xref: /xnu-8796.121.2/bsd/kern/kern_fork.c (revision c54f35ca767986246321eb901baf8f5ff7923f6a)
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 #include <os/log.h>
131 
132 #if CONFIG_MACF
133 #include <security/mac_framework.h>
134 #include <security/mac_mach_internal.h>
135 #endif
136 
137 #include <vm/vm_map.h>
138 #include <vm/vm_protos.h>
139 #include <vm/vm_shared_region.h>
140 
141 #include <sys/shm_internal.h>   /* for shmfork() */
142 #include <mach/task.h>          /* for thread_create() */
143 #include <mach/thread_act.h>    /* for thread_resume() */
144 
145 #include <sys/sdt.h>
146 
147 #if CONFIG_MEMORYSTATUS
148 #include <sys/kern_memorystatus.h>
149 #endif
150 
151 /* XXX routines which should have Mach prototypes, but don't */
152 void thread_set_parent(thread_t parent, int pid);
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_FLAGS_NONE : CLONEPROC_FLAGS_INHERIT_MEMORY)) == 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 			 * Also enable task ports for the new task for fork. In the spawn
331 			 * case, task ports enablement is delayed until after image activation
332 			 * since task map will be swapped during mach executable loading.
333 			 */
334 			ipc_task_enable(get_threadtask(child_thread));
335 		}
336 
337 // XXX BEGIN: wants to move to be common code (and safe)
338 #if CONFIG_MACF
339 		/*
340 		 * allow policies to associate the credential/label that
341 		 * we referenced from the parent ... with the child
342 		 * JMM - this really isn't safe, as we can drop that
343 		 *       association without informing the policy in other
344 		 *       situations (keep long enough to get policies changed)
345 		 */
346 		mac_cred_label_associate_fork(proc_ucred(child_proc), child_proc);
347 #endif
348 
349 		/*
350 		 * Propogate change of PID - may get new cred if auditing.
351 		 */
352 		set_security_token(child_proc);
353 
354 		AUDIT_ARG(pid, proc_getpid(child_proc));
355 
356 // XXX END: wants to move to be common code (and safe)
357 
358 		/*
359 		 * Blow thread state information; this is what gives the child
360 		 * process its "return" value from a fork() call.
361 		 *
362 		 * Note: this should probably move to fork() proper, since it
363 		 * is not relevent to spawn, and the value won't matter
364 		 * until we resume the child there.  If you are in here
365 		 * refactoring code, consider doing this at the same time.
366 		 */
367 		thread_set_child(child_thread, proc_getpid(child_proc));
368 
369 		child_proc->p_acflag = AFORK;   /* forked but not exec'ed */
370 
371 #if CONFIG_DTRACE
372 		dtrace_proc_fork(parent_proc, child_proc, spawn);
373 #endif  /* CONFIG_DTRACE */
374 		if (!spawn) {
375 			/*
376 			 * Of note, we need to initialize the bank context behind
377 			 * the protection of the proc_trans lock to prevent a race with exit.
378 			 */
379 			task_bank_init(get_threadtask(child_thread));
380 		}
381 
382 		break;
383 
384 	default:
385 		panic("fork1 called with unknown kind %d", kind);
386 		break;
387 	}
388 
389 
390 	/* return the thread pointer to the caller */
391 	*child_threadp = child_thread;
392 
393 bad:
394 	/*
395 	 * In the error case, we return a 0 value for the returned pid (but
396 	 * it is ignored in the trampoline due to the error return); this
397 	 * is probably not necessary.
398 	 */
399 	if (err) {
400 		(void)chgproccnt(uid, -1);
401 	}
402 
403 	return err;
404 }
405 
406 
407 
408 
409 /*
410  * fork_create_child
411  *
412  * Description:	Common operations associated with the creation of a child
413  *		process. Return with new task and first thread's control port movable
414  *      and not pinned.
415  *
416  * Parameters:	parent_task		parent task
417  *		parent_coalitions	parent's set of coalitions
418  *		child_proc			child process
419  *		inherit_memory		TRUE, if the parents address space is
420  *							to be inherited by the child
421  *		is_64bit_addr		TRUE, if the child being created will
422  *							be associated with a 64 bit address space
423  *		is_64bit_data		TRUE if the child being created will use a
424  *                                                       64-bit register state
425  *		in_exec				TRUE, if called from execve or posix spawn set exec
426  *							FALSE, if called from fork or vfexec
427  *
428  * Note:	This code is called in the fork() case, from the execve() call
429  *		graph, from the posix_spawn() call graph (which implicitly
430  *		includes a vfork() equivalent call, and in the system
431  *		bootstrap case.
432  *
433  *		It creates a new task and thread (and as a side effect of the
434  *		thread creation, a uthread) in the parent coalition set, which is
435  *		then associated with the process 'child'.  If the parent
436  *		process address space is to be inherited, then a flag
437  *		indicates that the newly created task should inherit this from
438  *		the child task.
439  *
440  *		As a special concession to bootstrapping the initial process
441  *		in the system, it's possible for 'parent_task' to be TASK_NULL;
442  *		in this case, 'inherit_memory' MUST be FALSE.
443  */
444 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)445 fork_create_child(task_t parent_task,
446     coalition_t *parent_coalitions,
447     proc_t child_proc,
448     int is_64bit_addr,
449     int is_64bit_data,
450     cloneproc_flags_t clone_flags)
451 {
452 	thread_t        child_thread = NULL;
453 	task_t          child_task;
454 	kern_return_t   result;
455 	proc_ro_t       proc_ro;
456 	bool inherit_memory = !!(clone_flags & CLONEPROC_FLAGS_INHERIT_MEMORY);
457 	bool in_exec = !!(clone_flags & CLONEPROC_FLAGS_FOR_EXEC);
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 	    (TRW_LRETURNWAIT | TRW_LRETURNWAITER),                     /* 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 		 * Drop the signal lock on the child which was taken on our
595 		 * behalf by forkproc()/cloneproc() to prevent signals being
596 		 * received by the child in a partially constructed state.
597 		 */
598 		proc_signalend(child_proc, 0);
599 		proc_transend(child_proc, 0);
600 
601 		/* flag the fork has occurred */
602 		proc_knote(parent_proc, NOTE_FORK | proc_getpid(child_proc));
603 		DTRACE_PROC1(create, proc_t, child_proc);
604 
605 #if CONFIG_DTRACE
606 		if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) {
607 			(*dtrace_proc_waitfor_hook)(child_proc);
608 		}
609 #endif
610 
611 		/*
612 		 * If current process died during the fork, the child would contain
613 		 * non consistent vmmap, kill the child and reap it internally.
614 		 */
615 		if (parent_proc->p_lflag & P_LEXIT || !thread_is_active(current_thread())) {
616 			task_terminate_internal(child_task);
617 			proc_list_lock();
618 			child_proc->p_listflag |= P_LIST_DEADPARENT;
619 			proc_list_unlock();
620 		}
621 
622 		/* "Return" to the child */
623 		task_clear_return_wait(get_threadtask(child_thread), TCRW_CLEAR_ALL_WAIT);
624 
625 		/* drop the extra references we got during the creation */
626 		task_deallocate(child_task);
627 		thread_deallocate(child_thread);
628 	}
629 
630 	return err;
631 }
632 
633 
634 /*
635  * cloneproc
636  *
637  * Description: Create a new process from a specified process.
638  *
639  * Parameters:	parent_task		The parent task to be cloned, or
640  *					TASK_NULL is task characteristics
641  *					are not to be inherited
642  *					be cloned, or TASK_NULL if the new
643  *					task is not to inherit the VM
644  *					characteristics of the parent
645  *		parent_proc		The parent process to be cloned
646  *		clone_flags		Clone flags to specify if the cloned
647  *					process should inherit memory,
648  *					marked as memory stat internal,
649  *					or if the cloneproc is called for exec.
650  *
651  * Returns:	!NULL			pointer to new child thread
652  *		NULL			Failure (unspecified)
653  *
654  * Note:	On return newly created child process has signal lock held
655  *		to block delivery of signal to it if called with lock set.
656  *		fork() code needs to explicity remove this lock before
657  *		signals can be delivered
658  *
659  *		In the case of bootstrap, this function can be called from
660  *		bsd_utaskbootstrap() in order to bootstrap the first process;
661  *		the net effect is to provide a uthread structure for the
662  *		kernel process associated with the kernel task.
663  *
664  * XXX:		Tristating using the value parent_task as the major key
665  *		and inherit_memory as the minor key is something we should
666  *		refactor later; we owe the current semantics, ultimately,
667  *		to the semantics of task_create_internal.  For now, we will
668  *		live with this being somewhat awkward.
669  */
670 thread_t
cloneproc(task_t parent_task,coalition_t * parent_coalitions,proc_t parent_proc,cloneproc_flags_t clone_flags)671 cloneproc(task_t parent_task, coalition_t *parent_coalitions, proc_t parent_proc, cloneproc_flags_t clone_flags)
672 {
673 #if !CONFIG_MEMORYSTATUS
674 #pragma unused(memstat_internal)
675 #endif
676 	task_t child_task;
677 	proc_t child_proc;
678 	thread_t child_thread = NULL;
679 	bool memstat_internal = !!(clone_flags & CLONEPROC_FLAGS_MEMSTAT_INTERNAL);
680 	bool in_exec = !!(clone_flags & CLONEPROC_FLAGS_FOR_EXEC);
681 
682 	if ((child_proc = forkproc(parent_proc, clone_flags)) == NULL) {
683 		/* Failed to allocate new process */
684 		goto bad;
685 	}
686 
687 	/*
688 	 * In the case where the parent_task is TASK_NULL (during the init path)
689 	 * we make the assumption that the register size will be the same as the
690 	 * address space size since there's no way to determine the possible
691 	 * register size until an image is exec'd.
692 	 *
693 	 * The only architecture that has different address space and register sizes
694 	 * (arm64_32) isn't being used within kernel-space, so the above assumption
695 	 * always holds true for the init path.
696 	 */
697 	const int parent_64bit_addr = parent_proc->p_flag & P_LP64;
698 	const int parent_64bit_data = (parent_task == TASK_NULL) ? parent_64bit_addr : task_get_64bit_data(parent_task);
699 
700 	child_thread = fork_create_child(parent_task,
701 	    parent_coalitions,
702 	    child_proc,
703 	    parent_64bit_addr,
704 	    parent_64bit_data,
705 	    clone_flags);
706 
707 	if (child_thread == NULL) {
708 		/*
709 		 * Failed to create thread; now we must deconstruct the new
710 		 * process previously obtained from forkproc().
711 		 */
712 		forkproc_free(child_proc);
713 		goto bad;
714 	}
715 
716 	child_task = get_threadtask(child_thread);
717 	if (parent_64bit_addr) {
718 		OSBitOrAtomic(P_LP64, (UInt32 *)&child_proc->p_flag);
719 		get_bsdthread_info(child_thread)->uu_flag |= UT_LP64;
720 	} else {
721 		OSBitAndAtomic(~((uint32_t)P_LP64), (UInt32 *)&child_proc->p_flag);
722 		get_bsdthread_info(child_thread)->uu_flag &= ~UT_LP64;
723 	}
724 
725 #if CONFIG_MEMORYSTATUS
726 	if (memstat_internal ||
727 	    (in_exec && (parent_proc->p_memstat_state & P_MEMSTAT_INTERNAL))) {
728 		proc_list_lock();
729 		child_proc->p_memstat_state |= P_MEMSTAT_INTERNAL;
730 		proc_list_unlock();
731 	}
732 	if (in_exec && parent_proc->p_memstat_relaunch_flags != P_MEMSTAT_RELAUNCH_UNKNOWN) {
733 		memorystatus_relaunch_flags_update(child_proc, parent_proc->p_memstat_relaunch_flags);
734 	}
735 #endif
736 
737 	/* make child visible */
738 	pinsertchild(parent_proc, child_proc, in_exec);
739 
740 	/*
741 	 * Make child runnable, set start time.
742 	 */
743 	child_proc->p_stat = SRUN;
744 bad:
745 	return child_thread;
746 }
747 
748 void
proc_set_sigact(proc_t p,int sig,user_addr_t sigact)749 proc_set_sigact(proc_t p, int sig, user_addr_t sigact)
750 {
751 	assert((sig > 0) && (sig < NSIG));
752 
753 	p->p_sigacts.ps_sigact[sig] = sigact;
754 }
755 
756 void
proc_set_trampact(proc_t p,int sig,user_addr_t trampact)757 proc_set_trampact(proc_t p, int sig, user_addr_t trampact)
758 {
759 	assert((sig > 0) && (sig < NSIG));
760 
761 	p->p_sigacts.ps_trampact[sig] = trampact;
762 }
763 
764 void
proc_set_sigact_trampact(proc_t p,int sig,user_addr_t sigact,user_addr_t trampact)765 proc_set_sigact_trampact(proc_t p, int sig, user_addr_t sigact, user_addr_t trampact)
766 {
767 	assert((sig > 0) && (sig < NSIG));
768 
769 	p->p_sigacts.ps_sigact[sig] = sigact;
770 	p->p_sigacts.ps_trampact[sig] = trampact;
771 }
772 
773 void
proc_reset_sigact(proc_t p,sigset_t sigs)774 proc_reset_sigact(proc_t p, sigset_t sigs)
775 {
776 	user_addr_t *sigacts = p->p_sigacts.ps_sigact;
777 	int nc;
778 
779 	while (sigs) {
780 		nc = ffs((unsigned int)sigs);
781 		if (sigacts[nc] != SIG_DFL) {
782 			sigacts[nc] = SIG_DFL;
783 		}
784 		sigs &= ~sigmask(nc);
785 	}
786 }
787 
788 /*
789  * Destroy a process structure that resulted from a call to forkproc(), but
790  * which must be returned to the system because of a subsequent failure
791  * preventing it from becoming active.
792  *
793  * Parameters:	p			The incomplete process from forkproc()
794  *
795  * Returns:	(void)
796  *
797  * Note:	This function should only be used in an error handler following
798  *		a call to forkproc().
799  *
800  *		Operations occur in reverse order of those in forkproc().
801  */
802 void
forkproc_free(proc_t p)803 forkproc_free(proc_t p)
804 {
805 	struct pgrp *pg;
806 
807 #if CONFIG_PERSONAS
808 	persona_proc_drop(p);
809 #endif /* CONFIG_PERSONAS */
810 
811 #if PSYNCH
812 	pth_proc_hashdelete(p);
813 #endif /* PSYNCH */
814 
815 	/* We held signal and a transition locks; drop them */
816 	proc_signalend(p, 0);
817 	proc_transend(p, 0);
818 
819 	/*
820 	 * If we have our own copy of the resource limits structure, we
821 	 * need to free it.  If it's a shared copy, we need to drop our
822 	 * reference on it.
823 	 */
824 	proc_limitdrop(p);
825 
826 #if SYSV_SHM
827 	/* Need to drop references to the shared memory segment(s), if any */
828 	if (p->vm_shm) {
829 		/*
830 		 * Use shmexec(): we have no address space, so no mappings
831 		 *
832 		 * XXX Yes, the routine is badly named.
833 		 */
834 		shmexec(p);
835 	}
836 #endif
837 
838 	/* Need to undo the effects of the fdt_fork(), if any */
839 	fdt_invalidate(p);
840 	fdt_destroy(p);
841 
842 	/*
843 	 * Drop the reference on a text vnode pointer, if any
844 	 * XXX This code is broken in forkproc(); see <rdar://4256419>;
845 	 * XXX if anyone ever uses this field, we will be extremely unhappy.
846 	 */
847 	if (p->p_textvp) {
848 		vnode_rele(p->p_textvp);
849 		p->p_textvp = NULL;
850 	}
851 
852 	/* Update the audit session proc count */
853 	AUDIT_SESSION_PROCEXIT(p);
854 
855 	lck_mtx_destroy(&p->p_mlock, &proc_mlock_grp);
856 	lck_mtx_destroy(&p->p_ucred_mlock, &proc_ucred_mlock_grp);
857 #if CONFIG_DTRACE
858 	lck_mtx_destroy(&p->p_dtrace_sprlock, &proc_lck_grp);
859 #endif
860 	lck_spin_destroy(&p->p_slock, &proc_slock_grp);
861 
862 	/* Release the credential reference */
863 	proc_set_ucred(p, NOCRED);
864 
865 	proc_list_lock();
866 	/* Decrement the count of processes in the system */
867 	nprocs--;
868 
869 	/* quit the group */
870 	pg = pgrp_leave_locked(p);
871 
872 	/* Take it out of process hash */
873 	assert((os_ref_get_raw_mask(&p->p_refcount) >> P_REF_BITS) == 1);
874 	assert((os_ref_get_raw_mask(&p->p_refcount) & P_REF_NEW) == P_REF_NEW);
875 	os_atomic_xor(&p->p_refcount, P_REF_NEW | P_REF_DEAD, relaxed);
876 
877 	/* Remove from hash if not a shadow proc */
878 	if (!proc_is_shadow(p)) {
879 		phash_remove_locked(p);
880 	}
881 
882 	proc_list_unlock();
883 
884 	pgrp_rele(pg);
885 
886 	thread_call_free(p->p_rcall);
887 
888 	/* Free allocated memory */
889 	zfree(proc_stats_zone, p->p_stats);
890 	p->p_stats = NULL;
891 	if (p->p_subsystem_root_path) {
892 		zfree(ZV_NAMEI, p->p_subsystem_root_path);
893 		p->p_subsystem_root_path = NULL;
894 	}
895 
896 	p->p_proc_ro = proc_ro_release_proc(p->p_proc_ro);
897 	if (p->p_proc_ro != NULL) {
898 		proc_ro_free(p->p_proc_ro);
899 		p->p_proc_ro = NULL;
900 	}
901 
902 	proc_checkdeadrefs(p);
903 	proc_wait_release(p);
904 }
905 
906 
907 /*
908  * forkproc
909  *
910  * Description:	Create a new process structure, given a parent process
911  *		structure.
912  *
913  * Parameters:	parent_proc		The parent process
914  *
915  * Returns:	!NULL			The new process structure
916  *		NULL			Error (insufficient free memory)
917  *
918  * Note:	When successful, the newly created process structure is
919  *		partially initialized; if a caller needs to deconstruct the
920  *		returned structure, they must call forkproc_free() to do so.
921  */
922 proc_t
forkproc(proc_t parent_proc,cloneproc_flags_t clone_flags)923 forkproc(proc_t parent_proc, cloneproc_flags_t clone_flags)
924 {
925 	static uint64_t nextuniqueid = 0;
926 	static pid_t lastpid = 0;
927 
928 	proc_t child_proc;      /* Our new process */
929 	int error = 0;
930 	struct pgrp *pg;
931 	uthread_t parent_uthread = current_uthread();
932 	rlim_t rlimit_cpu_cur;
933 	pid_t pid;
934 	struct proc_ro_data proc_ro_data = {};
935 	bool in_exec = !!(clone_flags & CLONEPROC_FLAGS_FOR_EXEC);
936 
937 	child_proc = zalloc_flags(proc_task_zone, Z_WAITOK | Z_ZERO);
938 
939 	child_proc->p_stats = zalloc_flags(proc_stats_zone, Z_WAITOK | Z_ZERO);
940 	child_proc->p_sigacts = parent_proc->p_sigacts;
941 	os_ref_init_mask(&child_proc->p_refcount, P_REF_BITS, &p_refgrp, P_REF_NEW);
942 	os_ref_init_raw(&child_proc->p_waitref, &p_refgrp);
943 	proc_ref_hold_proc_task_struct(child_proc);
944 
945 	/* allocate a callout for use by interval timers */
946 	child_proc->p_rcall = thread_call_allocate((thread_call_func_t)realitexpire, child_proc);
947 
948 
949 	/*
950 	 * Find an unused PID.
951 	 */
952 
953 	fdt_init(child_proc);
954 
955 	proc_list_lock();
956 
957 	if (!in_exec) {
958 		pid = lastpid;
959 		do {
960 			/*
961 			 * If the process ID prototype has wrapped around,
962 			 * restart somewhat above 0, as the low-numbered procs
963 			 * tend to include daemons that don't exit.
964 			 */
965 			if (++pid >= PID_MAX) {
966 				pid = 100;
967 			}
968 			if (pid == lastpid) {
969 				panic("Unable to allocate a new pid");
970 			}
971 
972 			/* if the pid stays in hash both for zombie and runniing state */
973 		} while (phash_find_locked(pid) != PROC_NULL ||
974 		    pghash_exists_locked(pid) ||
975 		    session_find_locked(pid) != SESSION_NULL);
976 
977 		lastpid = pid;
978 		nprocs++;
979 
980 		child_proc->p_pid = pid;
981 		proc_ro_data.p_idversion = OSIncrementAtomic(&nextpidversion);
982 		/* kernel process is handcrafted and not from fork, so start from 1 */
983 		proc_ro_data.p_uniqueid = ++nextuniqueid;
984 
985 		/* Insert in the hash, and inherit our group (and session) */
986 		phash_insert_locked(child_proc);
987 
988 		/* Check if the proc is from App Cryptex */
989 		if (parent_proc->p_ladvflag & P_RSR) {
990 			os_atomic_or(&child_proc->p_ladvflag, P_RSR, relaxed);
991 		}
992 	} else {
993 		/* For exec copy of the proc, copy the pid, pidversion and uniqueid of original proc */
994 		pid = parent_proc->p_pid;
995 		child_proc->p_pid = pid;
996 		proc_ro_data.p_idversion = parent_proc->p_proc_ro->p_idversion;
997 		proc_ro_data.p_uniqueid = parent_proc->p_proc_ro->p_uniqueid;
998 
999 		nprocs++;
1000 		os_atomic_or(&child_proc->p_refcount, P_REF_SHADOW, relaxed);
1001 	}
1002 	pg = pgrp_enter_locked(parent_proc, child_proc);
1003 	proc_list_unlock();
1004 
1005 	if (proc_ro_data.p_uniqueid == startup_serial_num_procs) {
1006 		/*
1007 		 * Turn off startup serial logging now that we have reached
1008 		 * the defined number of startup processes.
1009 		 */
1010 		startup_serial_logging_active = false;
1011 	}
1012 
1013 	/*
1014 	 * We've identified the PID we are going to use;
1015 	 * initialize the new process structure.
1016 	 */
1017 	child_proc->p_stat = SIDL;
1018 
1019 	/*
1020 	 * The zero'ing of the proc was at the allocation time due to need
1021 	 * for insertion to hash.  Copy the section that is to be copied
1022 	 * directly from the parent.
1023 	 */
1024 	child_proc->p_forkcopy = parent_proc->p_forkcopy;
1025 
1026 	proc_ro_data.syscall_filter_mask = proc_syscall_filter_mask(parent_proc);
1027 	proc_ro_data.p_platform_data = proc_get_ro(parent_proc)->p_platform_data;
1028 
1029 	/*
1030 	 * Some flags are inherited from the parent.
1031 	 * Duplicate sub-structures as needed.
1032 	 * Increase reference counts on shared objects.
1033 	 * The p_stats substruct is set in vm_fork.
1034 	 */
1035 #if CONFIG_DELAY_IDLE_SLEEP
1036 	child_proc->p_flag = (parent_proc->p_flag & (P_LP64 | P_TRANSLATED | P_DISABLE_ASLR | P_DELAYIDLESLEEP | P_SUGID | P_AFFINITY));
1037 #else /* CONFIG_DELAY_IDLE_SLEEP */
1038 	child_proc->p_flag = (parent_proc->p_flag & (P_LP64 | P_TRANSLATED | P_DISABLE_ASLR | P_SUGID | P_AFFINITY));
1039 #endif /* CONFIG_DELAY_IDLE_SLEEP */
1040 
1041 	child_proc->p_vfs_iopolicy = (parent_proc->p_vfs_iopolicy & (P_VFS_IOPOLICY_VALID_MASK));
1042 
1043 	child_proc->p_responsible_pid = parent_proc->p_responsible_pid;
1044 
1045 	/*
1046 	 * Note that if the current thread has an assumed identity, this
1047 	 * credential will be granted to the new process.
1048 	 */
1049 	kauth_cred_set(&proc_ro_data.p_ucred, kauth_cred_get());
1050 
1051 	lck_mtx_init(&child_proc->p_mlock, &proc_mlock_grp, &proc_lck_attr);
1052 	lck_mtx_init(&child_proc->p_ucred_mlock, &proc_ucred_mlock_grp, &proc_lck_attr);
1053 #if CONFIG_DTRACE
1054 	lck_mtx_init(&child_proc->p_dtrace_sprlock, &proc_lck_grp, &proc_lck_attr);
1055 #endif
1056 	lck_spin_init(&child_proc->p_slock, &proc_slock_grp, &proc_lck_attr);
1057 
1058 	klist_init(&child_proc->p_klist);
1059 
1060 	if (child_proc->p_textvp != NULLVP) {
1061 		/* bump references to the text vnode */
1062 		/* Need to hold iocount across the ref call */
1063 		if ((error = vnode_getwithref(child_proc->p_textvp)) == 0) {
1064 			error = vnode_ref(child_proc->p_textvp);
1065 			vnode_put(child_proc->p_textvp);
1066 		}
1067 
1068 		if (error != 0) {
1069 			child_proc->p_textvp = NULLVP;
1070 		}
1071 	}
1072 
1073 	/* Inherit the parent flags for code sign */
1074 	proc_ro_data.p_csflags = ((uint32_t)proc_getcsflags(parent_proc) & ~CS_KILLED);
1075 
1076 	child_proc->p_proc_ro = proc_ro_alloc(child_proc, &proc_ro_data, NULL, NULL);
1077 
1078 	/* update cred on proc */
1079 	proc_update_creds_onproc(child_proc);
1080 
1081 	/* update audit session proc count */
1082 	AUDIT_SESSION_PROCNEW(child_proc);
1083 
1084 	/*
1085 	 * Copy the parents per process open file table to the child; if
1086 	 * there is a per-thread current working directory, set the childs
1087 	 * per-process current working directory to that instead of the
1088 	 * parents.
1089 	 */
1090 	if (fdt_fork(&child_proc->p_fd, parent_proc, parent_uthread->uu_cdir, in_exec) != 0) {
1091 		forkproc_free(child_proc);
1092 		child_proc = NULL;
1093 		goto bad;
1094 	}
1095 
1096 #if SYSV_SHM
1097 	if (parent_proc->vm_shm && !in_exec) {
1098 		/* XXX may fail to attach shm to child */
1099 		(void)shmfork(parent_proc, child_proc);
1100 	}
1101 #endif
1102 
1103 	/*
1104 	 * Child inherits the parent's plimit
1105 	 */
1106 	proc_limitfork(parent_proc, child_proc);
1107 
1108 	rlimit_cpu_cur = proc_limitgetcur(child_proc, RLIMIT_CPU);
1109 	if (rlimit_cpu_cur != RLIM_INFINITY) {
1110 		child_proc->p_rlim_cpu.tv_sec = (rlimit_cpu_cur > __INT_MAX__) ? __INT_MAX__ : rlimit_cpu_cur;
1111 	}
1112 
1113 	if (in_exec) {
1114 		/* Keep the original start time for exec'ed proc */
1115 		child_proc->p_stats->ps_start = parent_proc->p_stats->ps_start;
1116 		child_proc->p_start.tv_sec = parent_proc->p_start.tv_sec;
1117 		child_proc->p_start.tv_usec = parent_proc->p_start.tv_usec;
1118 	} else {
1119 		/* Intialize new process stats, including start time */
1120 		/* <rdar://6640543> non-zeroed portion contains garbage AFAICT */
1121 		microtime_with_abstime(&child_proc->p_start, &child_proc->p_stats->ps_start);
1122 	}
1123 
1124 	if (pg->pg_session->s_ttyvp != NULL && parent_proc->p_flag & P_CONTROLT) {
1125 		os_atomic_or(&child_proc->p_flag, P_CONTROLT, relaxed);
1126 	}
1127 
1128 	/*
1129 	 * block all signals to reach the process.
1130 	 * no transition race should be occuring with the child yet,
1131 	 * but indicate that the process is in (the creation) transition.
1132 	 */
1133 	proc_signalstart(child_proc, 0);
1134 	proc_transstart(child_proc, 0, 0);
1135 
1136 	child_proc->p_pcaction = 0;
1137 
1138 	TAILQ_INIT(&child_proc->p_uthlist);
1139 	TAILQ_INIT(&child_proc->p_aio_activeq);
1140 	TAILQ_INIT(&child_proc->p_aio_doneq);
1141 
1142 	/*
1143 	 * Copy work queue information
1144 	 *
1145 	 * Note: This should probably only happen in the case where we are
1146 	 *	creating a child that is a copy of the parent; since this
1147 	 *	routine is called in the non-duplication case of vfork()
1148 	 *	or posix_spawn(), then this information should likely not
1149 	 *	be duplicated.
1150 	 *
1151 	 * <rdar://6640553> Work queue pointers that no longer point to code
1152 	 */
1153 	child_proc->p_wqthread = parent_proc->p_wqthread;
1154 	child_proc->p_threadstart = parent_proc->p_threadstart;
1155 	child_proc->p_pthsize = parent_proc->p_pthsize;
1156 	if ((parent_proc->p_lflag & P_LREGISTER) != 0) {
1157 		child_proc->p_lflag |= P_LREGISTER;
1158 	}
1159 	child_proc->p_dispatchqueue_offset = parent_proc->p_dispatchqueue_offset;
1160 	child_proc->p_dispatchqueue_serialno_offset = parent_proc->p_dispatchqueue_serialno_offset;
1161 	child_proc->p_dispatchqueue_label_offset = parent_proc->p_dispatchqueue_label_offset;
1162 	child_proc->p_return_to_kernel_offset = parent_proc->p_return_to_kernel_offset;
1163 	child_proc->p_mach_thread_self_offset = parent_proc->p_mach_thread_self_offset;
1164 	child_proc->p_pth_tsd_offset = parent_proc->p_pth_tsd_offset;
1165 	child_proc->p_pthread_wq_quantum_offset = parent_proc->p_pthread_wq_quantum_offset;
1166 #if PSYNCH
1167 	pth_proc_hashinit(child_proc);
1168 #endif /* PSYNCH */
1169 
1170 #if CONFIG_PERSONAS
1171 	child_proc->p_persona = NULL;
1172 	error = persona_proc_inherit(child_proc, parent_proc);
1173 	if (error != 0) {
1174 		printf("forkproc: persona_proc_inherit failed (persona %d being destroyed?)\n", persona_id_from_proc(parent_proc));
1175 		forkproc_free(child_proc);
1176 		child_proc = NULL;
1177 		goto bad;
1178 	}
1179 #endif
1180 
1181 #if CONFIG_MEMORYSTATUS
1182 	/* Memorystatus init */
1183 	child_proc->p_memstat_state = 0;
1184 	child_proc->p_memstat_effectivepriority = JETSAM_PRIORITY_DEFAULT;
1185 	child_proc->p_memstat_requestedpriority = JETSAM_PRIORITY_DEFAULT;
1186 	child_proc->p_memstat_assertionpriority = 0;
1187 	child_proc->p_memstat_userdata          = 0;
1188 	child_proc->p_memstat_idle_start        = 0;
1189 	child_proc->p_memstat_idle_delta        = 0;
1190 	child_proc->p_memstat_memlimit          = 0;
1191 	child_proc->p_memstat_memlimit_active   = 0;
1192 	child_proc->p_memstat_memlimit_inactive = 0;
1193 	child_proc->p_memstat_relaunch_flags    = P_MEMSTAT_RELAUNCH_UNKNOWN;
1194 #if CONFIG_FREEZE
1195 	child_proc->p_memstat_freeze_sharedanon_pages = 0;
1196 #endif
1197 	child_proc->p_memstat_dirty = 0;
1198 	child_proc->p_memstat_idledeadline = 0;
1199 #endif /* CONFIG_MEMORYSTATUS */
1200 
1201 	if (parent_proc->p_subsystem_root_path) {
1202 		size_t parent_length = strlen(parent_proc->p_subsystem_root_path) + 1;
1203 		assert(parent_length <= MAXPATHLEN);
1204 		child_proc->p_subsystem_root_path = zalloc_flags(ZV_NAMEI,
1205 		    Z_WAITOK | Z_ZERO);
1206 		memcpy(child_proc->p_subsystem_root_path, parent_proc->p_subsystem_root_path, parent_length);
1207 	}
1208 
1209 bad:
1210 	return child_proc;
1211 }
1212 
1213 void
proc_lock(proc_t p)1214 proc_lock(proc_t p)
1215 {
1216 	LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_NOTOWNED);
1217 	lck_mtx_lock(&p->p_mlock);
1218 }
1219 
1220 void
proc_unlock(proc_t p)1221 proc_unlock(proc_t p)
1222 {
1223 	lck_mtx_unlock(&p->p_mlock);
1224 }
1225 
1226 void
proc_spinlock(proc_t p)1227 proc_spinlock(proc_t p)
1228 {
1229 	lck_spin_lock_grp(&p->p_slock, &proc_slock_grp);
1230 }
1231 
1232 void
proc_spinunlock(proc_t p)1233 proc_spinunlock(proc_t p)
1234 {
1235 	lck_spin_unlock(&p->p_slock);
1236 }
1237 
1238 void
proc_list_lock(void)1239 proc_list_lock(void)
1240 {
1241 	lck_mtx_lock(&proc_list_mlock);
1242 }
1243 
1244 void
proc_list_unlock(void)1245 proc_list_unlock(void)
1246 {
1247 	lck_mtx_unlock(&proc_list_mlock);
1248 }
1249 
1250 void
proc_ucred_lock(proc_t p)1251 proc_ucred_lock(proc_t p)
1252 {
1253 	lck_mtx_lock(&p->p_ucred_mlock);
1254 }
1255 
1256 void
proc_ucred_unlock(proc_t p)1257 proc_ucred_unlock(proc_t p)
1258 {
1259 	lck_mtx_unlock(&p->p_ucred_mlock);
1260 }
1261 
1262 void
proc_update_creds_onproc(proc_t p)1263 proc_update_creds_onproc(proc_t p)
1264 {
1265 	kauth_cred_t cred = proc_ucred(p);
1266 
1267 	p->p_uid = kauth_cred_getuid(cred);
1268 	p->p_gid = kauth_cred_getgid(cred);
1269 	p->p_ruid = kauth_cred_getruid(cred);
1270 	p->p_rgid = kauth_cred_getrgid(cred);
1271 	p->p_svuid = kauth_cred_getsvuid(cred);
1272 	p->p_svgid = kauth_cred_getsvgid(cred);
1273 }
1274 
1275 
1276 bool
uthread_is64bit(struct uthread * uth)1277 uthread_is64bit(struct uthread *uth)
1278 {
1279 	return uth->uu_flag & UT_LP64;
1280 }
1281 
1282 void
uthread_init(task_t task,uthread_t uth,thread_ro_t tro_tpl,int workq_thread)1283 uthread_init(task_t task, uthread_t uth, thread_ro_t tro_tpl, int workq_thread)
1284 {
1285 	uthread_t uth_parent = current_uthread();
1286 
1287 	lck_spin_init(&uth->uu_rethrottle_lock, &rethrottle_lock_grp,
1288 	    LCK_ATTR_NULL);
1289 
1290 	/*
1291 	 * Lazily set the thread on the kernel VFS context
1292 	 * to the first thread made which will be vm_pageout_scan_thread.
1293 	 */
1294 	if (__improbable(vfs_context0.vc_thread == NULL)) {
1295 		extern thread_t vm_pageout_scan_thread;
1296 
1297 		assert(task == kernel_task);
1298 		assert(get_machthread(uth) == vm_pageout_scan_thread);
1299 		vfs_context0.vc_thread = get_machthread(uth);
1300 	}
1301 
1302 	if (task_get_64bit_addr(task)) {
1303 		uth->uu_flag |= UT_LP64;
1304 	}
1305 
1306 	/*
1307 	 * Thread inherits credential from the creating thread, if both
1308 	 * are in the same task.
1309 	 *
1310 	 * If the creating thread has no credential or is from another
1311 	 * task we can leave the new thread credential NULL.  If it needs
1312 	 * one later, it will be lazily assigned from the task's process.
1313 	 */
1314 	if (task == kernel_task) {
1315 		kauth_cred_set(&tro_tpl->tro_cred, vfs_context0.vc_ucred);
1316 		tro_tpl->tro_proc = kernproc;
1317 		tro_tpl->tro_proc_ro = kernproc->p_proc_ro;
1318 	} else if (!task_is_a_corpse(task)) {
1319 		thread_ro_t curtro = current_thread_ro();
1320 		proc_t p = get_bsdtask_info(task);
1321 
1322 		if (task == curtro->tro_task &&
1323 		    ((curtro->tro_flags & TRO_SETUID) == 0 || !workq_thread)) {
1324 			kauth_cred_set(&tro_tpl->tro_cred, curtro->tro_cred);
1325 			tro_tpl->tro_flags = (curtro->tro_flags & TRO_SETUID);
1326 			tro_tpl->tro_proc_ro = curtro->tro_proc_ro;
1327 		} else {
1328 			kauth_cred_t cred = kauth_cred_proc_ref(p);
1329 			kauth_cred_set_and_unref(&tro_tpl->tro_cred, &cred);
1330 			tro_tpl->tro_proc_ro = task_get_ro(task);
1331 		}
1332 		tro_tpl->tro_proc = p;
1333 
1334 		proc_lock(p);
1335 		if (workq_thread) {
1336 			/* workq_thread threads will not inherit masks */
1337 			uth->uu_sigmask = ~workq_threadmask;
1338 		} else if (uth_parent->uu_flag & UT_SAS_OLDMASK) {
1339 			uth->uu_sigmask = uth_parent->uu_oldmask;
1340 		} else {
1341 			uth->uu_sigmask = uth_parent->uu_sigmask;
1342 		}
1343 
1344 		TAILQ_INSERT_TAIL(&p->p_uthlist, uth, uu_list);
1345 		proc_unlock(p);
1346 
1347 #if CONFIG_DTRACE
1348 		if (p->p_dtrace_ptss_pages != NULL) {
1349 			uth->t_dtrace_scratch = dtrace_ptss_claim_entry(p);
1350 		}
1351 #endif
1352 	} else {
1353 		tro_tpl->tro_proc_ro = task_get_ro(task);
1354 	}
1355 
1356 	uth->uu_pending_sigreturn = 0;
1357 	uthread_init_proc_refcount(uth);
1358 }
1359 
1360 mach_port_name_t
uthread_joiner_port(struct uthread * uth)1361 uthread_joiner_port(struct uthread *uth)
1362 {
1363 	return uth->uu_save.uus_bsdthread_terminate.kport;
1364 }
1365 
1366 user_addr_t
uthread_joiner_address(uthread_t uth)1367 uthread_joiner_address(uthread_t uth)
1368 {
1369 	return uth->uu_save.uus_bsdthread_terminate.ulock_addr;
1370 }
1371 
1372 void
uthread_joiner_wake(task_t task,uthread_t uth)1373 uthread_joiner_wake(task_t task, uthread_t uth)
1374 {
1375 	struct _bsdthread_terminate bts = uth->uu_save.uus_bsdthread_terminate;
1376 
1377 	assert(bts.ulock_addr);
1378 	bzero(&uth->uu_save.uus_bsdthread_terminate, sizeof(bts));
1379 
1380 	int flags = UL_UNFAIR_LOCK | ULF_WAKE_ALL | ULF_WAKE_ALLOW_NON_OWNER;
1381 	(void)ulock_wake(task, flags, bts.ulock_addr, 0);
1382 	mach_port_deallocate(get_task_ipcspace(task), bts.kport);
1383 }
1384 
1385 /*
1386  * This routine frees the thread name field of the uthread_t structure. Split out of
1387  * uthread_cleanup() so thread name does not get deallocated while generating a corpse fork.
1388  */
1389 void
uthread_cleanup_name(uthread_t uth)1390 uthread_cleanup_name(uthread_t uth)
1391 {
1392 	/*
1393 	 * <rdar://17834538>
1394 	 * Set pth_name to NULL before calling free().
1395 	 * Previously there was a race condition in the
1396 	 * case this code was executing during a stackshot
1397 	 * where the stackshot could try and copy pth_name
1398 	 * after it had been freed and before if was marked
1399 	 * as null.
1400 	 */
1401 	if (uth->pth_name != NULL) {
1402 		void *pth_name = uth->pth_name;
1403 		uth->pth_name = NULL;
1404 		kfree_data(pth_name, MAXTHREADNAMESIZE);
1405 	}
1406 	return;
1407 }
1408 
1409 /*
1410  * This routine frees all the BSD context in uthread except the credential.
1411  * It does not free the uthread structure as well
1412  */
1413 void
uthread_cleanup(uthread_t uth,thread_ro_t tro)1414 uthread_cleanup(uthread_t uth, thread_ro_t tro)
1415 {
1416 	task_t task = tro->tro_task;
1417 	proc_t p    = tro->tro_proc;
1418 
1419 	uthread_assert_zero_proc_refcount(uth);
1420 
1421 	if (uth->uu_lowpri_window || uth->uu_throttle_info) {
1422 		/*
1423 		 * task is marked as a low priority I/O type
1424 		 * and we've somehow managed to not dismiss the throttle
1425 		 * through the normal exit paths back to user space...
1426 		 * no need to throttle this thread since its going away
1427 		 * but we do need to update our bookeeping w/r to throttled threads
1428 		 *
1429 		 * Calling this routine will clean up any throttle info reference
1430 		 * still inuse by the thread.
1431 		 */
1432 		throttle_lowpri_io(0);
1433 	}
1434 
1435 #if CONFIG_AUDIT
1436 	/*
1437 	 * Per-thread audit state should never last beyond system
1438 	 * call return.  Since we don't audit the thread creation/
1439 	 * removal, the thread state pointer should never be
1440 	 * non-NULL when we get here.
1441 	 */
1442 	assert(uth->uu_ar == NULL);
1443 #endif
1444 
1445 	if (uth->uu_select.nbytes) {
1446 		select_cleanup_uthread(&uth->uu_select);
1447 	}
1448 
1449 	if (uth->uu_cdir) {
1450 		vnode_rele(uth->uu_cdir);
1451 		uth->uu_cdir = NULLVP;
1452 	}
1453 
1454 	if (uth->uu_selset) {
1455 		select_set_free(uth->uu_selset);
1456 		uth->uu_selset = NULL;
1457 	}
1458 
1459 	os_reason_free(uth->uu_exit_reason);
1460 
1461 	if ((task != kernel_task) && p) {
1462 		/*
1463 		 * Remove the thread from the process list and
1464 		 * transfer [appropriate] pending signals to the process.
1465 		 * Do not remove the uthread from proc uthlist for exec
1466 		 * copy task, since they does not have a ref on proc and
1467 		 * would not have been added to the list.
1468 		 */
1469 		if (uth->uu_kqr_bound) {
1470 			kqueue_threadreq_unbind(p, uth->uu_kqr_bound);
1471 		}
1472 
1473 		if (get_bsdtask_info(task) == p) {
1474 			proc_lock(p);
1475 			TAILQ_REMOVE(&p->p_uthlist, uth, uu_list);
1476 			p->p_siglist |= (uth->uu_siglist & execmask & (~p->p_sigignore | sigcantmask));
1477 			proc_unlock(p);
1478 		}
1479 
1480 #if CONFIG_DTRACE
1481 		struct dtrace_ptss_page_entry *tmpptr = uth->t_dtrace_scratch;
1482 		uth->t_dtrace_scratch = NULL;
1483 		if (tmpptr != NULL) {
1484 			dtrace_ptss_release_entry(p, tmpptr);
1485 		}
1486 #endif
1487 	} else {
1488 		assert(!uth->uu_kqr_bound);
1489 	}
1490 }
1491 
1492 /* This routine releases the credential stored in uthread */
1493 void
uthread_cred_ref(struct ucred * ucred)1494 uthread_cred_ref(struct ucred *ucred)
1495 {
1496 	kauth_cred_ref(ucred);
1497 }
1498 
1499 void
uthread_cred_free(struct ucred * ucred)1500 uthread_cred_free(struct ucred *ucred)
1501 {
1502 	kauth_cred_set(&ucred, NOCRED);
1503 }
1504 
1505 /* This routine frees the uthread structure held in thread structure */
1506 void
uthread_destroy(uthread_t uth)1507 uthread_destroy(uthread_t uth)
1508 {
1509 	uthread_destroy_proc_refcount(uth);
1510 
1511 	if (uth->t_tombstone) {
1512 		kfree_type(struct doc_tombstone, uth->t_tombstone);
1513 		uth->t_tombstone = NULL;
1514 	}
1515 
1516 #if CONFIG_DEBUG_SYSCALL_REJECTION
1517 	size_t const bitstr_len = BITMAP_SIZE(mach_trap_count + nsysent);
1518 
1519 	if (uth->syscall_rejection_mask) {
1520 		kfree_data(uth->syscall_rejection_mask, bitstr_len);
1521 		uth->syscall_rejection_mask = NULL;
1522 	}
1523 
1524 	if (uth->syscall_rejection_once_mask) {
1525 		kfree_data(uth->syscall_rejection_once_mask, bitstr_len);
1526 		uth->syscall_rejection_once_mask = NULL;
1527 	}
1528 #endif /* CONFIG_DEBUG_SYSCALL_REJECTION */
1529 
1530 	lck_spin_destroy(&uth->uu_rethrottle_lock, &rethrottle_lock_grp);
1531 
1532 	uthread_cleanup_name(uth);
1533 }
1534 
1535 user_addr_t
thread_get_sigreturn_token(thread_t thread)1536 thread_get_sigreturn_token(thread_t thread)
1537 {
1538 	uthread_t ut = (struct uthread *) get_bsdthread_info(thread);
1539 	return ut->uu_sigreturn_token;
1540 }
1541 
1542 uint32_t
thread_get_sigreturn_diversifier(thread_t thread)1543 thread_get_sigreturn_diversifier(thread_t thread)
1544 {
1545 	uthread_t ut = (struct uthread *) get_bsdthread_info(thread);
1546 	return ut->uu_sigreturn_diversifier;
1547 }
1548