xref: /xnu-10002.61.3/bsd/kern/kern_fork.c (revision 0f4c859e951fba394238ab619495c4e1d54d0f34)
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 extern void act_thread_catt(void *ctx);
153 void thread_set_child(thread_t child, int pid);
154 boolean_t thread_is_active(thread_t thread);
155 void *act_thread_csave(void);
156 extern boolean_t task_is_exec_copy(task_t);
157 int nextpidversion = 0;
158 
159 void ipc_task_enable(task_t task);
160 
161 proc_t forkproc(proc_t, cloneproc_flags_t);
162 void forkproc_free(proc_t);
163 thread_t fork_create_child(task_t parent_task,
164     coalition_t *parent_coalitions,
165     proc_t child,
166     int is_64bit_addr,
167     int is_64bit_data,
168     cloneproc_flags_t clone_flags);
169 
170 __private_extern__ const size_t uthread_size = sizeof(struct uthread);
171 static LCK_GRP_DECLARE(rethrottle_lock_grp, "rethrottle");
172 
173 os_refgrp_decl(, p_refgrp, "proc", NULL);
174 
175 extern const size_t task_alignment;
176 const size_t proc_alignment = _Alignof(struct proc);
177 
178 extern size_t task_struct_size;
179 size_t proc_struct_size = sizeof(struct proc);
180 size_t proc_and_task_size;
181 
182 ZONE_DECLARE_ID(ZONE_ID_PROC_TASK, struct proc);
183 SECURITY_READ_ONLY_LATE(zone_t) proc_task_zone;
184 
185 KALLOC_TYPE_DEFINE(proc_stats_zone, struct pstats, KT_DEFAULT);
186 
187 /*
188  * fork1
189  *
190  * Description:	common code used by all new process creation other than the
191  *		bootstrap of the initial process on the system
192  *
193  * Parameters: parent_proc		parent process of the process being
194  *		child_threadp		pointer to location to receive the
195  *					Mach thread_t of the child process
196  *					created
197  *		kind			kind of creation being requested
198  *		coalitions		if spawn, the set of coalitions the
199  *					child process should join, or NULL to
200  *					inherit the parent's. On non-spawns,
201  *					this param is ignored and the child
202  *					always inherits the parent's
203  *					coalitions.
204  *
205  * Notes:	Permissable values for 'kind':
206  *
207  *		PROC_CREATE_FORK	Create a complete process which will
208  *					return actively running in both the
209  *					parent and the child; the child copies
210  *					the parent address space.
211  *		PROC_CREATE_SPAWN	Create a complete process which will
212  *					return actively running in the parent
213  *					only after returning actively running
214  *					in the child; the child address space
215  *					is newly created by an image activator,
216  *					after which the child is run.
217  *
218  *		At first it may seem strange that we return the child thread
219  *		address rather than process structure, since the process is
220  *		the only part guaranteed to be "new"; however, since we do
221  *		not actualy adjust other references between Mach and BSD, this
222  *		is the only method which guarantees us the ability to get
223  *		back to the other information.
224  */
225 int
fork1(proc_t parent_proc,thread_t * child_threadp,int kind,coalition_t * coalitions)226 fork1(proc_t parent_proc, thread_t *child_threadp, int kind, coalition_t *coalitions)
227 {
228 	proc_t child_proc = NULL;       /* set in switch, but compiler... */
229 	thread_t child_thread = NULL;
230 	uid_t uid;
231 	size_t count;
232 	int err = 0;
233 	int spawn = 0;
234 	rlim_t rlimit_nproc_cur;
235 
236 	/*
237 	 * Although process entries are dynamically created, we still keep
238 	 * a global limit on the maximum number we will create.  Don't allow
239 	 * a nonprivileged user to use the last process; don't let root
240 	 * exceed the limit. The variable nprocs is the current number of
241 	 * processes, maxproc is the limit.
242 	 */
243 	uid = kauth_getruid();
244 	proc_list_lock();
245 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
246 #if (DEVELOPMENT || DEBUG) && !defined(XNU_TARGET_OS_OSX)
247 		/*
248 		 * On the development kernel, panic so that the fact that we hit
249 		 * the process limit is obvious, as this may very well wedge the
250 		 * system.
251 		 */
252 		panic("The process table is full; parent pid=%d", proc_getpid(parent_proc));
253 #endif
254 		proc_list_unlock();
255 		tablefull("proc");
256 		return EAGAIN;
257 	}
258 	proc_list_unlock();
259 
260 	/*
261 	 * Increment the count of procs running with this uid. Don't allow
262 	 * a nonprivileged user to exceed their current limit, which is
263 	 * always less than what an rlim_t can hold.
264 	 * (locking protection is provided by list lock held in chgproccnt)
265 	 */
266 	count = chgproccnt(uid, 1);
267 	rlimit_nproc_cur = proc_limitgetcur(parent_proc, RLIMIT_NPROC);
268 	if (uid != 0 &&
269 	    (rlim_t)count > rlimit_nproc_cur) {
270 #if (DEVELOPMENT || DEBUG) && !defined(XNU_TARGET_OS_OSX)
271 		/*
272 		 * On the development kernel, panic so that the fact that we hit
273 		 * the per user process limit is obvious.  This may be less dire
274 		 * than hitting the global process limit, but we cannot rely on
275 		 * that.
276 		 */
277 		panic("The per-user process limit has been hit; parent pid=%d, uid=%d", proc_getpid(parent_proc), uid);
278 #endif
279 		err = EAGAIN;
280 		goto bad;
281 	}
282 
283 #if CONFIG_MACF
284 	/*
285 	 * Determine if MAC policies applied to the process will allow
286 	 * it to fork.  This is an advisory-only check.
287 	 */
288 	err = mac_proc_check_fork(parent_proc);
289 	if (err != 0) {
290 		goto bad;
291 	}
292 #endif
293 
294 	switch (kind) {
295 	case PROC_CREATE_SPAWN:
296 		/*
297 		 * A spawned process differs from a forked process in that
298 		 * the spawned process does not carry around the parents
299 		 * baggage with regard to address space copying, dtrace,
300 		 * and so on.
301 		 */
302 		spawn = 1;
303 
304 		OS_FALLTHROUGH;
305 
306 	case PROC_CREATE_FORK:
307 		/*
308 		 * When we clone the parent process, we are going to inherit
309 		 * its task attributes and memory, since when we fork, we
310 		 * will, in effect, create a duplicate of it, with only minor
311 		 * differences.  Contrarily, spawned processes do not inherit.
312 		 */
313 		if ((child_thread = cloneproc(proc_task(parent_proc),
314 		    spawn ? coalitions : NULL,
315 		    parent_proc,
316 		    spawn ? CLONEPROC_FLAGS_NONE : CLONEPROC_FLAGS_INHERIT_MEMORY)) == NULL) {
317 			/* Failed to create thread */
318 			err = EAGAIN;
319 			goto bad;
320 		}
321 
322 		/* child_proc = child_thread->task->proc; */
323 		child_proc = (proc_t)(get_bsdtask_info(get_threadtask(child_thread)));
324 
325 		if (!spawn) {
326 			/* Copy current thread state into the child thread (only for fork) */
327 			thread_dup(child_thread);
328 		}
329 
330 // XXX BEGIN: wants to move to be common code (and safe)
331 #if CONFIG_MACF
332 		/*
333 		 * allow policies to associate the credential/label that
334 		 * we referenced from the parent ... with the child
335 		 * JMM - this really isn't safe, as we can drop that
336 		 *       association without informing the policy in other
337 		 *       situations (keep long enough to get policies changed)
338 		 */
339 		mac_cred_label_associate_fork(proc_ucred_unsafe(child_proc),
340 		    child_proc);
341 #endif
342 
343 		/*
344 		 * Propogate change of PID - may get new cred if auditing.
345 		 */
346 		set_security_token(child_proc, proc_ucred_unsafe(child_proc));
347 
348 		AUDIT_ARG(pid, proc_getpid(child_proc));
349 
350 // XXX END: wants to move to be common code (and safe)
351 
352 		/*
353 		 * Blow thread state information; this is what gives the child
354 		 * process its "return" value from a fork() call.
355 		 *
356 		 * Note: this should probably move to fork() proper, since it
357 		 * is not relevent to spawn, and the value won't matter
358 		 * until we resume the child there.  If you are in here
359 		 * refactoring code, consider doing this at the same time.
360 		 */
361 		thread_set_child(child_thread, proc_getpid(child_proc));
362 
363 		child_proc->p_acflag = AFORK;   /* forked but not exec'ed */
364 
365 #if CONFIG_DTRACE
366 		dtrace_proc_fork(parent_proc, child_proc, spawn);
367 #endif  /* CONFIG_DTRACE */
368 		if (!spawn) {
369 			/*
370 			 * Of note, we need to initialize the bank context behind
371 			 * the protection of the proc_trans lock to prevent a race with exit.
372 			 */
373 			task_bank_init(get_threadtask(child_thread));
374 		}
375 
376 		break;
377 
378 	default:
379 		panic("fork1 called with unknown kind %d", kind);
380 		break;
381 	}
382 
383 
384 	/* return the thread pointer to the caller */
385 	*child_threadp = child_thread;
386 
387 bad:
388 	/*
389 	 * In the error case, we return a 0 value for the returned pid (but
390 	 * it is ignored in the trampoline due to the error return); this
391 	 * is probably not necessary.
392 	 */
393 	if (err) {
394 		(void)chgproccnt(uid, -1);
395 	}
396 
397 	return err;
398 }
399 
400 
401 
402 
403 /*
404  * fork_create_child
405  *
406  * Description:	Common operations associated with the creation of a child
407  *		process. Return with new task and first thread's control port movable
408  *      and not pinned.
409  *
410  * Parameters:	parent_task		parent task
411  *		parent_coalitions	parent's set of coalitions
412  *		child_proc			child process
413  *		inherit_memory		TRUE, if the parents address space is
414  *							to be inherited by the child
415  *		is_64bit_addr		TRUE, if the child being created will
416  *							be associated with a 64 bit address space
417  *		is_64bit_data		TRUE if the child being created will use a
418  *                                                       64-bit register state
419  *		in_exec				TRUE, if called from execve or posix spawn set exec
420  *							FALSE, if called from fork or vfexec
421  *
422  * Note:	This code is called in the fork() case, from the execve() call
423  *		graph, from the posix_spawn() call graph (which implicitly
424  *		includes a vfork() equivalent call, and in the system
425  *		bootstrap case.
426  *
427  *		It creates a new task and thread (and as a side effect of the
428  *		thread creation, a uthread) in the parent coalition set, which is
429  *		then associated with the process 'child'.  If the parent
430  *		process address space is to be inherited, then a flag
431  *		indicates that the newly created task should inherit this from
432  *		the child task.
433  *
434  *		As a special concession to bootstrapping the initial process
435  *		in the system, it's possible for 'parent_task' to be TASK_NULL;
436  *		in this case, 'inherit_memory' MUST be FALSE.
437  */
438 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)439 fork_create_child(task_t parent_task,
440     coalition_t *parent_coalitions,
441     proc_t child_proc,
442     int is_64bit_addr,
443     int is_64bit_data,
444     cloneproc_flags_t clone_flags)
445 {
446 	thread_t        child_thread = NULL;
447 	task_t          child_task;
448 	kern_return_t   result;
449 	proc_ro_t       proc_ro;
450 	bool inherit_memory = !!(clone_flags & CLONEPROC_FLAGS_INHERIT_MEMORY);
451 	bool in_exec = !!(clone_flags & CLONEPROC_FLAGS_FOR_EXEC);
452 	/*
453 	 * Exec complete hook should be called for spawn and exec, but not for fork.
454 	 */
455 	uint8_t returnwaitflags = (!inherit_memory ? TRW_LEXEC_COMPLETE : 0) |
456 	    (TRW_LRETURNWAIT | TRW_LRETURNWAITER);
457 
458 	proc_ro = proc_get_ro(child_proc);
459 	if (proc_ro_task(proc_ro) != NULL) {
460 		panic("Proc_ro_task for newly created proc %p is not NULL", child_proc);
461 	}
462 
463 	child_task = proc_get_task_raw(child_proc);
464 
465 	/*
466 	 * Create a new task for the child process, IPC access to the new task will
467 	 * be set up after task has been fully initialized.
468 	 */
469 	result = task_create_internal(parent_task,
470 	    proc_ro,
471 	    parent_coalitions,
472 	    inherit_memory,
473 	    is_64bit_addr,
474 	    is_64bit_data,
475 	    TF_NONE,
476 	    TF_NONE,
477 	    in_exec ? TPF_EXEC_COPY : TPF_NONE,           /* Mark the task exec copy if in execve */
478 	    returnwaitflags,                              /* All created threads will wait in task_wait_to_return */
479 	    child_task);
480 	if (result != KERN_SUCCESS) {
481 		printf("%s: task_create_internal failed.  Code: %d\n",
482 		    __func__, result);
483 		goto bad;
484 	}
485 
486 	/* Set the child proc process to child task */
487 	proc_set_task(child_proc, child_task);
488 
489 	/* Set child task process to child proc */
490 	set_bsdtask_info(child_task, child_proc);
491 
492 	/* Propagate CPU limit timer from parent */
493 	if (timerisset(&child_proc->p_rlim_cpu)) {
494 		task_vtimer_set(child_task, TASK_VTIMER_RLIM);
495 	}
496 
497 	/*
498 	 * Set child process BSD visible scheduler priority if nice value
499 	 * inherited from parent
500 	 */
501 	if (child_proc->p_nice != 0) {
502 		resetpriority(child_proc);
503 	}
504 
505 	/*
506 	 * Create main thread for the child process. Its control port is not immovable/pinned
507 	 * until main_thread_set_immovable_pinned().
508 	 *
509 	 * The new thread is waiting on the event triggered by 'task_clear_return_wait'
510 	 */
511 	result = main_thread_create_waiting(child_task,
512 	    (thread_continue_t)task_wait_to_return,
513 	    task_get_return_wait_event(child_task),
514 	    &child_thread);
515 
516 	if (result != KERN_SUCCESS) {
517 		printf("%s: thread_create failed. Code: %d\n",
518 		    __func__, result);
519 		task_deallocate(child_task);
520 		child_task = NULL;
521 	}
522 
523 	/*
524 	 * Tag thread as being the first thread in its task.
525 	 */
526 	thread_set_tag(child_thread, THREAD_TAG_MAINTHREAD);
527 
528 bad:
529 	thread_yield_internal(1);
530 
531 	return child_thread;
532 }
533 
534 
535 /*
536  * fork
537  *
538  * Description:	fork system call.
539  *
540  * Parameters:	parent			Parent process to fork
541  *		uap (void)		[unused]
542  *		retval			Return value
543  *
544  * Returns:	0			Success
545  *		EAGAIN			Resource unavailable, try again
546  *
547  * Notes:	Attempts to create a new child process which inherits state
548  *		from the parent process.  If successful, the call returns
549  *		having created an initially suspended child process with an
550  *		extra Mach task and thread reference, for which the thread
551  *		is initially suspended.  Until we resume the child process,
552  *		it is not yet running.
553  *
554  *		The return information to the child is contained in the
555  *		thread state structure of the new child, and does not
556  *		become visible to the child through a normal return process,
557  *		since it never made the call into the kernel itself in the
558  *		first place.
559  *
560  *		After resuming the thread, this function returns directly to
561  *		the parent process which invoked the fork() system call.
562  *
563  * Important:	The child thread_resume occurs before the parent returns;
564  *		depending on scheduling latency, this means that it is not
565  *		deterministic as to whether the parent or child is scheduled
566  *		to run first.  It is entirely possible that the child could
567  *		run to completion prior to the parent running.
568  */
569 int
fork(proc_t parent_proc,__unused struct fork_args * uap,int32_t * retval)570 fork(proc_t parent_proc, __unused struct fork_args *uap, int32_t *retval)
571 {
572 	thread_t child_thread;
573 	int err;
574 
575 	retval[1] = 0;          /* flag parent return for user space */
576 
577 	if ((err = fork1(parent_proc, &child_thread, PROC_CREATE_FORK, NULL)) == 0) {
578 		task_t child_task;
579 		proc_t child_proc;
580 
581 		/* Return to the parent */
582 		child_proc = (proc_t)get_bsdthreadtask_info(child_thread);
583 		retval[0] = proc_getpid(child_proc);
584 
585 		child_task = (task_t)get_threadtask(child_thread);
586 		assert(child_task != TASK_NULL);
587 
588 		/* task_control_port_options has been inherited from parent, apply it */
589 		task_set_immovable_pinned(child_task);
590 		main_thread_set_immovable_pinned(child_thread);
591 
592 		/*
593 		 * Since the task ports for this new task are now set to be immovable,
594 		 * we can enable them.
595 		 */
596 		ipc_task_enable(get_threadtask(child_thread));
597 
598 		/*
599 		 * Drop the signal lock on the child which was taken on our
600 		 * behalf by forkproc()/cloneproc() to prevent signals being
601 		 * received by the child in a partially constructed state.
602 		 */
603 		proc_signalend(child_proc, 0);
604 		proc_transend(child_proc, 0);
605 
606 		/* flag the fork has occurred */
607 		proc_knote(parent_proc, NOTE_FORK | proc_getpid(child_proc));
608 		DTRACE_PROC1(create, proc_t, child_proc);
609 
610 #if CONFIG_DTRACE
611 		if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) {
612 			(*dtrace_proc_waitfor_hook)(child_proc);
613 		}
614 #endif
615 
616 		/*
617 		 * If current process died during the fork, the child would contain
618 		 * non consistent vmmap, kill the child and reap it internally.
619 		 */
620 		if (parent_proc->p_lflag & P_LEXIT || !thread_is_active(current_thread())) {
621 			task_terminate_internal(child_task);
622 			proc_list_lock();
623 			child_proc->p_listflag |= P_LIST_DEADPARENT;
624 			proc_list_unlock();
625 		}
626 
627 		/* "Return" to the child */
628 		task_clear_return_wait(get_threadtask(child_thread), TCRW_CLEAR_ALL_WAIT);
629 
630 		/* drop the extra references we got during the creation */
631 		task_deallocate(child_task);
632 		thread_deallocate(child_thread);
633 	}
634 
635 	return err;
636 }
637 
638 
639 /*
640  * cloneproc
641  *
642  * Description: Create a new process from a specified process.
643  *
644  * Parameters:	parent_task		The parent task to be cloned, or
645  *					TASK_NULL is task characteristics
646  *					are not to be inherited
647  *					be cloned, or TASK_NULL if the new
648  *					task is not to inherit the VM
649  *					characteristics of the parent
650  *		parent_proc		The parent process to be cloned
651  *		clone_flags		Clone flags to specify if the cloned
652  *					process should inherit memory,
653  *					marked as memory stat internal,
654  *					or if the cloneproc is called for exec.
655  *
656  * Returns:	!NULL			pointer to new child thread
657  *		NULL			Failure (unspecified)
658  *
659  * Note:	On return newly created child process has signal lock held
660  *		to block delivery of signal to it if called with lock set.
661  *		fork() code needs to explicity remove this lock before
662  *		signals can be delivered
663  *
664  *		In the case of bootstrap, this function can be called from
665  *		bsd_utaskbootstrap() in order to bootstrap the first process;
666  *		the net effect is to provide a uthread structure for the
667  *		kernel process associated with the kernel task.
668  *
669  * XXX:		Tristating using the value parent_task as the major key
670  *		and inherit_memory as the minor key is something we should
671  *		refactor later; we owe the current semantics, ultimately,
672  *		to the semantics of task_create_internal.  For now, we will
673  *		live with this being somewhat awkward.
674  */
675 thread_t
cloneproc(task_t parent_task,coalition_t * parent_coalitions,proc_t parent_proc,cloneproc_flags_t clone_flags)676 cloneproc(task_t parent_task, coalition_t *parent_coalitions, proc_t parent_proc, cloneproc_flags_t clone_flags)
677 {
678 #if !CONFIG_MEMORYSTATUS
679 #pragma unused(memstat_internal)
680 #endif
681 	task_t child_task;
682 	proc_t child_proc;
683 	thread_t child_thread = NULL;
684 	bool memstat_internal = !!(clone_flags & CLONEPROC_FLAGS_MEMSTAT_INTERNAL);
685 	bool in_exec = !!(clone_flags & CLONEPROC_FLAGS_FOR_EXEC);
686 
687 	if ((child_proc = forkproc(parent_proc, clone_flags)) == NULL) {
688 		/* Failed to allocate new process */
689 		goto bad;
690 	}
691 
692 	/*
693 	 * In the case where the parent_task is TASK_NULL (during the init path)
694 	 * we make the assumption that the register size will be the same as the
695 	 * address space size since there's no way to determine the possible
696 	 * register size until an image is exec'd.
697 	 *
698 	 * The only architecture that has different address space and register sizes
699 	 * (arm64_32) isn't being used within kernel-space, so the above assumption
700 	 * always holds true for the init path.
701 	 */
702 	const int parent_64bit_addr = parent_proc->p_flag & P_LP64;
703 	const int parent_64bit_data = (parent_task == TASK_NULL) ? parent_64bit_addr : task_get_64bit_data(parent_task);
704 
705 	child_thread = fork_create_child(parent_task,
706 	    parent_coalitions,
707 	    child_proc,
708 	    parent_64bit_addr,
709 	    parent_64bit_data,
710 	    clone_flags);
711 
712 	if (child_thread == NULL) {
713 		/*
714 		 * Failed to create thread; now we must deconstruct the new
715 		 * process previously obtained from forkproc().
716 		 */
717 		forkproc_free(child_proc);
718 		goto bad;
719 	}
720 
721 	child_task = get_threadtask(child_thread);
722 	if (parent_64bit_addr) {
723 		OSBitOrAtomic(P_LP64, (UInt32 *)&child_proc->p_flag);
724 		get_bsdthread_info(child_thread)->uu_flag |= UT_LP64;
725 	} else {
726 		OSBitAndAtomic(~((uint32_t)P_LP64), (UInt32 *)&child_proc->p_flag);
727 		get_bsdthread_info(child_thread)->uu_flag &= ~UT_LP64;
728 	}
729 
730 #if CONFIG_MEMORYSTATUS
731 	if (memstat_internal ||
732 	    (in_exec && (parent_proc->p_memstat_state & P_MEMSTAT_INTERNAL))) {
733 		proc_list_lock();
734 		child_proc->p_memstat_state |= P_MEMSTAT_INTERNAL;
735 		proc_list_unlock();
736 	}
737 	if (in_exec && parent_proc->p_memstat_relaunch_flags != P_MEMSTAT_RELAUNCH_UNKNOWN) {
738 		memorystatus_relaunch_flags_update(child_proc, parent_proc->p_memstat_relaunch_flags);
739 	}
740 #endif
741 
742 	/* make child visible */
743 	pinsertchild(parent_proc, child_proc, in_exec);
744 
745 	/*
746 	 * Make child runnable, set start time.
747 	 */
748 	child_proc->p_stat = SRUN;
749 bad:
750 	return child_thread;
751 }
752 
753 void
proc_set_sigact(proc_t p,int sig,user_addr_t sigact)754 proc_set_sigact(proc_t p, int sig, user_addr_t sigact)
755 {
756 	assert((sig > 0) && (sig < NSIG));
757 
758 	p->p_sigacts.ps_sigact[sig] = sigact;
759 }
760 
761 void
proc_set_trampact(proc_t p,int sig,user_addr_t trampact)762 proc_set_trampact(proc_t p, int sig, user_addr_t trampact)
763 {
764 	assert((sig > 0) && (sig < NSIG));
765 
766 	p->p_sigacts.ps_trampact[sig] = trampact;
767 }
768 
769 void
proc_set_sigact_trampact(proc_t p,int sig,user_addr_t sigact,user_addr_t trampact)770 proc_set_sigact_trampact(proc_t p, int sig, user_addr_t sigact, user_addr_t trampact)
771 {
772 	assert((sig > 0) && (sig < NSIG));
773 
774 	p->p_sigacts.ps_sigact[sig] = sigact;
775 	p->p_sigacts.ps_trampact[sig] = trampact;
776 }
777 
778 void
proc_reset_sigact(proc_t p,sigset_t sigs)779 proc_reset_sigact(proc_t p, sigset_t sigs)
780 {
781 	user_addr_t *sigacts = p->p_sigacts.ps_sigact;
782 	int nc;
783 
784 	while (sigs) {
785 		nc = ffs((unsigned int)sigs);
786 		if (sigacts[nc] != SIG_DFL) {
787 			sigacts[nc] = SIG_DFL;
788 		}
789 		sigs &= ~sigmask(nc);
790 	}
791 }
792 
793 /*
794  * Destroy a process structure that resulted from a call to forkproc(), but
795  * which must be returned to the system because of a subsequent failure
796  * preventing it from becoming active.
797  *
798  * Parameters:	p			The incomplete process from forkproc()
799  *
800  * Returns:	(void)
801  *
802  * Note:	This function should only be used in an error handler following
803  *		a call to forkproc().
804  *
805  *		Operations occur in reverse order of those in forkproc().
806  */
807 void
forkproc_free(proc_t p)808 forkproc_free(proc_t p)
809 {
810 	struct pgrp *pg;
811 
812 #if CONFIG_PERSONAS
813 	persona_proc_drop(p);
814 #endif /* CONFIG_PERSONAS */
815 
816 #if PSYNCH
817 	pth_proc_hashdelete(p);
818 #endif /* PSYNCH */
819 
820 	/* We held signal and a transition locks; drop them */
821 	proc_signalend(p, 0);
822 	proc_transend(p, 0);
823 
824 	/*
825 	 * If we have our own copy of the resource limits structure, we
826 	 * need to free it.  If it's a shared copy, we need to drop our
827 	 * reference on it.
828 	 */
829 	proc_limitdrop(p);
830 
831 #if SYSV_SHM
832 	/* Need to drop references to the shared memory segment(s), if any */
833 	if (p->vm_shm) {
834 		/*
835 		 * Use shmexec(): we have no address space, so no mappings
836 		 *
837 		 * XXX Yes, the routine is badly named.
838 		 */
839 		shmexec(p);
840 	}
841 #endif
842 
843 	/* Need to undo the effects of the fdt_fork(), if any */
844 	fdt_invalidate(p);
845 	fdt_destroy(p);
846 
847 	/*
848 	 * Drop the reference on a text vnode pointer, if any
849 	 * XXX This code is broken in forkproc(); see <rdar://4256419>;
850 	 * XXX if anyone ever uses this field, we will be extremely unhappy.
851 	 */
852 	if (p->p_textvp) {
853 		vnode_rele(p->p_textvp);
854 		p->p_textvp = NULL;
855 	}
856 
857 	/* Update the audit session proc count */
858 	AUDIT_SESSION_PROCEXIT(p);
859 
860 	lck_mtx_destroy(&p->p_mlock, &proc_mlock_grp);
861 	lck_mtx_destroy(&p->p_ucred_mlock, &proc_ucred_mlock_grp);
862 #if CONFIG_AUDIT
863 	lck_mtx_destroy(&p->p_audit_mlock, &proc_ucred_mlock_grp);
864 #endif /* CONFIG_AUDIT */
865 #if CONFIG_DTRACE
866 	lck_mtx_destroy(&p->p_dtrace_sprlock, &proc_lck_grp);
867 #endif
868 	lck_spin_destroy(&p->p_slock, &proc_slock_grp);
869 
870 	proc_list_lock();
871 	/* Decrement the count of processes in the system */
872 	nprocs--;
873 
874 	/* quit the group */
875 	pg = pgrp_leave_locked(p);
876 
877 	/* Take it out of process hash */
878 	assert((os_ref_get_raw_mask(&p->p_refcount) >> P_REF_BITS) == 1);
879 	assert((os_ref_get_raw_mask(&p->p_refcount) & P_REF_NEW) == P_REF_NEW);
880 	os_atomic_xor(&p->p_refcount, P_REF_NEW | P_REF_DEAD, relaxed);
881 
882 	/* Remove from hash if not a shadow proc */
883 	if (!proc_is_shadow(p)) {
884 		phash_remove_locked(p);
885 	}
886 
887 	proc_list_unlock();
888 
889 	pgrp_rele(pg);
890 
891 	thread_call_free(p->p_rcall);
892 
893 	/* Free allocated memory */
894 	zfree(proc_stats_zone, p->p_stats);
895 	p->p_stats = NULL;
896 	if (p->p_subsystem_root_path) {
897 		zfree(ZV_NAMEI, p->p_subsystem_root_path);
898 		p->p_subsystem_root_path = NULL;
899 	}
900 
901 	proc_checkdeadrefs(p);
902 	proc_wait_release(p);
903 }
904 
905 
906 /*
907  * forkproc
908  *
909  * Description:	Create a new process structure, given a parent process
910  *		structure.
911  *
912  * Parameters:	parent_proc		The parent process
913  *
914  * Returns:	!NULL			The new process structure
915  *		NULL			Error (insufficient free memory)
916  *
917  * Note:	When successful, the newly created process structure is
918  *		partially initialized; if a caller needs to deconstruct the
919  *		returned structure, they must call forkproc_free() to do so.
920  */
921 proc_t
forkproc(proc_t parent_proc,cloneproc_flags_t clone_flags)922 forkproc(proc_t parent_proc, cloneproc_flags_t clone_flags)
923 {
924 	static uint64_t nextuniqueid = 0;
925 	static pid_t lastpid = 0;
926 
927 	proc_t child_proc;      /* Our new process */
928 	int error = 0;
929 	struct pgrp *pg;
930 	uthread_t parent_uthread = current_uthread();
931 	rlim_t rlimit_cpu_cur;
932 	pid_t pid;
933 	struct proc_ro_data proc_ro_data = {};
934 	bool in_exec = !!(clone_flags & CLONEPROC_FLAGS_FOR_EXEC);
935 
936 	child_proc = zalloc_flags(proc_task_zone, Z_WAITOK | Z_ZERO);
937 
938 	child_proc->p_stats = zalloc_flags(proc_stats_zone, Z_WAITOK | Z_ZERO);
939 	child_proc->p_sigacts = parent_proc->p_sigacts;
940 	os_ref_init_mask(&child_proc->p_refcount, P_REF_BITS, &p_refgrp, P_REF_NEW);
941 	os_ref_init_raw(&child_proc->p_waitref, &p_refgrp);
942 	proc_ref_hold_proc_task_struct(child_proc);
943 
944 	/* allocate a callout for use by interval timers */
945 	child_proc->p_rcall = thread_call_allocate((thread_call_func_t)realitexpire, child_proc);
946 
947 
948 	/*
949 	 * Find an unused PID.
950 	 */
951 
952 	fdt_init(child_proc);
953 
954 	proc_list_lock();
955 
956 	if (!in_exec) {
957 		pid = lastpid;
958 		do {
959 			/*
960 			 * If the process ID prototype has wrapped around,
961 			 * restart somewhat above 0, as the low-numbered procs
962 			 * tend to include daemons that don't exit.
963 			 */
964 			if (++pid >= PID_MAX) {
965 				pid = 100;
966 			}
967 			if (pid == lastpid) {
968 				panic("Unable to allocate a new pid");
969 			}
970 
971 			/* if the pid stays in hash both for zombie and runniing state */
972 		} while (phash_find_locked(pid) != PROC_NULL ||
973 		    pghash_exists_locked(pid) ||
974 		    session_find_locked(pid) != SESSION_NULL);
975 
976 		lastpid = pid;
977 		nprocs++;
978 
979 		child_proc->p_pid = pid;
980 		proc_ro_data.p_idversion = OSIncrementAtomic(&nextpidversion);
981 		/* kernel process is handcrafted and not from fork, so start from 1 */
982 		proc_ro_data.p_uniqueid = ++nextuniqueid;
983 
984 		/* Insert in the hash, and inherit our group (and session) */
985 		phash_insert_locked(child_proc);
986 
987 		/* Check if the proc is from App Cryptex */
988 		if (parent_proc->p_ladvflag & P_RSR) {
989 			os_atomic_or(&child_proc->p_ladvflag, P_RSR, relaxed);
990 		}
991 	} else {
992 		/* For exec copy of the proc, copy the pid, pidversion and uniqueid of original proc */
993 		pid = parent_proc->p_pid;
994 		child_proc->p_pid = pid;
995 		proc_ro_data.p_idversion = parent_proc->p_proc_ro->p_idversion;
996 		proc_ro_data.p_uniqueid = parent_proc->p_proc_ro->p_uniqueid;
997 
998 		nprocs++;
999 		os_atomic_or(&child_proc->p_refcount, P_REF_SHADOW, relaxed);
1000 	}
1001 	pg = pgrp_enter_locked(parent_proc, child_proc);
1002 	proc_list_unlock();
1003 
1004 	if (proc_ro_data.p_uniqueid == startup_serial_num_procs) {
1005 		/*
1006 		 * Turn off startup serial logging now that we have reached
1007 		 * the defined number of startup processes.
1008 		 */
1009 		startup_serial_logging_active = false;
1010 	}
1011 
1012 	/*
1013 	 * We've identified the PID we are going to use;
1014 	 * initialize the new process structure.
1015 	 */
1016 	child_proc->p_stat = SIDL;
1017 
1018 	/*
1019 	 * The zero'ing of the proc was at the allocation time due to need
1020 	 * for insertion to hash.  Copy the section that is to be copied
1021 	 * directly from the parent.
1022 	 */
1023 	child_proc->p_forkcopy = parent_proc->p_forkcopy;
1024 
1025 	proc_ro_data.syscall_filter_mask = proc_syscall_filter_mask(parent_proc);
1026 	proc_ro_data.p_platform_data = proc_get_ro(parent_proc)->p_platform_data;
1027 
1028 	/*
1029 	 * Some flags are inherited from the parent.
1030 	 * Duplicate sub-structures as needed.
1031 	 * Increase reference counts on shared objects.
1032 	 * The p_stats substruct is set in vm_fork.
1033 	 */
1034 #if CONFIG_DELAY_IDLE_SLEEP
1035 	child_proc->p_flag = (parent_proc->p_flag & (P_LP64 | P_TRANSLATED | P_DISABLE_ASLR | P_DELAYIDLESLEEP | P_SUGID | P_AFFINITY));
1036 #else /* CONFIG_DELAY_IDLE_SLEEP */
1037 	child_proc->p_flag = (parent_proc->p_flag & (P_LP64 | P_TRANSLATED | P_DISABLE_ASLR | P_SUGID | P_AFFINITY));
1038 #endif /* CONFIG_DELAY_IDLE_SLEEP */
1039 
1040 	child_proc->p_vfs_iopolicy = (parent_proc->p_vfs_iopolicy & (P_VFS_IOPOLICY_INHERITED_MASK));
1041 
1042 	child_proc->p_responsible_pid = parent_proc->p_responsible_pid;
1043 
1044 	/*
1045 	 * Note that if the current thread has an assumed identity, this
1046 	 * credential will be granted to the new process.
1047 	 */
1048 	kauth_cred_set(&proc_ro_data.p_ucred.__smr_ptr, kauth_cred_get());
1049 
1050 	lck_mtx_init(&child_proc->p_mlock, &proc_mlock_grp, &proc_lck_attr);
1051 	lck_mtx_init(&child_proc->p_ucred_mlock, &proc_ucred_mlock_grp, &proc_lck_attr);
1052 #if CONFIG_AUDIT
1053 	lck_mtx_init(&child_proc->p_audit_mlock, &proc_ucred_mlock_grp, &proc_lck_attr);
1054 #endif /* CONFIG_AUDIT */
1055 #if CONFIG_DTRACE
1056 	lck_mtx_init(&child_proc->p_dtrace_sprlock, &proc_lck_grp, &proc_lck_attr);
1057 #endif
1058 	lck_spin_init(&child_proc->p_slock, &proc_slock_grp, &proc_lck_attr);
1059 
1060 	klist_init(&child_proc->p_klist);
1061 
1062 	if (child_proc->p_textvp != NULLVP) {
1063 		/* bump references to the text vnode */
1064 		/* Need to hold iocount across the ref call */
1065 		if ((error = vnode_getwithref(child_proc->p_textvp)) == 0) {
1066 			error = vnode_ref(child_proc->p_textvp);
1067 			vnode_put(child_proc->p_textvp);
1068 		}
1069 
1070 		if (error != 0) {
1071 			child_proc->p_textvp = NULLVP;
1072 		}
1073 	}
1074 
1075 	/* Inherit the parent flags for code sign */
1076 	proc_ro_data.p_csflags = ((uint32_t)proc_getcsflags(parent_proc) & ~CS_KILLED);
1077 
1078 	child_proc->p_proc_ro = proc_ro_alloc(child_proc, &proc_ro_data, NULL, NULL);
1079 
1080 	/* update cred on proc */
1081 	proc_update_creds_onproc(child_proc, proc_ucred_unsafe(child_proc));
1082 
1083 	/* update audit session proc count */
1084 	AUDIT_SESSION_PROCNEW(child_proc);
1085 
1086 	/*
1087 	 * Copy the parents per process open file table to the child; if
1088 	 * there is a per-thread current working directory, set the childs
1089 	 * per-process current working directory to that instead of the
1090 	 * parents.
1091 	 */
1092 	if (fdt_fork(&child_proc->p_fd, parent_proc, parent_uthread->uu_cdir, in_exec) != 0) {
1093 		forkproc_free(child_proc);
1094 		child_proc = NULL;
1095 		goto bad;
1096 	}
1097 
1098 #if SYSV_SHM
1099 	if (parent_proc->vm_shm && !in_exec) {
1100 		/* XXX may fail to attach shm to child */
1101 		(void)shmfork(parent_proc, child_proc);
1102 	}
1103 #endif
1104 
1105 	/*
1106 	 * Child inherits the parent's plimit
1107 	 */
1108 	proc_limitfork(parent_proc, child_proc);
1109 
1110 	rlimit_cpu_cur = proc_limitgetcur(child_proc, RLIMIT_CPU);
1111 	if (rlimit_cpu_cur != RLIM_INFINITY) {
1112 		child_proc->p_rlim_cpu.tv_sec = (rlimit_cpu_cur > __INT_MAX__) ? __INT_MAX__ : rlimit_cpu_cur;
1113 	}
1114 
1115 	if (in_exec) {
1116 		/* Keep the original start time for exec'ed proc */
1117 		child_proc->p_stats->ps_start = parent_proc->p_stats->ps_start;
1118 		child_proc->p_start.tv_sec = parent_proc->p_start.tv_sec;
1119 		child_proc->p_start.tv_usec = parent_proc->p_start.tv_usec;
1120 	} else {
1121 		/* Intialize new process stats, including start time */
1122 		/* <rdar://6640543> non-zeroed portion contains garbage AFAICT */
1123 		microtime_with_abstime(&child_proc->p_start, &child_proc->p_stats->ps_start);
1124 	}
1125 
1126 	if (pg->pg_session->s_ttyvp != NULL && parent_proc->p_flag & P_CONTROLT) {
1127 		os_atomic_or(&child_proc->p_flag, P_CONTROLT, relaxed);
1128 	}
1129 
1130 	/*
1131 	 * block all signals to reach the process.
1132 	 * no transition race should be occuring with the child yet,
1133 	 * but indicate that the process is in (the creation) transition.
1134 	 */
1135 	proc_signalstart(child_proc, 0);
1136 	proc_transstart(child_proc, 0, 0);
1137 
1138 	child_proc->p_pcaction = 0;
1139 
1140 	TAILQ_INIT(&child_proc->p_uthlist);
1141 	TAILQ_INIT(&child_proc->p_aio_activeq);
1142 	TAILQ_INIT(&child_proc->p_aio_doneq);
1143 
1144 	/*
1145 	 * Copy work queue information
1146 	 *
1147 	 * Note: This should probably only happen in the case where we are
1148 	 *	creating a child that is a copy of the parent; since this
1149 	 *	routine is called in the non-duplication case of vfork()
1150 	 *	or posix_spawn(), then this information should likely not
1151 	 *	be duplicated.
1152 	 *
1153 	 * <rdar://6640553> Work queue pointers that no longer point to code
1154 	 */
1155 	child_proc->p_wqthread = parent_proc->p_wqthread;
1156 	child_proc->p_threadstart = parent_proc->p_threadstart;
1157 	child_proc->p_pthsize = parent_proc->p_pthsize;
1158 	if ((parent_proc->p_lflag & P_LREGISTER) != 0) {
1159 		child_proc->p_lflag |= P_LREGISTER;
1160 	}
1161 	child_proc->p_dispatchqueue_offset = parent_proc->p_dispatchqueue_offset;
1162 	child_proc->p_dispatchqueue_serialno_offset = parent_proc->p_dispatchqueue_serialno_offset;
1163 	child_proc->p_dispatchqueue_label_offset = parent_proc->p_dispatchqueue_label_offset;
1164 	child_proc->p_return_to_kernel_offset = parent_proc->p_return_to_kernel_offset;
1165 	child_proc->p_mach_thread_self_offset = parent_proc->p_mach_thread_self_offset;
1166 	child_proc->p_pth_tsd_offset = parent_proc->p_pth_tsd_offset;
1167 	child_proc->p_pthread_wq_quantum_offset = parent_proc->p_pthread_wq_quantum_offset;
1168 #if PSYNCH
1169 	pth_proc_hashinit(child_proc);
1170 #endif /* PSYNCH */
1171 
1172 #if CONFIG_PERSONAS
1173 	child_proc->p_persona = NULL;
1174 	if (parent_proc->p_persona) {
1175 		struct persona *persona = proc_persona_get(parent_proc);
1176 
1177 		if (persona) {
1178 			error = persona_proc_adopt(child_proc, persona, NULL);
1179 			if (error != 0) {
1180 				printf("forkproc: persona_proc_inherit failed (persona %d being destroyed?)\n",
1181 				    persona_get_id(persona));
1182 				forkproc_free(child_proc);
1183 				child_proc = NULL;
1184 				goto bad;
1185 			}
1186 		}
1187 	}
1188 #endif
1189 
1190 #if CONFIG_MEMORYSTATUS
1191 	/* Memorystatus init */
1192 	child_proc->p_memstat_state = 0;
1193 	child_proc->p_memstat_effectivepriority = JETSAM_PRIORITY_DEFAULT;
1194 	child_proc->p_memstat_requestedpriority = JETSAM_PRIORITY_DEFAULT;
1195 	child_proc->p_memstat_assertionpriority = 0;
1196 	child_proc->p_memstat_userdata          = 0;
1197 	child_proc->p_memstat_idle_start        = 0;
1198 	child_proc->p_memstat_idle_delta        = 0;
1199 	child_proc->p_memstat_memlimit          = 0;
1200 	child_proc->p_memstat_memlimit_active   = 0;
1201 	child_proc->p_memstat_memlimit_inactive = 0;
1202 	child_proc->p_memstat_relaunch_flags    = P_MEMSTAT_RELAUNCH_UNKNOWN;
1203 #if CONFIG_FREEZE
1204 	child_proc->p_memstat_freeze_sharedanon_pages = 0;
1205 #endif
1206 	child_proc->p_memstat_dirty = 0;
1207 	child_proc->p_memstat_idledeadline = 0;
1208 #endif /* CONFIG_MEMORYSTATUS */
1209 
1210 	if (parent_proc->p_subsystem_root_path) {
1211 		size_t parent_length = strlen(parent_proc->p_subsystem_root_path) + 1;
1212 		assert(parent_length <= MAXPATHLEN);
1213 		child_proc->p_subsystem_root_path = zalloc_flags(ZV_NAMEI,
1214 		    Z_WAITOK | Z_ZERO);
1215 		memcpy(child_proc->p_subsystem_root_path, parent_proc->p_subsystem_root_path, parent_length);
1216 	}
1217 
1218 bad:
1219 	return child_proc;
1220 }
1221 
1222 void
proc_lock(proc_t p)1223 proc_lock(proc_t p)
1224 {
1225 	LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_NOTOWNED);
1226 	lck_mtx_lock(&p->p_mlock);
1227 }
1228 
1229 void
proc_unlock(proc_t p)1230 proc_unlock(proc_t p)
1231 {
1232 	lck_mtx_unlock(&p->p_mlock);
1233 }
1234 
1235 void
proc_spinlock(proc_t p)1236 proc_spinlock(proc_t p)
1237 {
1238 	lck_spin_lock_grp(&p->p_slock, &proc_slock_grp);
1239 }
1240 
1241 void
proc_spinunlock(proc_t p)1242 proc_spinunlock(proc_t p)
1243 {
1244 	lck_spin_unlock(&p->p_slock);
1245 }
1246 
1247 void
proc_list_lock(void)1248 proc_list_lock(void)
1249 {
1250 	lck_mtx_lock(&proc_list_mlock);
1251 }
1252 
1253 void
proc_list_unlock(void)1254 proc_list_unlock(void)
1255 {
1256 	lck_mtx_unlock(&proc_list_mlock);
1257 }
1258 
1259 void
proc_ucred_lock(proc_t p)1260 proc_ucred_lock(proc_t p)
1261 {
1262 	lck_mtx_lock(&p->p_ucred_mlock);
1263 }
1264 
1265 void
proc_ucred_unlock(proc_t p)1266 proc_ucred_unlock(proc_t p)
1267 {
1268 	lck_mtx_unlock(&p->p_ucred_mlock);
1269 }
1270 
1271 void
proc_update_creds_onproc(proc_t p,kauth_cred_t cred)1272 proc_update_creds_onproc(proc_t p, kauth_cred_t cred)
1273 {
1274 	p->p_uid = kauth_cred_getuid(cred);
1275 	p->p_gid = kauth_cred_getgid(cred);
1276 	p->p_ruid = kauth_cred_getruid(cred);
1277 	p->p_rgid = kauth_cred_getrgid(cred);
1278 	p->p_svuid = kauth_cred_getsvuid(cred);
1279 	p->p_svgid = kauth_cred_getsvgid(cred);
1280 }
1281 
1282 
1283 bool
uthread_is64bit(struct uthread * uth)1284 uthread_is64bit(struct uthread *uth)
1285 {
1286 	return uth->uu_flag & UT_LP64;
1287 }
1288 
1289 void
uthread_init(task_t task,uthread_t uth,thread_ro_t tro_tpl,int workq_thread)1290 uthread_init(task_t task, uthread_t uth, thread_ro_t tro_tpl, int workq_thread)
1291 {
1292 	uthread_t uth_parent = current_uthread();
1293 
1294 	lck_spin_init(&uth->uu_rethrottle_lock, &rethrottle_lock_grp,
1295 	    LCK_ATTR_NULL);
1296 
1297 	/*
1298 	 * Lazily set the thread on the kernel VFS context
1299 	 * to the first thread made which will be vm_pageout_scan_thread.
1300 	 */
1301 	if (__improbable(vfs_context0.vc_thread == NULL)) {
1302 		extern thread_t vm_pageout_scan_thread;
1303 
1304 		assert(task == kernel_task);
1305 		assert(get_machthread(uth) == vm_pageout_scan_thread);
1306 		vfs_context0.vc_thread = get_machthread(uth);
1307 	}
1308 
1309 	if (task_get_64bit_addr(task)) {
1310 		uth->uu_flag |= UT_LP64;
1311 	}
1312 
1313 	/*
1314 	 * Thread inherits credential from the creating thread, if both
1315 	 * are in the same task.
1316 	 *
1317 	 * If the creating thread has no credential or is from another
1318 	 * task we can leave the new thread credential NULL.  If it needs
1319 	 * one later, it will be lazily assigned from the task's process.
1320 	 */
1321 	if (task == kernel_task) {
1322 		kauth_cred_set(&tro_tpl->tro_cred, vfs_context0.vc_ucred);
1323 		tro_tpl->tro_proc = kernproc;
1324 		tro_tpl->tro_proc_ro = kernproc->p_proc_ro;
1325 	} else if (!task_is_a_corpse(task)) {
1326 		thread_ro_t curtro = current_thread_ro();
1327 		proc_t p = get_bsdtask_info(task);
1328 
1329 		if (task == curtro->tro_task &&
1330 		    ((curtro->tro_flags & TRO_SETUID) == 0 || !workq_thread)) {
1331 			kauth_cred_set(&tro_tpl->tro_cred, curtro->tro_cred);
1332 			tro_tpl->tro_flags = (curtro->tro_flags & TRO_SETUID);
1333 			tro_tpl->tro_proc_ro = curtro->tro_proc_ro;
1334 		} else {
1335 			kauth_cred_t cred = kauth_cred_proc_ref(p);
1336 			kauth_cred_set_and_unref(&tro_tpl->tro_cred, &cred);
1337 			tro_tpl->tro_proc_ro = task_get_ro(task);
1338 		}
1339 		tro_tpl->tro_proc = p;
1340 
1341 		proc_lock(p);
1342 		if (workq_thread) {
1343 			/* workq_thread threads will not inherit masks */
1344 			uth->uu_sigmask = ~workq_threadmask;
1345 		} else if (uth_parent->uu_flag & UT_SAS_OLDMASK) {
1346 			uth->uu_sigmask = uth_parent->uu_oldmask;
1347 		} else {
1348 			uth->uu_sigmask = uth_parent->uu_sigmask;
1349 		}
1350 
1351 		TAILQ_INSERT_TAIL(&p->p_uthlist, uth, uu_list);
1352 		proc_unlock(p);
1353 
1354 #if CONFIG_DTRACE
1355 		if (p->p_dtrace_ptss_pages != NULL) {
1356 			uth->t_dtrace_scratch = dtrace_ptss_claim_entry(p);
1357 		}
1358 #endif
1359 	} else {
1360 		tro_tpl->tro_proc_ro = task_get_ro(task);
1361 	}
1362 
1363 	uth->uu_pending_sigreturn = 0;
1364 	uthread_init_proc_refcount(uth);
1365 }
1366 
1367 mach_port_name_t
uthread_joiner_port(struct uthread * uth)1368 uthread_joiner_port(struct uthread *uth)
1369 {
1370 	return uth->uu_save.uus_bsdthread_terminate.kport;
1371 }
1372 
1373 user_addr_t
uthread_joiner_address(uthread_t uth)1374 uthread_joiner_address(uthread_t uth)
1375 {
1376 	return uth->uu_save.uus_bsdthread_terminate.ulock_addr;
1377 }
1378 
1379 void
uthread_joiner_wake(task_t task,uthread_t uth)1380 uthread_joiner_wake(task_t task, uthread_t uth)
1381 {
1382 	struct _bsdthread_terminate bts = uth->uu_save.uus_bsdthread_terminate;
1383 
1384 	assert(bts.ulock_addr);
1385 	bzero(&uth->uu_save.uus_bsdthread_terminate, sizeof(bts));
1386 
1387 	int flags = UL_UNFAIR_LOCK | ULF_WAKE_ALL | ULF_WAKE_ALLOW_NON_OWNER;
1388 	(void)ulock_wake(task, flags, bts.ulock_addr, 0);
1389 	mach_port_deallocate(get_task_ipcspace(task), bts.kport);
1390 }
1391 
1392 /*
1393  * This routine frees the thread name field of the uthread_t structure. Split out of
1394  * uthread_cleanup() so thread name does not get deallocated while generating a corpse fork.
1395  */
1396 void
uthread_cleanup_name(uthread_t uth)1397 uthread_cleanup_name(uthread_t uth)
1398 {
1399 	/*
1400 	 * <rdar://17834538>
1401 	 * Set pth_name to NULL before calling free().
1402 	 * Previously there was a race condition in the
1403 	 * case this code was executing during a stackshot
1404 	 * where the stackshot could try and copy pth_name
1405 	 * after it had been freed and before if was marked
1406 	 * as null.
1407 	 */
1408 	if (uth->pth_name != NULL) {
1409 		void *pth_name = uth->pth_name;
1410 		uth->pth_name = NULL;
1411 		kfree_data(pth_name, MAXTHREADNAMESIZE);
1412 	}
1413 	return;
1414 }
1415 
1416 /*
1417  * This routine frees all the BSD context in uthread except the credential.
1418  * It does not free the uthread structure as well
1419  */
1420 void
uthread_cleanup(uthread_t uth,thread_ro_t tro)1421 uthread_cleanup(uthread_t uth, thread_ro_t tro)
1422 {
1423 	task_t task = tro->tro_task;
1424 	proc_t p    = tro->tro_proc;
1425 
1426 	uthread_assert_zero_proc_refcount(uth);
1427 
1428 	if (uth->uu_lowpri_window || uth->uu_throttle_info) {
1429 		/*
1430 		 * task is marked as a low priority I/O type
1431 		 * and we've somehow managed to not dismiss the throttle
1432 		 * through the normal exit paths back to user space...
1433 		 * no need to throttle this thread since its going away
1434 		 * but we do need to update our bookeeping w/r to throttled threads
1435 		 *
1436 		 * Calling this routine will clean up any throttle info reference
1437 		 * still inuse by the thread.
1438 		 */
1439 		throttle_lowpri_io(0);
1440 	}
1441 
1442 #if CONFIG_AUDIT
1443 	/*
1444 	 * Per-thread audit state should never last beyond system
1445 	 * call return.  Since we don't audit the thread creation/
1446 	 * removal, the thread state pointer should never be
1447 	 * non-NULL when we get here.
1448 	 */
1449 	assert(uth->uu_ar == NULL);
1450 #endif
1451 
1452 	if (uth->uu_select.nbytes) {
1453 		select_cleanup_uthread(&uth->uu_select);
1454 	}
1455 
1456 	if (uth->uu_cdir) {
1457 		vnode_rele(uth->uu_cdir);
1458 		uth->uu_cdir = NULLVP;
1459 	}
1460 
1461 	if (uth->uu_selset) {
1462 		select_set_free(uth->uu_selset);
1463 		uth->uu_selset = NULL;
1464 	}
1465 
1466 	os_reason_free(uth->uu_exit_reason);
1467 
1468 	if ((task != kernel_task) && p) {
1469 		/*
1470 		 * Remove the thread from the process list and
1471 		 * transfer [appropriate] pending signals to the process.
1472 		 * Do not remove the uthread from proc uthlist for exec
1473 		 * copy task, since they does not have a ref on proc and
1474 		 * would not have been added to the list.
1475 		 */
1476 		if (uth->uu_kqr_bound) {
1477 			kqueue_threadreq_unbind(p, uth->uu_kqr_bound);
1478 		}
1479 
1480 		if (get_bsdtask_info(task) == p) {
1481 			proc_lock(p);
1482 			TAILQ_REMOVE(&p->p_uthlist, uth, uu_list);
1483 			p->p_siglist |= (uth->uu_siglist & execmask & (~p->p_sigignore | sigcantmask));
1484 			proc_unlock(p);
1485 		}
1486 
1487 #if CONFIG_DTRACE
1488 		struct dtrace_ptss_page_entry *tmpptr = uth->t_dtrace_scratch;
1489 		uth->t_dtrace_scratch = NULL;
1490 		if (tmpptr != NULL) {
1491 			dtrace_ptss_release_entry(p, tmpptr);
1492 		}
1493 #endif
1494 	} else {
1495 		assert(!uth->uu_kqr_bound);
1496 	}
1497 }
1498 
1499 /* This routine releases the credential stored in uthread */
1500 void
uthread_cred_ref(struct ucred * ucred)1501 uthread_cred_ref(struct ucred *ucred)
1502 {
1503 	kauth_cred_ref(ucred);
1504 }
1505 
1506 void
uthread_cred_free(struct ucred * ucred)1507 uthread_cred_free(struct ucred *ucred)
1508 {
1509 	kauth_cred_set(&ucred, NOCRED);
1510 }
1511 
1512 /* This routine frees the uthread structure held in thread structure */
1513 void
uthread_destroy(uthread_t uth)1514 uthread_destroy(uthread_t uth)
1515 {
1516 	uthread_destroy_proc_refcount(uth);
1517 
1518 	if (uth->t_tombstone) {
1519 		kfree_type(struct doc_tombstone, uth->t_tombstone);
1520 		uth->t_tombstone = NULL;
1521 	}
1522 
1523 #if CONFIG_DEBUG_SYSCALL_REJECTION
1524 	size_t const bitstr_len = BITMAP_SIZE(mach_trap_count + nsysent);
1525 
1526 	if (uth->syscall_rejection_mask) {
1527 		kfree_data(uth->syscall_rejection_mask, bitstr_len);
1528 		uth->syscall_rejection_mask = NULL;
1529 	}
1530 
1531 	if (uth->syscall_rejection_once_mask) {
1532 		kfree_data(uth->syscall_rejection_once_mask, bitstr_len);
1533 		uth->syscall_rejection_once_mask = NULL;
1534 	}
1535 #endif /* CONFIG_DEBUG_SYSCALL_REJECTION */
1536 
1537 	lck_spin_destroy(&uth->uu_rethrottle_lock, &rethrottle_lock_grp);
1538 
1539 	uthread_cleanup_name(uth);
1540 }
1541 
1542 user_addr_t
thread_get_sigreturn_token(thread_t thread)1543 thread_get_sigreturn_token(thread_t thread)
1544 {
1545 	uthread_t ut = (struct uthread *) get_bsdthread_info(thread);
1546 	return ut->uu_sigreturn_token;
1547 }
1548 
1549 uint32_t
thread_get_sigreturn_diversifier(thread_t thread)1550 thread_get_sigreturn_diversifier(thread_t thread)
1551 {
1552 	uthread_t ut = (struct uthread *) get_bsdthread_info(thread);
1553 	return ut->uu_sigreturn_diversifier;
1554 }
1555