_proc->p_memstat_idle_start        = 0;
1207 	child_proc->p_memstat_idle_delta        = 0;
1208 	child_proc->p_memstat_memlimit          = 0;
1209 	child_proc->p_memstat_memlimit_active   = 0;
1210 	child_proc->p_memstat_memlimit_inactive = 0;
1211 	child_proc->p_memstat_relaunch_flags    = P_MEMSTAT_RELAUNCH_UNKNOWN;
1212 #if CONFIG_FREEZE
1213 	child_proc->p_memstat_freeze_sharedanon_pages = 0;
1214 #endif
1215 	child_proc->p_memstat_dirty = 0;
1216 	child_proc->p_memstat_idledeadline = 0;
1217 #endif /* CONFIG_MEMORYSTATUS */
1218 
1219 	if (parent_proc->p_subsystem_root_path) {
1220 		size_t parent_length = strlen(parent_proc->p_subsystem_root_path) + 1;
1221 		assert(parent_length <= MAXPATHLEN);
1222 		child_proc->p_subsystem_root_path = zalloc_flags(ZV_NAMEI,
1223 		    Z_WAITOK | Z_ZERO);
1224 		memcpy(child_proc->p_subsystem_root_path, parent_proc->p_subsystem_root_path, parent_length);
1225 	}
1226 
1227 bad:
1228 	return child_proc;
1229 }
1230 
1231 void
proc_lock(proc_t p)1232 proc_lock(proc_t p)
1233 {
1234 	LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_NOTOWNED);
1235 	lck_mtx_lock(&p->p_mlock);
1236 }
1237 
1238 void
proc_unlock(proc_t p)1239 proc_unlock(proc_t p)
1240 {
1241 	lck_mtx_unlock(&p->p_mlock);
1242 }
1243 
1244 void
proc_spinlock(proc_t p)1245 proc_spinlock(proc_t p)
1246 {
1247 	lck_spin_lock_grp(&p->p_slock, &proc_slock_grp);
1248 }
1249 
1250 void
proc_spinunlock(proc_t p)1251 proc_spinunlock(proc_t p)
1252 {
1253 	lck_spin_unlock(&p->p_slock);
1254 }
1255 
1256 void
proc_list_lock(void)1257 proc_list_lock(void)
1258 {
1259 	lck_mtx_lock(&proc_list_mlock);
1260 }
1261 
1262 void
proc_list_unlock(void)1263 proc_list_unlock(void)
1264 {
1265 	lck_mtx_unlock(&proc_list_mlock);
1266 }
1267 
1268 void
proc_ucred_lock(proc_t p)1269 proc_ucred_lock(proc_t p)
1270 {
1271 	lck_mtx_lock(&p->p_ucred_mlock);
1272 }
1273 
1274 void
proc_ucred_unlock(proc_t p)1275 proc_ucred_unlock(proc_t p)
1276 {
1277 	lck_mtx_unlock(&p->p_ucred_mlock);
1278 }
1279 
1280 void
proc_update_creds_onproc(proc_t p,kauth_cred_t cred)1281 proc_update_creds_onproc(proc_t p, kauth_cred_t cred)
1282 {
1283 	p->p_uid = kauth_cred_getuid(cred);
1284 	p->p_gid = kauth_cred_getgid(cred);
1285 	p->p_ruid = kauth_cred_getruid(cred);
1286 	p->p_rgid = kauth_cred_getrgid(cred);
1287 	p->p_svuid = kauth_cred_getsvuid(cred);
1288 	p->p_svgid = kauth_cred_getsvgid(cred);
1289 }
1290 
1291 
1292 bool
uthread_is64bit(struct uthread * uth)1293 uthread_is64bit(struct uthread *uth)
1294 {
1295 	return uth->uu_flag & UT_LP64;
1296 }
1297 
1298 void
uthread_init(task_t task,uthread_t uth,thread_ro_t tro_tpl,int workq_thread)1299 uthread_init(task_t task, uthread_t uth, thread_ro_t tro_tpl, int workq_thread)
1300 {
1301 	uthread_t uth_parent = current_uthread();
1302 
1303 	lck_spin_init(&uth->uu_rethrottle_lock, &rethrottle_lock_grp,
1304 	    LCK_ATTR_NULL);
1305 
1306 	/*
1307 	 * Lazily set the thread on the kernel VFS context
1308 	 * to the first thread made which will be vm_pageout_scan_thread.
1309 	 */
1310 	if (__improbable(vfs_context0.vc_thread == NULL)) {
1311 		extern thread_t vm_pageout_scan_thread;
1312 
1313 		assert(task == kernel_task);
1314 		assert(get_machthread(uth) == vm_pageout_scan_thread);
1315 		vfs_context0.vc_thread = get_machthread(uth);
1316 	}
1317 
1318 	if (task_get_64bit_addr(task)) {
1319 		uth->uu_flag |= UT_LP64;
1320 	}
1321 
1322 	/*
1323 	 * Thread inherits credential from the creating thread, if both
1324 	 * are in the same task.
1325 	 *
1326 	 * If the creating thread has no credential or is from another
1327 	 * task we can leave the new thread credential NULL.  If it needs
1328 	 * one later, it will be lazily assigned from the task's process.
1329 	 */
1330 	if (task == kernel_task) {
1331 		kauth_cred_set(&tro_tpl->tro_cred, vfs_context0.vc_ucred);
1332 		kauth_cred_set(&tro_tpl->tro_realcred, vfs_context0.vc_ucred);
1333 		tro_tpl->tro_proc = kernproc;
1334 		tro_tpl->tro_proc_ro = kernproc->p_proc_ro;
1335 	} else if (!task_is_a_corpse(task)) {
1336 		thread_ro_t curtro = current_thread_ro();
1337 		proc_t p = get_bsdtask_info(task);
1338 
1339 		if (task == curtro->tro_task) {
1340 			kauth_cred_set(&tro_tpl->tro_realcred,
1341 			    curtro->tro_realcred);
1342 			if (workq_thread) {
1343 				kauth_cred_set(&tro_tpl->tro_cred,
1344 				    curtro->tro_realcred);
1345 			} else {
1346 				kauth_cred_set(&tro_tpl->tro_cred,
1347 				    curtro->tro_cred);
1348 			}
1349 			tro_tpl->tro_proc_ro = curtro->tro_proc_ro;
1350 		} else {
1351 			kauth_cred_t cred = kauth_cred_proc_ref(p);
1352 			kauth_cred_set(&tro_tpl->tro_realcred, cred);
1353 			kauth_cred_set(&tro_tpl->tro_cred, cred);
1354 			kauth_cred_unref(&cred);
1355 			tro_tpl->tro_proc_ro = task_get_ro(task);
1356 		}
1357 		tro_tpl->tro_proc = p;
1358 
1359 		proc_lock(p);
1360 		if (workq_thread) {
1361 			/* workq_thread threads will not inherit masks */
1362 			uth->uu_sigmask = ~workq_threadmask;
1363 		} else if (uth_parent->uu_flag & UT_SAS_OLDMASK) {
1364 			uth->uu_sigmask = uth_parent->uu_oldmask;
1365 		} else {
1366 			uth->uu_sigmask = uth_parent->uu_sigmask;
1367 		}
1368 
1369 		TAILQ_INSERT_TAIL(&p->p_uthlist, uth, uu_list);
1370 		proc_unlock(p);
1371 
1372 #if CONFIG_DTRACE
1373 		if (p->p_dtrace_ptss_pages != NULL) {
1374 			uth->t_dtrace_scratch = dtrace_ptss_claim_entry(p);
1375 		}
1376 #endif
1377 	} else {
1378 		tro_tpl->tro_proc_ro = task_get_ro(task);
1379 	}
1380 
1381 	uth->uu_pending_sigreturn = 0;
1382 	uthread_init_proc_refcount(uth);
1383 }
1384 
1385 mach_port_name_t
uthread_joiner_port(struct uthread * uth)1386 uthread_joiner_port(struct uthread *uth)
1387 {
1388 	return uth->uu_save.uus_bsdthread_terminate.kport;
1389 }
1390 
1391 user_addr_t
uthread_joiner_address(uthread_t uth)1392 uthread_joiner_address(uthread_t uth)
1393 {
1394 	return uth->uu_save.uus_bsdthread_terminate.ulock_addr;
1395 }
1396 
1397 void
uthread_joiner_wake(task_t task,uthread_t uth)1398 uthread_joiner_wake(task_t task, uthread_t uth)
1399 {
1400 	struct _bsdthread_terminate bts = uth->uu_save.uus_bsdthread_terminate;
1401 
1402 	assert(bts.ulock_addr);
1403 	bzero(&uth->uu_save.uus_bsdthread_terminate, sizeof(bts));
1404 
1405 	int flags = UL_UNFAIR_LOCK | ULF_WAKE_ALL | ULF_WAKE_ALLOW_NON_OWNER;
1406 	(void)ulock_wake(task, flags, bts.ulock_addr, 0);
1407 	mach_port_deallocate(get_task_ipcspace(task), bts.kport);
1408 }
1409 
1410 /*
1411  * This routine frees the thread name field of the uthread_t structure. Split out of
1412  * uthread_cleanup() so thread name does not get deallocated while generating a corpse fork.
1413  */
1414 void
uthread_cleanup_name(uthread_t uth)1415 uthread_cleanup_name(uthread_t uth)
1416 {
1417 	/*
1418 	 * <rdar://17834538>
1419 	 * Set pth_name to NULL before calling free().
1420 	 * Previously there was a race condition in the
1421 	 * case this code was executing during a stackshot
1422 	 * where the stackshot could try and copy pth_name
1423 	 * after it had been freed and before if was marked
1424 	 * as null.
1425 	 */
1426 	if (uth->pth_name != NULL) {
1427 		void *pth_name = uth->pth_name;
1428 		uth->pth_name = NULL;
1429 		kfree_data(pth_name, MAXTHREADNAMESIZE);
1430 	}
1431 	return;
1432 }
1433 
1434 /*
1435  * This routine frees all the BSD context in uthread except the credential.
1436  * It does not free the uthread structure as well
1437  */
1438 void
uthread_cleanup(uthread_t uth,thread_ro_t tro)1439 uthread_cleanup(uthread_t uth, thread_ro_t tro)
1440 {
1441 	task_t task = tro->tro_task;
1442 	proc_t p    = tro->tro_proc;
1443 
1444 	uthread_assert_zero_proc_refcount(uth);
1445 
1446 	if (uth->uu_lowpri_window || uth->uu_throttle_info) {
1447 		/*
1448 		 * task is marked as a low priority I/O type
1449 		 * and we've somehow managed to not dismiss the throttle
1450 		 * through the normal exit paths back to user space...
1451 		 * no need to throttle this thread since its going away
1452 		 * but we do need to update our bookeeping w/r to throttled threads
1453 		 *
1454 		 * Calling this routine will clean up any throttle info reference
1455 		 * still inuse by the thread.
1456 		 */
1457 		throttle_lowpri_io(0);
1458 	}
1459 
1460 #if CONFIG_AUDIT
1461 	/*
1462 	 * Per-thread audit state should never last beyond system
1463 	 * call return.  Since we don't audit the thread creation/
1464 	 * removal, the thread state pointer should never be
1465 	 * non-NULL when we get here.
1466 	 */
1467 	assert(uth->uu_ar == NULL);
1468 #endif
1469 
1470 	if (uth->uu_select.nbytes) {
1471 		select_cleanup_uthread(&uth->uu_select);
1472 	}
1473 
1474 	if (uth->uu_cdir) {
1475 		vnode_rele(uth->uu_cdir);
1476 		uth->uu_cdir = NULLVP;
1477 	}
1478 
1479 	if (uth->uu_selset) {
1480 		select_set_free(uth->uu_selset);
1481 		uth->uu_selset = NULL;
1482 	}
1483 
1484 	os_reason_free(uth->uu_exit_reason);
1485 
1486 	if ((task != kernel_task) && p) {
1487 		/*
1488 		 * Remove the thread from the process list and
1489 		 * transfer [appropriate] pending signals to the process.
1490 		 * Do not remove the uthread from proc uthlist for exec
1491 		 * copy task, since they does not have a ref on proc and
1492 		 * would not have been added to the list.
1493 		 */
1494 		if (uth->uu_kqr_bound) {
1495 			kqueue_threadreq_unbind(p, uth->uu_kqr_bound);
1496 		}
1497 
1498 		if (get_bsdtask_info(task) == p) {
1499 			proc_lock(p);
1500 			TAILQ_REMOVE(&p->p_uthlist, uth, uu_list);
1501 			p->p_siglist |= (uth->uu_siglist & execmask & (~p->p_sigignore | sigcantmask));
1502 			proc_unlock(p);
1503 		}
1504 
1505 #if CONFIG_DTRACE
1506 		struct dtrace_ptss_page_entry *tmpptr = uth->t_dtrace_scratch;
1507 		uth->t_dtrace_scratch = NULL;
1508 		if (tmpptr != NULL) {
1509 			dtrace_ptss_release_entry(p, tmpptr);
1510 		}
1511 #endif
1512 	} else {
1513 		assert(!uth->uu_kqr_bound);
1514 	}
1515 }
1516 
1517 /* This routine releases the credential stored in uthread */
1518 void
uthread_cred_ref(struct ucred * ucred)1519 uthread_cred_ref(struct ucred *ucred)
1520 {
1521 	kauth_cred_ref(ucred);
1522 }
1523 
1524 void
uthread_cred_free(struct ucred * ucred)1525 uthread_cred_free(struct ucred *ucred)
1526 {
1527 	kauth_cred_set(&ucred, NOCRED);
1528 }
1529 
1530 /* This routine frees the uthread structure held in thread structure */
1531 void
uthread_destroy(uthread_t uth)1532 uthread_destroy(uthread_t uth)
1533 {
1534 	uthread_destroy_proc_refcount(uth);
1535 
1536 	if (uth->t_tombstone) {
1537 		kfree_type(struct doc_tombstone, uth->t_tombstone);
1538 		uth->t_tombstone = NULL;
1539 	}
1540 
1541 #if CONFIG_DEBUG_SYSCALL_REJECTION
1542 	size_t const bitstr_len = BITMAP_SIZE(mach_trap_count + nsysent);
1543 
1544 	if (uth->syscall_rejection_mask) {
1545 		kfree_data(uth->syscall_rejection_mask, bitstr_len);
1546 		uth->syscall_rejection_mask = NULL;
1547 	}
1548 
1549 	if (uth->syscall_rejection_once_mask) {
1550 		kfree_data(uth->syscall_rejection_once_mask, bitstr_len);
1551 		uth->syscall_rejection_once_mask = NULL;
1552 	}
1553 #endif /* CONFIG_DEBUG_SYSCALL_REJECTION */
1554 
1555 	lck_spin_destroy(&uth->uu_rethrottle_lock, &rethrottle_lock_grp);
1556 
1557 	uthread_cleanup_name(uth);
1558 }
1559 
1560 user_addr_t
thread_get_sigreturn_token(thread_t thread)1561 thread_get_sigreturn_token(thread_t thread)
1562 {
1563 	uthread_t ut = (struct uthread *) get_bsdthread_info(thread);
1564 	return ut->uu_sigreturn_token;
1565 }
1566 
1567 uint32_t
thread_get_sigreturn_diversifier(thread_t thread)1568 thread_get_sigreturn_diversifier(thread_t thread)
1569 {
1570 	uthread_t ut = (struct uthread *) get_bsdthread_info(thread);
1571 	return ut->uu_sigreturn_diversifier;
1572 }
1573