1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/kernel/exit.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/mm.h>
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/task_io_accounting_ops.h>
52 #include <linux/blkdev.h>
53 #include <linux/task_work.h>
54 #include <linux/fs_struct.h>
55 #include <linux/init_task.h>
56 #include <linux/perf_event.h>
57 #include <trace/events/sched.h>
58 #include <linux/hw_breakpoint.h>
59 #include <linux/oom.h>
60 #include <linux/writeback.h>
61 #include <linux/shm.h>
62 #include <linux/kcov.h>
63 #include <linux/random.h>
64 #include <linux/rcuwait.h>
65 #include <linux/compat.h>
66 #include <linux/io_uring.h>
67 #include <linux/kprobes.h>
68 #include <linux/rethook.h>
69 
70 #include <linux/uaccess.h>
71 #include <asm/unistd.h>
72 #include <asm/mmu_context.h>
73 
__unhash_process(struct task_struct * p,bool group_dead)74 static void __unhash_process(struct task_struct *p, bool group_dead)
75 {
76 	nr_threads--;
77 	detach_pid(p, PIDTYPE_PID);
78 	if (group_dead) {
79 		detach_pid(p, PIDTYPE_TGID);
80 		detach_pid(p, PIDTYPE_PGID);
81 		detach_pid(p, PIDTYPE_SID);
82 
83 		list_del_rcu(&p->tasks);
84 		list_del_init(&p->sibling);
85 		__this_cpu_dec(process_counts);
86 	}
87 	list_del_rcu(&p->thread_group);
88 	list_del_rcu(&p->thread_node);
89 }
90 
91 /*
92  * This function expects the tasklist_lock write-locked.
93  */
__exit_signal(struct task_struct * tsk)94 static void __exit_signal(struct task_struct *tsk)
95 {
96 	struct signal_struct *sig = tsk->signal;
97 	bool group_dead = thread_group_leader(tsk);
98 	struct sighand_struct *sighand;
99 	struct tty_struct *tty;
100 	u64 utime, stime;
101 
102 	sighand = rcu_dereference_check(tsk->sighand,
103 					lockdep_tasklist_lock_is_held());
104 	spin_lock(&sighand->siglock);
105 
106 #ifdef CONFIG_POSIX_TIMERS
107 	posix_cpu_timers_exit(tsk);
108 	if (group_dead)
109 		posix_cpu_timers_exit_group(tsk);
110 #endif
111 
112 	if (group_dead) {
113 		tty = sig->tty;
114 		sig->tty = NULL;
115 	} else {
116 		/*
117 		 * If there is any task waiting for the group exit
118 		 * then notify it:
119 		 */
120 		if (sig->notify_count > 0 && !--sig->notify_count)
121 			wake_up_process(sig->group_exec_task);
122 
123 		if (tsk == sig->curr_target)
124 			sig->curr_target = next_thread(tsk);
125 	}
126 
127 	add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
128 			      sizeof(unsigned long long));
129 
130 	/*
131 	 * Accumulate here the counters for all threads as they die. We could
132 	 * skip the group leader because it is the last user of signal_struct,
133 	 * but we want to avoid the race with thread_group_cputime() which can
134 	 * see the empty ->thread_head list.
135 	 */
136 	task_cputime(tsk, &utime, &stime);
137 	write_seqlock(&sig->stats_lock);
138 	sig->utime += utime;
139 	sig->stime += stime;
140 	sig->gtime += task_gtime(tsk);
141 	sig->min_flt += tsk->min_flt;
142 	sig->maj_flt += tsk->maj_flt;
143 	sig->nvcsw += tsk->nvcsw;
144 	sig->nivcsw += tsk->nivcsw;
145 	sig->inblock += task_io_get_inblock(tsk);
146 	sig->oublock += task_io_get_oublock(tsk);
147 	task_io_accounting_add(&sig->ioac, &tsk->ioac);
148 	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
149 	sig->nr_threads--;
150 	__unhash_process(tsk, group_dead);
151 	write_sequnlock(&sig->stats_lock);
152 
153 	/*
154 	 * Do this under ->siglock, we can race with another thread
155 	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
156 	 */
157 	flush_sigqueue(&tsk->pending);
158 	tsk->sighand = NULL;
159 	spin_unlock(&sighand->siglock);
160 
161 	__cleanup_sighand(sighand);
162 	clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
163 	if (group_dead) {
164 		flush_sigqueue(&sig->shared_pending);
165 		tty_kref_put(tty);
166 	}
167 }
168 
delayed_put_task_struct(struct rcu_head * rhp)169 static void delayed_put_task_struct(struct rcu_head *rhp)
170 {
171 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
172 
173 	kprobe_flush_task(tsk);
174 	rethook_flush_task(tsk);
175 	perf_event_delayed_put(tsk);
176 	trace_sched_process_free(tsk);
177 	put_task_struct(tsk);
178 }
179 
put_task_struct_rcu_user(struct task_struct * task)180 void put_task_struct_rcu_user(struct task_struct *task)
181 {
182 	if (refcount_dec_and_test(&task->rcu_users))
183 		call_rcu(&task->rcu, delayed_put_task_struct);
184 }
185 
release_task(struct task_struct * p)186 void release_task(struct task_struct *p)
187 {
188 	struct task_struct *leader;
189 	struct pid *thread_pid;
190 	int zap_leader;
191 repeat:
192 	/* don't need to get the RCU readlock here - the process is dead and
193 	 * can't be modifying its own credentials. But shut RCU-lockdep up */
194 	rcu_read_lock();
195 	dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
196 	rcu_read_unlock();
197 
198 	cgroup_release(p);
199 
200 	write_lock_irq(&tasklist_lock);
201 	ptrace_release_task(p);
202 	thread_pid = get_pid(p->thread_pid);
203 	__exit_signal(p);
204 
205 	/*
206 	 * If we are the last non-leader member of the thread
207 	 * group, and the leader is zombie, then notify the
208 	 * group leader's parent process. (if it wants notification.)
209 	 */
210 	zap_leader = 0;
211 	leader = p->group_leader;
212 	if (leader != p && thread_group_empty(leader)
213 			&& leader->exit_state == EXIT_ZOMBIE) {
214 		/*
215 		 * If we were the last child thread and the leader has
216 		 * exited already, and the leader's parent ignores SIGCHLD,
217 		 * then we are the one who should release the leader.
218 		 */
219 		zap_leader = do_notify_parent(leader, leader->exit_signal);
220 		if (zap_leader)
221 			leader->exit_state = EXIT_DEAD;
222 	}
223 
224 	write_unlock_irq(&tasklist_lock);
225 	seccomp_filter_release(p);
226 	proc_flush_pid(thread_pid);
227 	put_pid(thread_pid);
228 	release_thread(p);
229 	put_task_struct_rcu_user(p);
230 
231 	p = leader;
232 	if (unlikely(zap_leader))
233 		goto repeat;
234 }
235 
rcuwait_wake_up(struct rcuwait * w)236 int rcuwait_wake_up(struct rcuwait *w)
237 {
238 	int ret = 0;
239 	struct task_struct *task;
240 
241 	rcu_read_lock();
242 
243 	/*
244 	 * Order condition vs @task, such that everything prior to the load
245 	 * of @task is visible. This is the condition as to why the user called
246 	 * rcuwait_wake() in the first place. Pairs with set_current_state()
247 	 * barrier (A) in rcuwait_wait_event().
248 	 *
249 	 *    WAIT                WAKE
250 	 *    [S] tsk = current	  [S] cond = true
251 	 *        MB (A)	      MB (B)
252 	 *    [L] cond		  [L] tsk
253 	 */
254 	smp_mb(); /* (B) */
255 
256 	task = rcu_dereference(w->task);
257 	if (task)
258 		ret = wake_up_process(task);
259 	rcu_read_unlock();
260 
261 	return ret;
262 }
263 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
264 
265 /*
266  * Determine if a process group is "orphaned", according to the POSIX
267  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
268  * by terminal-generated stop signals.  Newly orphaned process groups are
269  * to receive a SIGHUP and a SIGCONT.
270  *
271  * "I ask you, have you ever known what it is to be an orphan?"
272  */
will_become_orphaned_pgrp(struct pid * pgrp,struct task_struct * ignored_task)273 static int will_become_orphaned_pgrp(struct pid *pgrp,
274 					struct task_struct *ignored_task)
275 {
276 	struct task_struct *p;
277 
278 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
279 		if ((p == ignored_task) ||
280 		    (p->exit_state && thread_group_empty(p)) ||
281 		    is_global_init(p->real_parent))
282 			continue;
283 
284 		if (task_pgrp(p->real_parent) != pgrp &&
285 		    task_session(p->real_parent) == task_session(p))
286 			return 0;
287 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
288 
289 	return 1;
290 }
291 
is_current_pgrp_orphaned(void)292 int is_current_pgrp_orphaned(void)
293 {
294 	int retval;
295 
296 	read_lock(&tasklist_lock);
297 	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
298 	read_unlock(&tasklist_lock);
299 
300 	return retval;
301 }
302 
has_stopped_jobs(struct pid * pgrp)303 static bool has_stopped_jobs(struct pid *pgrp)
304 {
305 	struct task_struct *p;
306 
307 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
308 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
309 			return true;
310 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
311 
312 	return false;
313 }
314 
315 /*
316  * Check to see if any process groups have become orphaned as
317  * a result of our exiting, and if they have any stopped jobs,
318  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
319  */
320 static void
kill_orphaned_pgrp(struct task_struct * tsk,struct task_struct * parent)321 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
322 {
323 	struct pid *pgrp = task_pgrp(tsk);
324 	struct task_struct *ignored_task = tsk;
325 
326 	if (!parent)
327 		/* exit: our father is in a different pgrp than
328 		 * we are and we were the only connection outside.
329 		 */
330 		parent = tsk->real_parent;
331 	else
332 		/* reparent: our child is in a different pgrp than
333 		 * we are, and it was the only connection outside.
334 		 */
335 		ignored_task = NULL;
336 
337 	if (task_pgrp(parent) != pgrp &&
338 	    task_session(parent) == task_session(tsk) &&
339 	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
340 	    has_stopped_jobs(pgrp)) {
341 		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
342 		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
343 	}
344 }
345 
coredump_task_exit(struct task_struct * tsk)346 static void coredump_task_exit(struct task_struct *tsk)
347 {
348 	struct core_state *core_state;
349 
350 	/*
351 	 * Serialize with any possible pending coredump.
352 	 * We must hold siglock around checking core_state
353 	 * and setting PF_POSTCOREDUMP.  The core-inducing thread
354 	 * will increment ->nr_threads for each thread in the
355 	 * group without PF_POSTCOREDUMP set.
356 	 */
357 	spin_lock_irq(&tsk->sighand->siglock);
358 	tsk->flags |= PF_POSTCOREDUMP;
359 	core_state = tsk->signal->core_state;
360 	spin_unlock_irq(&tsk->sighand->siglock);
361 	if (core_state) {
362 		struct core_thread self;
363 
364 		self.task = current;
365 		if (self.task->flags & PF_SIGNALED)
366 			self.next = xchg(&core_state->dumper.next, &self);
367 		else
368 			self.task = NULL;
369 		/*
370 		 * Implies mb(), the result of xchg() must be visible
371 		 * to core_state->dumper.
372 		 */
373 		if (atomic_dec_and_test(&core_state->nr_threads))
374 			complete(&core_state->startup);
375 
376 		for (;;) {
377 			set_current_state(TASK_UNINTERRUPTIBLE);
378 			if (!self.task) /* see coredump_finish() */
379 				break;
380 			freezable_schedule();
381 		}
382 		__set_current_state(TASK_RUNNING);
383 	}
384 }
385 
386 #ifdef CONFIG_MEMCG
387 /*
388  * A task is exiting.   If it owned this mm, find a new owner for the mm.
389  */
mm_update_next_owner(struct mm_struct * mm)390 void mm_update_next_owner(struct mm_struct *mm)
391 {
392 	struct task_struct *c, *g, *p = current;
393 
394 retry:
395 	/*
396 	 * If the exiting or execing task is not the owner, it's
397 	 * someone else's problem.
398 	 */
399 	if (mm->owner != p)
400 		return;
401 	/*
402 	 * The current owner is exiting/execing and there are no other
403 	 * candidates.  Do not leave the mm pointing to a possibly
404 	 * freed task structure.
405 	 */
406 	if (atomic_read(&mm->mm_users) <= 1) {
407 		WRITE_ONCE(mm->owner, NULL);
408 		return;
409 	}
410 
411 	read_lock(&tasklist_lock);
412 	/*
413 	 * Search in the children
414 	 */
415 	list_for_each_entry(c, &p->children, sibling) {
416 		if (c->mm == mm)
417 			goto assign_new_owner;
418 	}
419 
420 	/*
421 	 * Search in the siblings
422 	 */
423 	list_for_each_entry(c, &p->real_parent->children, sibling) {
424 		if (c->mm == mm)
425 			goto assign_new_owner;
426 	}
427 
428 	/*
429 	 * Search through everything else, we should not get here often.
430 	 */
431 	for_each_process(g) {
432 		if (g->flags & PF_KTHREAD)
433 			continue;
434 		for_each_thread(g, c) {
435 			if (c->mm == mm)
436 				goto assign_new_owner;
437 			if (c->mm)
438 				break;
439 		}
440 	}
441 	read_unlock(&tasklist_lock);
442 	/*
443 	 * We found no owner yet mm_users > 1: this implies that we are
444 	 * most likely racing with swapoff (try_to_unuse()) or /proc or
445 	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
446 	 */
447 	WRITE_ONCE(mm->owner, NULL);
448 	return;
449 
450 assign_new_owner:
451 	BUG_ON(c == p);
452 	get_task_struct(c);
453 	/*
454 	 * The task_lock protects c->mm from changing.
455 	 * We always want mm->owner->mm == mm
456 	 */
457 	task_lock(c);
458 	/*
459 	 * Delay read_unlock() till we have the task_lock()
460 	 * to ensure that c does not slip away underneath us
461 	 */
462 	read_unlock(&tasklist_lock);
463 	if (c->mm != mm) {
464 		task_unlock(c);
465 		put_task_struct(c);
466 		goto retry;
467 	}
468 	WRITE_ONCE(mm->owner, c);
469 	task_unlock(c);
470 	put_task_struct(c);
471 }
472 #endif /* CONFIG_MEMCG */
473 
474 /*
475  * Turn us into a lazy TLB process if we
476  * aren't already..
477  */
exit_mm(void)478 static void exit_mm(void)
479 {
480 	struct mm_struct *mm = current->mm;
481 
482 	exit_mm_release(current, mm);
483 	if (!mm)
484 		return;
485 	sync_mm_rss(mm);
486 	mmap_read_lock(mm);
487 	mmgrab(mm);
488 	BUG_ON(mm != current->active_mm);
489 	/* more a memory barrier than a real lock */
490 	task_lock(current);
491 	/*
492 	 * When a thread stops operating on an address space, the loop
493 	 * in membarrier_private_expedited() may not observe that
494 	 * tsk->mm, and the loop in membarrier_global_expedited() may
495 	 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
496 	 * rq->membarrier_state, so those would not issue an IPI.
497 	 * Membarrier requires a memory barrier after accessing
498 	 * user-space memory, before clearing tsk->mm or the
499 	 * rq->membarrier_state.
500 	 */
501 	smp_mb__after_spinlock();
502 	local_irq_disable();
503 	current->mm = NULL;
504 	membarrier_update_current_mm(NULL);
505 	enter_lazy_tlb(mm, current);
506 	local_irq_enable();
507 	task_unlock(current);
508 	mmap_read_unlock(mm);
509 	mm_update_next_owner(mm);
510 	mmput(mm);
511 	if (test_thread_flag(TIF_MEMDIE))
512 		exit_oom_victim();
513 }
514 
find_alive_thread(struct task_struct * p)515 static struct task_struct *find_alive_thread(struct task_struct *p)
516 {
517 	struct task_struct *t;
518 
519 	for_each_thread(p, t) {
520 		if (!(t->flags & PF_EXITING))
521 			return t;
522 	}
523 	return NULL;
524 }
525 
find_child_reaper(struct task_struct * father,struct list_head * dead)526 static struct task_struct *find_child_reaper(struct task_struct *father,
527 						struct list_head *dead)
528 	__releases(&tasklist_lock)
529 	__acquires(&tasklist_lock)
530 {
531 	struct pid_namespace *pid_ns = task_active_pid_ns(father);
532 	struct task_struct *reaper = pid_ns->child_reaper;
533 	struct task_struct *p, *n;
534 
535 	if (likely(reaper != father))
536 		return reaper;
537 
538 	reaper = find_alive_thread(father);
539 	if (reaper) {
540 		pid_ns->child_reaper = reaper;
541 		return reaper;
542 	}
543 
544 	write_unlock_irq(&tasklist_lock);
545 
546 	list_for_each_entry_safe(p, n, dead, ptrace_entry) {
547 		list_del_init(&p->ptrace_entry);
548 		release_task(p);
549 	}
550 
551 	zap_pid_ns_processes(pid_ns);
552 	write_lock_irq(&tasklist_lock);
553 
554 	return father;
555 }
556 
557 /*
558  * When we die, we re-parent all our children, and try to:
559  * 1. give them to another thread in our thread group, if such a member exists
560  * 2. give it to the first ancestor process which prctl'd itself as a
561  *    child_subreaper for its children (like a service manager)
562  * 3. give it to the init process (PID 1) in our pid namespace
563  */
find_new_reaper(struct task_struct * father,struct task_struct * child_reaper)564 static struct task_struct *find_new_reaper(struct task_struct *father,
565 					   struct task_struct *child_reaper)
566 {
567 	struct task_struct *thread, *reaper;
568 
569 	thread = find_alive_thread(father);
570 	if (thread)
571 		return thread;
572 
573 	if (father->signal->has_child_subreaper) {
574 		unsigned int ns_level = task_pid(father)->level;
575 		/*
576 		 * Find the first ->is_child_subreaper ancestor in our pid_ns.
577 		 * We can't check reaper != child_reaper to ensure we do not
578 		 * cross the namespaces, the exiting parent could be injected
579 		 * by setns() + fork().
580 		 * We check pid->level, this is slightly more efficient than
581 		 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
582 		 */
583 		for (reaper = father->real_parent;
584 		     task_pid(reaper)->level == ns_level;
585 		     reaper = reaper->real_parent) {
586 			if (reaper == &init_task)
587 				break;
588 			if (!reaper->signal->is_child_subreaper)
589 				continue;
590 			thread = find_alive_thread(reaper);
591 			if (thread)
592 				return thread;
593 		}
594 	}
595 
596 	return child_reaper;
597 }
598 
599 /*
600 * Any that need to be release_task'd are put on the @dead list.
601  */
reparent_leader(struct task_struct * father,struct task_struct * p,struct list_head * dead)602 static void reparent_leader(struct task_struct *father, struct task_struct *p,
603 				struct list_head *dead)
604 {
605 	if (unlikely(p->exit_state == EXIT_DEAD))
606 		return;
607 
608 	/* We don't want people slaying init. */
609 	p->exit_signal = SIGCHLD;
610 
611 	/* If it has exited notify the new parent about this child's death. */
612 	if (!p->ptrace &&
613 	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
614 		if (do_notify_parent(p, p->exit_signal)) {
615 			p->exit_state = EXIT_DEAD;
616 			list_add(&p->ptrace_entry, dead);
617 		}
618 	}
619 
620 	kill_orphaned_pgrp(p, father);
621 }
622 
623 /*
624  * This does two things:
625  *
626  * A.  Make init inherit all the child processes
627  * B.  Check to see if any process groups have become orphaned
628  *	as a result of our exiting, and if they have any stopped
629  *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
630  */
forget_original_parent(struct task_struct * father,struct list_head * dead)631 static void forget_original_parent(struct task_struct *father,
632 					struct list_head *dead)
633 {
634 	struct task_struct *p, *t, *reaper;
635 
636 	if (unlikely(!list_empty(&father->ptraced)))
637 		exit_ptrace(father, dead);
638 
639 	/* Can drop and reacquire tasklist_lock */
640 	reaper = find_child_reaper(father, dead);
641 	if (list_empty(&father->children))
642 		return;
643 
644 	reaper = find_new_reaper(father, reaper);
645 	list_for_each_entry(p, &father->children, sibling) {
646 		for_each_thread(p, t) {
647 			RCU_INIT_POINTER(t->real_parent, reaper);
648 			BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
649 			if (likely(!t->ptrace))
650 				t->parent = t->real_parent;
651 			if (t->pdeath_signal)
652 				group_send_sig_info(t->pdeath_signal,
653 						    SEND_SIG_NOINFO, t,
654 						    PIDTYPE_TGID);
655 		}
656 		/*
657 		 * If this is a threaded reparent there is no need to
658 		 * notify anyone anything has happened.
659 		 */
660 		if (!same_thread_group(reaper, father))
661 			reparent_leader(father, p, dead);
662 	}
663 	list_splice_tail_init(&father->children, &reaper->children);
664 }
665 
666 /*
667  * Send signals to all our closest relatives so that they know
668  * to properly mourn us..
669  */
exit_notify(struct task_struct * tsk,int group_dead)670 static void exit_notify(struct task_struct *tsk, int group_dead)
671 {
672 	bool autoreap;
673 	struct task_struct *p, *n;
674 	LIST_HEAD(dead);
675 
676 	write_lock_irq(&tasklist_lock);
677 	forget_original_parent(tsk, &dead);
678 
679 	if (group_dead)
680 		kill_orphaned_pgrp(tsk->group_leader, NULL);
681 
682 	tsk->exit_state = EXIT_ZOMBIE;
683 	if (unlikely(tsk->ptrace)) {
684 		int sig = thread_group_leader(tsk) &&
685 				thread_group_empty(tsk) &&
686 				!ptrace_reparented(tsk) ?
687 			tsk->exit_signal : SIGCHLD;
688 		autoreap = do_notify_parent(tsk, sig);
689 	} else if (thread_group_leader(tsk)) {
690 		autoreap = thread_group_empty(tsk) &&
691 			do_notify_parent(tsk, tsk->exit_signal);
692 	} else {
693 		autoreap = true;
694 	}
695 
696 	if (autoreap) {
697 		tsk->exit_state = EXIT_DEAD;
698 		list_add(&tsk->ptrace_entry, &dead);
699 	}
700 
701 	/* mt-exec, de_thread() is waiting for group leader */
702 	if (unlikely(tsk->signal->notify_count < 0))
703 		wake_up_process(tsk->signal->group_exec_task);
704 	write_unlock_irq(&tasklist_lock);
705 
706 	list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
707 		list_del_init(&p->ptrace_entry);
708 		release_task(p);
709 	}
710 }
711 
712 #ifdef CONFIG_DEBUG_STACK_USAGE
check_stack_usage(void)713 static void check_stack_usage(void)
714 {
715 	static DEFINE_SPINLOCK(low_water_lock);
716 	static int lowest_to_date = THREAD_SIZE;
717 	unsigned long free;
718 
719 	free = stack_not_used(current);
720 
721 	if (free >= lowest_to_date)
722 		return;
723 
724 	spin_lock(&low_water_lock);
725 	if (free < lowest_to_date) {
726 		pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
727 			current->comm, task_pid_nr(current), free);
728 		lowest_to_date = free;
729 	}
730 	spin_unlock(&low_water_lock);
731 }
732 #else
check_stack_usage(void)733 static inline void check_stack_usage(void) {}
734 #endif
735 
do_exit(long code)736 void __noreturn do_exit(long code)
737 {
738 	struct task_struct *tsk = current;
739 	int group_dead;
740 
741 	WARN_ON(tsk->plug);
742 
743 	kcov_task_exit(tsk);
744 
745 	coredump_task_exit(tsk);
746 	ptrace_event(PTRACE_EVENT_EXIT, code);
747 
748 	validate_creds_for_do_exit(tsk);
749 
750 	io_uring_files_cancel();
751 	exit_signals(tsk);  /* sets PF_EXITING */
752 
753 	/* sync mm's RSS info before statistics gathering */
754 	if (tsk->mm)
755 		sync_mm_rss(tsk->mm);
756 	acct_update_integrals(tsk);
757 	group_dead = atomic_dec_and_test(&tsk->signal->live);
758 	if (group_dead) {
759 		/*
760 		 * If the last thread of global init has exited, panic
761 		 * immediately to get a useable coredump.
762 		 */
763 		if (unlikely(is_global_init(tsk)))
764 			panic("Attempted to kill init! exitcode=0x%08x\n",
765 				tsk->signal->group_exit_code ?: (int)code);
766 
767 #ifdef CONFIG_POSIX_TIMERS
768 		hrtimer_cancel(&tsk->signal->real_timer);
769 		exit_itimers(tsk);
770 #endif
771 		if (tsk->mm)
772 			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
773 	}
774 	acct_collect(code, group_dead);
775 	if (group_dead)
776 		tty_audit_exit();
777 	audit_free(tsk);
778 
779 	tsk->exit_code = code;
780 	taskstats_exit(tsk, group_dead);
781 
782 	exit_mm();
783 
784 	if (group_dead)
785 		acct_process();
786 	trace_sched_process_exit(tsk);
787 
788 	exit_sem(tsk);
789 	exit_shm(tsk);
790 	exit_files(tsk);
791 	exit_fs(tsk);
792 	if (group_dead)
793 		disassociate_ctty(1);
794 	exit_task_namespaces(tsk);
795 	exit_task_work(tsk);
796 	exit_thread(tsk);
797 
798 	/*
799 	 * Flush inherited counters to the parent - before the parent
800 	 * gets woken up by child-exit notifications.
801 	 *
802 	 * because of cgroup mode, must be called before cgroup_exit()
803 	 */
804 	perf_event_exit_task(tsk);
805 
806 	sched_autogroup_exit_task(tsk);
807 	cgroup_exit(tsk);
808 
809 	/*
810 	 * FIXME: do that only when needed, using sched_exit tracepoint
811 	 */
812 	flush_ptrace_hw_breakpoint(tsk);
813 
814 	exit_tasks_rcu_start();
815 	exit_notify(tsk, group_dead);
816 	proc_exit_connector(tsk);
817 	mpol_put_task_policy(tsk);
818 #ifdef CONFIG_FUTEX
819 	if (unlikely(current->pi_state_cache))
820 		kfree(current->pi_state_cache);
821 #endif
822 	/*
823 	 * Make sure we are holding no locks:
824 	 */
825 	debug_check_no_locks_held();
826 
827 	if (tsk->io_context)
828 		exit_io_context(tsk);
829 
830 	if (tsk->splice_pipe)
831 		free_pipe_info(tsk->splice_pipe);
832 
833 	if (tsk->task_frag.page)
834 		put_page(tsk->task_frag.page);
835 
836 	validate_creds_for_do_exit(tsk);
837 	exit_task_stack_account(tsk);
838 
839 	check_stack_usage();
840 	preempt_disable();
841 	if (tsk->nr_dirtied)
842 		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
843 	exit_rcu();
844 	exit_tasks_rcu_finish();
845 
846 	lockdep_free_task(tsk);
847 	do_task_dead();
848 }
849 
make_task_dead(int signr)850 void __noreturn make_task_dead(int signr)
851 {
852 	/*
853 	 * Take the task off the cpu after something catastrophic has
854 	 * happened.
855 	 *
856 	 * We can get here from a kernel oops, sometimes with preemption off.
857 	 * Start by checking for critical errors.
858 	 * Then fix up important state like USER_DS and preemption.
859 	 * Then do everything else.
860 	 */
861 	struct task_struct *tsk = current;
862 
863 	if (unlikely(in_interrupt()))
864 		panic("Aiee, killing interrupt handler!");
865 	if (unlikely(!tsk->pid))
866 		panic("Attempted to kill the idle task!");
867 
868 	if (unlikely(in_atomic())) {
869 		pr_info("note: %s[%d] exited with preempt_count %d\n",
870 			current->comm, task_pid_nr(current),
871 			preempt_count());
872 		preempt_count_set(PREEMPT_ENABLED);
873 	}
874 
875 	/*
876 	 * We're taking recursive faults here in make_task_dead. Safest is to just
877 	 * leave this task alone and wait for reboot.
878 	 */
879 	if (unlikely(tsk->flags & PF_EXITING)) {
880 		pr_alert("Fixing recursive fault but reboot is needed!\n");
881 		futex_exit_recursive(tsk);
882 		tsk->exit_state = EXIT_DEAD;
883 		refcount_inc(&tsk->rcu_users);
884 		do_task_dead();
885 	}
886 
887 	do_exit(signr);
888 }
889 
SYSCALL_DEFINE1(exit,int,error_code)890 SYSCALL_DEFINE1(exit, int, error_code)
891 {
892 	do_exit((error_code&0xff)<<8);
893 }
894 
895 /*
896  * Take down every thread in the group.  This is called by fatal signals
897  * as well as by sys_exit_group (below).
898  */
899 void __noreturn
do_group_exit(int exit_code)900 do_group_exit(int exit_code)
901 {
902 	struct signal_struct *sig = current->signal;
903 
904 	if (sig->flags & SIGNAL_GROUP_EXIT)
905 		exit_code = sig->group_exit_code;
906 	else if (sig->group_exec_task)
907 		exit_code = 0;
908 	else if (!thread_group_empty(current)) {
909 		struct sighand_struct *const sighand = current->sighand;
910 
911 		spin_lock_irq(&sighand->siglock);
912 		if (sig->flags & SIGNAL_GROUP_EXIT)
913 			/* Another thread got here before we took the lock.  */
914 			exit_code = sig->group_exit_code;
915 		else if (sig->group_exec_task)
916 			exit_code = 0;
917 		else {
918 			sig->group_exit_code = exit_code;
919 			sig->flags = SIGNAL_GROUP_EXIT;
920 			zap_other_threads(current);
921 		}
922 		spin_unlock_irq(&sighand->siglock);
923 	}
924 
925 	do_exit(exit_code);
926 	/* NOTREACHED */
927 }
928 
929 /*
930  * this kills every thread in the thread group. Note that any externally
931  * wait4()-ing process will get the correct exit code - even if this
932  * thread is not the thread group leader.
933  */
SYSCALL_DEFINE1(exit_group,int,error_code)934 SYSCALL_DEFINE1(exit_group, int, error_code)
935 {
936 	do_group_exit((error_code & 0xff) << 8);
937 	/* NOTREACHED */
938 	return 0;
939 }
940 
941 struct waitid_info {
942 	pid_t pid;
943 	uid_t uid;
944 	int status;
945 	int cause;
946 };
947 
948 struct wait_opts {
949 	enum pid_type		wo_type;
950 	int			wo_flags;
951 	struct pid		*wo_pid;
952 
953 	struct waitid_info	*wo_info;
954 	int			wo_stat;
955 	struct rusage		*wo_rusage;
956 
957 	wait_queue_entry_t		child_wait;
958 	int			notask_error;
959 };
960 
eligible_pid(struct wait_opts * wo,struct task_struct * p)961 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
962 {
963 	return	wo->wo_type == PIDTYPE_MAX ||
964 		task_pid_type(p, wo->wo_type) == wo->wo_pid;
965 }
966 
967 static int
eligible_child(struct wait_opts * wo,bool ptrace,struct task_struct * p)968 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
969 {
970 	if (!eligible_pid(wo, p))
971 		return 0;
972 
973 	/*
974 	 * Wait for all children (clone and not) if __WALL is set or
975 	 * if it is traced by us.
976 	 */
977 	if (ptrace || (wo->wo_flags & __WALL))
978 		return 1;
979 
980 	/*
981 	 * Otherwise, wait for clone children *only* if __WCLONE is set;
982 	 * otherwise, wait for non-clone children *only*.
983 	 *
984 	 * Note: a "clone" child here is one that reports to its parent
985 	 * using a signal other than SIGCHLD, or a non-leader thread which
986 	 * we can only see if it is traced by us.
987 	 */
988 	if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
989 		return 0;
990 
991 	return 1;
992 }
993 
994 /*
995  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
996  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
997  * the lock and this task is uninteresting.  If we return nonzero, we have
998  * released the lock and the system call should return.
999  */
wait_task_zombie(struct wait_opts * wo,struct task_struct * p)1000 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1001 {
1002 	int state, status;
1003 	pid_t pid = task_pid_vnr(p);
1004 	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1005 	struct waitid_info *infop;
1006 
1007 	if (!likely(wo->wo_flags & WEXITED))
1008 		return 0;
1009 
1010 	if (unlikely(wo->wo_flags & WNOWAIT)) {
1011 		status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1012 			? p->signal->group_exit_code : p->exit_code;
1013 		get_task_struct(p);
1014 		read_unlock(&tasklist_lock);
1015 		sched_annotate_sleep();
1016 		if (wo->wo_rusage)
1017 			getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1018 		put_task_struct(p);
1019 		goto out_info;
1020 	}
1021 	/*
1022 	 * Move the task's state to DEAD/TRACE, only one thread can do this.
1023 	 */
1024 	state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1025 		EXIT_TRACE : EXIT_DEAD;
1026 	if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1027 		return 0;
1028 	/*
1029 	 * We own this thread, nobody else can reap it.
1030 	 */
1031 	read_unlock(&tasklist_lock);
1032 	sched_annotate_sleep();
1033 
1034 	/*
1035 	 * Check thread_group_leader() to exclude the traced sub-threads.
1036 	 */
1037 	if (state == EXIT_DEAD && thread_group_leader(p)) {
1038 		struct signal_struct *sig = p->signal;
1039 		struct signal_struct *psig = current->signal;
1040 		unsigned long maxrss;
1041 		u64 tgutime, tgstime;
1042 
1043 		/*
1044 		 * The resource counters for the group leader are in its
1045 		 * own task_struct.  Those for dead threads in the group
1046 		 * are in its signal_struct, as are those for the child
1047 		 * processes it has previously reaped.  All these
1048 		 * accumulate in the parent's signal_struct c* fields.
1049 		 *
1050 		 * We don't bother to take a lock here to protect these
1051 		 * p->signal fields because the whole thread group is dead
1052 		 * and nobody can change them.
1053 		 *
1054 		 * psig->stats_lock also protects us from our sub-theads
1055 		 * which can reap other children at the same time. Until
1056 		 * we change k_getrusage()-like users to rely on this lock
1057 		 * we have to take ->siglock as well.
1058 		 *
1059 		 * We use thread_group_cputime_adjusted() to get times for
1060 		 * the thread group, which consolidates times for all threads
1061 		 * in the group including the group leader.
1062 		 */
1063 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1064 		spin_lock_irq(&current->sighand->siglock);
1065 		write_seqlock(&psig->stats_lock);
1066 		psig->cutime += tgutime + sig->cutime;
1067 		psig->cstime += tgstime + sig->cstime;
1068 		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1069 		psig->cmin_flt +=
1070 			p->min_flt + sig->min_flt + sig->cmin_flt;
1071 		psig->cmaj_flt +=
1072 			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1073 		psig->cnvcsw +=
1074 			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1075 		psig->cnivcsw +=
1076 			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1077 		psig->cinblock +=
1078 			task_io_get_inblock(p) +
1079 			sig->inblock + sig->cinblock;
1080 		psig->coublock +=
1081 			task_io_get_oublock(p) +
1082 			sig->oublock + sig->coublock;
1083 		maxrss = max(sig->maxrss, sig->cmaxrss);
1084 		if (psig->cmaxrss < maxrss)
1085 			psig->cmaxrss = maxrss;
1086 		task_io_accounting_add(&psig->ioac, &p->ioac);
1087 		task_io_accounting_add(&psig->ioac, &sig->ioac);
1088 		write_sequnlock(&psig->stats_lock);
1089 		spin_unlock_irq(&current->sighand->siglock);
1090 	}
1091 
1092 	if (wo->wo_rusage)
1093 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1094 	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1095 		? p->signal->group_exit_code : p->exit_code;
1096 	wo->wo_stat = status;
1097 
1098 	if (state == EXIT_TRACE) {
1099 		write_lock_irq(&tasklist_lock);
1100 		/* We dropped tasklist, ptracer could die and untrace */
1101 		ptrace_unlink(p);
1102 
1103 		/* If parent wants a zombie, don't release it now */
1104 		state = EXIT_ZOMBIE;
1105 		if (do_notify_parent(p, p->exit_signal))
1106 			state = EXIT_DEAD;
1107 		p->exit_state = state;
1108 		write_unlock_irq(&tasklist_lock);
1109 	}
1110 	if (state == EXIT_DEAD)
1111 		release_task(p);
1112 
1113 out_info:
1114 	infop = wo->wo_info;
1115 	if (infop) {
1116 		if ((status & 0x7f) == 0) {
1117 			infop->cause = CLD_EXITED;
1118 			infop->status = status >> 8;
1119 		} else {
1120 			infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1121 			infop->status = status & 0x7f;
1122 		}
1123 		infop->pid = pid;
1124 		infop->uid = uid;
1125 	}
1126 
1127 	return pid;
1128 }
1129 
task_stopped_code(struct task_struct * p,bool ptrace)1130 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1131 {
1132 	if (ptrace) {
1133 		if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1134 			return &p->exit_code;
1135 	} else {
1136 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1137 			return &p->signal->group_exit_code;
1138 	}
1139 	return NULL;
1140 }
1141 
1142 /**
1143  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1144  * @wo: wait options
1145  * @ptrace: is the wait for ptrace
1146  * @p: task to wait for
1147  *
1148  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1149  *
1150  * CONTEXT:
1151  * read_lock(&tasklist_lock), which is released if return value is
1152  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1153  *
1154  * RETURNS:
1155  * 0 if wait condition didn't exist and search for other wait conditions
1156  * should continue.  Non-zero return, -errno on failure and @p's pid on
1157  * success, implies that tasklist_lock is released and wait condition
1158  * search should terminate.
1159  */
wait_task_stopped(struct wait_opts * wo,int ptrace,struct task_struct * p)1160 static int wait_task_stopped(struct wait_opts *wo,
1161 				int ptrace, struct task_struct *p)
1162 {
1163 	struct waitid_info *infop;
1164 	int exit_code, *p_code, why;
1165 	uid_t uid = 0; /* unneeded, required by compiler */
1166 	pid_t pid;
1167 
1168 	/*
1169 	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1170 	 */
1171 	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1172 		return 0;
1173 
1174 	if (!task_stopped_code(p, ptrace))
1175 		return 0;
1176 
1177 	exit_code = 0;
1178 	spin_lock_irq(&p->sighand->siglock);
1179 
1180 	p_code = task_stopped_code(p, ptrace);
1181 	if (unlikely(!p_code))
1182 		goto unlock_sig;
1183 
1184 	exit_code = *p_code;
1185 	if (!exit_code)
1186 		goto unlock_sig;
1187 
1188 	if (!unlikely(wo->wo_flags & WNOWAIT))
1189 		*p_code = 0;
1190 
1191 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1192 unlock_sig:
1193 	spin_unlock_irq(&p->sighand->siglock);
1194 	if (!exit_code)
1195 		return 0;
1196 
1197 	/*
1198 	 * Now we are pretty sure this task is interesting.
1199 	 * Make sure it doesn't get reaped out from under us while we
1200 	 * give up the lock and then examine it below.  We don't want to
1201 	 * keep holding onto the tasklist_lock while we call getrusage and
1202 	 * possibly take page faults for user memory.
1203 	 */
1204 	get_task_struct(p);
1205 	pid = task_pid_vnr(p);
1206 	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1207 	read_unlock(&tasklist_lock);
1208 	sched_annotate_sleep();
1209 	if (wo->wo_rusage)
1210 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1211 	put_task_struct(p);
1212 
1213 	if (likely(!(wo->wo_flags & WNOWAIT)))
1214 		wo->wo_stat = (exit_code << 8) | 0x7f;
1215 
1216 	infop = wo->wo_info;
1217 	if (infop) {
1218 		infop->cause = why;
1219 		infop->status = exit_code;
1220 		infop->pid = pid;
1221 		infop->uid = uid;
1222 	}
1223 	return pid;
1224 }
1225 
1226 /*
1227  * Handle do_wait work for one task in a live, non-stopped state.
1228  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1229  * the lock and this task is uninteresting.  If we return nonzero, we have
1230  * released the lock and the system call should return.
1231  */
wait_task_continued(struct wait_opts * wo,struct task_struct * p)1232 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1233 {
1234 	struct waitid_info *infop;
1235 	pid_t pid;
1236 	uid_t uid;
1237 
1238 	if (!unlikely(wo->wo_flags & WCONTINUED))
1239 		return 0;
1240 
1241 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1242 		return 0;
1243 
1244 	spin_lock_irq(&p->sighand->siglock);
1245 	/* Re-check with the lock held.  */
1246 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1247 		spin_unlock_irq(&p->sighand->siglock);
1248 		return 0;
1249 	}
1250 	if (!unlikely(wo->wo_flags & WNOWAIT))
1251 		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1252 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1253 	spin_unlock_irq(&p->sighand->siglock);
1254 
1255 	pid = task_pid_vnr(p);
1256 	get_task_struct(p);
1257 	read_unlock(&tasklist_lock);
1258 	sched_annotate_sleep();
1259 	if (wo->wo_rusage)
1260 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1261 	put_task_struct(p);
1262 
1263 	infop = wo->wo_info;
1264 	if (!infop) {
1265 		wo->wo_stat = 0xffff;
1266 	} else {
1267 		infop->cause = CLD_CONTINUED;
1268 		infop->pid = pid;
1269 		infop->uid = uid;
1270 		infop->status = SIGCONT;
1271 	}
1272 	return pid;
1273 }
1274 
1275 /*
1276  * Consider @p for a wait by @parent.
1277  *
1278  * -ECHILD should be in ->notask_error before the first call.
1279  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1280  * Returns zero if the search for a child should continue;
1281  * then ->notask_error is 0 if @p is an eligible child,
1282  * or still -ECHILD.
1283  */
wait_consider_task(struct wait_opts * wo,int ptrace,struct task_struct * p)1284 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1285 				struct task_struct *p)
1286 {
1287 	/*
1288 	 * We can race with wait_task_zombie() from another thread.
1289 	 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1290 	 * can't confuse the checks below.
1291 	 */
1292 	int exit_state = READ_ONCE(p->exit_state);
1293 	int ret;
1294 
1295 	if (unlikely(exit_state == EXIT_DEAD))
1296 		return 0;
1297 
1298 	ret = eligible_child(wo, ptrace, p);
1299 	if (!ret)
1300 		return ret;
1301 
1302 	if (unlikely(exit_state == EXIT_TRACE)) {
1303 		/*
1304 		 * ptrace == 0 means we are the natural parent. In this case
1305 		 * we should clear notask_error, debugger will notify us.
1306 		 */
1307 		if (likely(!ptrace))
1308 			wo->notask_error = 0;
1309 		return 0;
1310 	}
1311 
1312 	if (likely(!ptrace) && unlikely(p->ptrace)) {
1313 		/*
1314 		 * If it is traced by its real parent's group, just pretend
1315 		 * the caller is ptrace_do_wait() and reap this child if it
1316 		 * is zombie.
1317 		 *
1318 		 * This also hides group stop state from real parent; otherwise
1319 		 * a single stop can be reported twice as group and ptrace stop.
1320 		 * If a ptracer wants to distinguish these two events for its
1321 		 * own children it should create a separate process which takes
1322 		 * the role of real parent.
1323 		 */
1324 		if (!ptrace_reparented(p))
1325 			ptrace = 1;
1326 	}
1327 
1328 	/* slay zombie? */
1329 	if (exit_state == EXIT_ZOMBIE) {
1330 		/* we don't reap group leaders with subthreads */
1331 		if (!delay_group_leader(p)) {
1332 			/*
1333 			 * A zombie ptracee is only visible to its ptracer.
1334 			 * Notification and reaping will be cascaded to the
1335 			 * real parent when the ptracer detaches.
1336 			 */
1337 			if (unlikely(ptrace) || likely(!p->ptrace))
1338 				return wait_task_zombie(wo, p);
1339 		}
1340 
1341 		/*
1342 		 * Allow access to stopped/continued state via zombie by
1343 		 * falling through.  Clearing of notask_error is complex.
1344 		 *
1345 		 * When !@ptrace:
1346 		 *
1347 		 * If WEXITED is set, notask_error should naturally be
1348 		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1349 		 * so, if there are live subthreads, there are events to
1350 		 * wait for.  If all subthreads are dead, it's still safe
1351 		 * to clear - this function will be called again in finite
1352 		 * amount time once all the subthreads are released and
1353 		 * will then return without clearing.
1354 		 *
1355 		 * When @ptrace:
1356 		 *
1357 		 * Stopped state is per-task and thus can't change once the
1358 		 * target task dies.  Only continued and exited can happen.
1359 		 * Clear notask_error if WCONTINUED | WEXITED.
1360 		 */
1361 		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1362 			wo->notask_error = 0;
1363 	} else {
1364 		/*
1365 		 * @p is alive and it's gonna stop, continue or exit, so
1366 		 * there always is something to wait for.
1367 		 */
1368 		wo->notask_error = 0;
1369 	}
1370 
1371 	/*
1372 	 * Wait for stopped.  Depending on @ptrace, different stopped state
1373 	 * is used and the two don't interact with each other.
1374 	 */
1375 	ret = wait_task_stopped(wo, ptrace, p);
1376 	if (ret)
1377 		return ret;
1378 
1379 	/*
1380 	 * Wait for continued.  There's only one continued state and the
1381 	 * ptracer can consume it which can confuse the real parent.  Don't
1382 	 * use WCONTINUED from ptracer.  You don't need or want it.
1383 	 */
1384 	return wait_task_continued(wo, p);
1385 }
1386 
1387 /*
1388  * Do the work of do_wait() for one thread in the group, @tsk.
1389  *
1390  * -ECHILD should be in ->notask_error before the first call.
1391  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1392  * Returns zero if the search for a child should continue; then
1393  * ->notask_error is 0 if there were any eligible children,
1394  * or still -ECHILD.
1395  */
do_wait_thread(struct wait_opts * wo,struct task_struct * tsk)1396 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1397 {
1398 	struct task_struct *p;
1399 
1400 	list_for_each_entry(p, &tsk->children, sibling) {
1401 		int ret = wait_consider_task(wo, 0, p);
1402 
1403 		if (ret)
1404 			return ret;
1405 	}
1406 
1407 	return 0;
1408 }
1409 
ptrace_do_wait(struct wait_opts * wo,struct task_struct * tsk)1410 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1411 {
1412 	struct task_struct *p;
1413 
1414 	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1415 		int ret = wait_consider_task(wo, 1, p);
1416 
1417 		if (ret)
1418 			return ret;
1419 	}
1420 
1421 	return 0;
1422 }
1423 
child_wait_callback(wait_queue_entry_t * wait,unsigned mode,int sync,void * key)1424 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1425 				int sync, void *key)
1426 {
1427 	struct wait_opts *wo = container_of(wait, struct wait_opts,
1428 						child_wait);
1429 	struct task_struct *p = key;
1430 
1431 	if (!eligible_pid(wo, p))
1432 		return 0;
1433 
1434 	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1435 		return 0;
1436 
1437 	return default_wake_function(wait, mode, sync, key);
1438 }
1439 
__wake_up_parent(struct task_struct * p,struct task_struct * parent)1440 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1441 {
1442 	__wake_up_sync_key(&parent->signal->wait_chldexit,
1443 			   TASK_INTERRUPTIBLE, p);
1444 }
1445 
is_effectively_child(struct wait_opts * wo,bool ptrace,struct task_struct * target)1446 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1447 				 struct task_struct *target)
1448 {
1449 	struct task_struct *parent =
1450 		!ptrace ? target->real_parent : target->parent;
1451 
1452 	return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1453 				     same_thread_group(current, parent));
1454 }
1455 
1456 /*
1457  * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1458  * and tracee lists to find the target task.
1459  */
do_wait_pid(struct wait_opts * wo)1460 static int do_wait_pid(struct wait_opts *wo)
1461 {
1462 	bool ptrace;
1463 	struct task_struct *target;
1464 	int retval;
1465 
1466 	ptrace = false;
1467 	target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1468 	if (target && is_effectively_child(wo, ptrace, target)) {
1469 		retval = wait_consider_task(wo, ptrace, target);
1470 		if (retval)
1471 			return retval;
1472 	}
1473 
1474 	ptrace = true;
1475 	target = pid_task(wo->wo_pid, PIDTYPE_PID);
1476 	if (target && target->ptrace &&
1477 	    is_effectively_child(wo, ptrace, target)) {
1478 		retval = wait_consider_task(wo, ptrace, target);
1479 		if (retval)
1480 			return retval;
1481 	}
1482 
1483 	return 0;
1484 }
1485 
do_wait(struct wait_opts * wo)1486 static long do_wait(struct wait_opts *wo)
1487 {
1488 	int retval;
1489 
1490 	trace_sched_process_wait(wo->wo_pid);
1491 
1492 	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1493 	wo->child_wait.private = current;
1494 	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1495 repeat:
1496 	/*
1497 	 * If there is nothing that can match our criteria, just get out.
1498 	 * We will clear ->notask_error to zero if we see any child that
1499 	 * might later match our criteria, even if we are not able to reap
1500 	 * it yet.
1501 	 */
1502 	wo->notask_error = -ECHILD;
1503 	if ((wo->wo_type < PIDTYPE_MAX) &&
1504 	   (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1505 		goto notask;
1506 
1507 	set_current_state(TASK_INTERRUPTIBLE);
1508 	read_lock(&tasklist_lock);
1509 
1510 	if (wo->wo_type == PIDTYPE_PID) {
1511 		retval = do_wait_pid(wo);
1512 		if (retval)
1513 			goto end;
1514 	} else {
1515 		struct task_struct *tsk = current;
1516 
1517 		do {
1518 			retval = do_wait_thread(wo, tsk);
1519 			if (retval)
1520 				goto end;
1521 
1522 			retval = ptrace_do_wait(wo, tsk);
1523 			if (retval)
1524 				goto end;
1525 
1526 			if (wo->wo_flags & __WNOTHREAD)
1527 				break;
1528 		} while_each_thread(current, tsk);
1529 	}
1530 	read_unlock(&tasklist_lock);
1531 
1532 notask:
1533 	retval = wo->notask_error;
1534 	if (!retval && !(wo->wo_flags & WNOHANG)) {
1535 		retval = -ERESTARTSYS;
1536 		if (!signal_pending(current)) {
1537 			schedule();
1538 			goto repeat;
1539 		}
1540 	}
1541 end:
1542 	__set_current_state(TASK_RUNNING);
1543 	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1544 	return retval;
1545 }
1546 
kernel_waitid(int which,pid_t upid,struct waitid_info * infop,int options,struct rusage * ru)1547 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1548 			  int options, struct rusage *ru)
1549 {
1550 	struct wait_opts wo;
1551 	struct pid *pid = NULL;
1552 	enum pid_type type;
1553 	long ret;
1554 	unsigned int f_flags = 0;
1555 
1556 	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1557 			__WNOTHREAD|__WCLONE|__WALL))
1558 		return -EINVAL;
1559 	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1560 		return -EINVAL;
1561 
1562 	switch (which) {
1563 	case P_ALL:
1564 		type = PIDTYPE_MAX;
1565 		break;
1566 	case P_PID:
1567 		type = PIDTYPE_PID;
1568 		if (upid <= 0)
1569 			return -EINVAL;
1570 
1571 		pid = find_get_pid(upid);
1572 		break;
1573 	case P_PGID:
1574 		type = PIDTYPE_PGID;
1575 		if (upid < 0)
1576 			return -EINVAL;
1577 
1578 		if (upid)
1579 			pid = find_get_pid(upid);
1580 		else
1581 			pid = get_task_pid(current, PIDTYPE_PGID);
1582 		break;
1583 	case P_PIDFD:
1584 		type = PIDTYPE_PID;
1585 		if (upid < 0)
1586 			return -EINVAL;
1587 
1588 		pid = pidfd_get_pid(upid, &f_flags);
1589 		if (IS_ERR(pid))
1590 			return PTR_ERR(pid);
1591 
1592 		break;
1593 	default:
1594 		return -EINVAL;
1595 	}
1596 
1597 	wo.wo_type	= type;
1598 	wo.wo_pid	= pid;
1599 	wo.wo_flags	= options;
1600 	wo.wo_info	= infop;
1601 	wo.wo_rusage	= ru;
1602 	if (f_flags & O_NONBLOCK)
1603 		wo.wo_flags |= WNOHANG;
1604 
1605 	ret = do_wait(&wo);
1606 	if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1607 		ret = -EAGAIN;
1608 
1609 	put_pid(pid);
1610 	return ret;
1611 }
1612 
SYSCALL_DEFINE5(waitid,int,which,pid_t,upid,struct siginfo __user *,infop,int,options,struct rusage __user *,ru)1613 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1614 		infop, int, options, struct rusage __user *, ru)
1615 {
1616 	struct rusage r;
1617 	struct waitid_info info = {.status = 0};
1618 	long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1619 	int signo = 0;
1620 
1621 	if (err > 0) {
1622 		signo = SIGCHLD;
1623 		err = 0;
1624 		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1625 			return -EFAULT;
1626 	}
1627 	if (!infop)
1628 		return err;
1629 
1630 	if (!user_write_access_begin(infop, sizeof(*infop)))
1631 		return -EFAULT;
1632 
1633 	unsafe_put_user(signo, &infop->si_signo, Efault);
1634 	unsafe_put_user(0, &infop->si_errno, Efault);
1635 	unsafe_put_user(info.cause, &infop->si_code, Efault);
1636 	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1637 	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1638 	unsafe_put_user(info.status, &infop->si_status, Efault);
1639 	user_write_access_end();
1640 	return err;
1641 Efault:
1642 	user_write_access_end();
1643 	return -EFAULT;
1644 }
1645 
kernel_wait4(pid_t upid,int __user * stat_addr,int options,struct rusage * ru)1646 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1647 		  struct rusage *ru)
1648 {
1649 	struct wait_opts wo;
1650 	struct pid *pid = NULL;
1651 	enum pid_type type;
1652 	long ret;
1653 
1654 	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1655 			__WNOTHREAD|__WCLONE|__WALL))
1656 		return -EINVAL;
1657 
1658 	/* -INT_MIN is not defined */
1659 	if (upid == INT_MIN)
1660 		return -ESRCH;
1661 
1662 	if (upid == -1)
1663 		type = PIDTYPE_MAX;
1664 	else if (upid < 0) {
1665 		type = PIDTYPE_PGID;
1666 		pid = find_get_pid(-upid);
1667 	} else if (upid == 0) {
1668 		type = PIDTYPE_PGID;
1669 		pid = get_task_pid(current, PIDTYPE_PGID);
1670 	} else /* upid > 0 */ {
1671 		type = PIDTYPE_PID;
1672 		pid = find_get_pid(upid);
1673 	}
1674 
1675 	wo.wo_type	= type;
1676 	wo.wo_pid	= pid;
1677 	wo.wo_flags	= options | WEXITED;
1678 	wo.wo_info	= NULL;
1679 	wo.wo_stat	= 0;
1680 	wo.wo_rusage	= ru;
1681 	ret = do_wait(&wo);
1682 	put_pid(pid);
1683 	if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1684 		ret = -EFAULT;
1685 
1686 	return ret;
1687 }
1688 
kernel_wait(pid_t pid,int * stat)1689 int kernel_wait(pid_t pid, int *stat)
1690 {
1691 	struct wait_opts wo = {
1692 		.wo_type	= PIDTYPE_PID,
1693 		.wo_pid		= find_get_pid(pid),
1694 		.wo_flags	= WEXITED,
1695 	};
1696 	int ret;
1697 
1698 	ret = do_wait(&wo);
1699 	if (ret > 0 && wo.wo_stat)
1700 		*stat = wo.wo_stat;
1701 	put_pid(wo.wo_pid);
1702 	return ret;
1703 }
1704 
SYSCALL_DEFINE4(wait4,pid_t,upid,int __user *,stat_addr,int,options,struct rusage __user *,ru)1705 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1706 		int, options, struct rusage __user *, ru)
1707 {
1708 	struct rusage r;
1709 	long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1710 
1711 	if (err > 0) {
1712 		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1713 			return -EFAULT;
1714 	}
1715 	return err;
1716 }
1717 
1718 #ifdef __ARCH_WANT_SYS_WAITPID
1719 
1720 /*
1721  * sys_waitpid() remains for compatibility. waitpid() should be
1722  * implemented by calling sys_wait4() from libc.a.
1723  */
SYSCALL_DEFINE3(waitpid,pid_t,pid,int __user *,stat_addr,int,options)1724 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1725 {
1726 	return kernel_wait4(pid, stat_addr, options, NULL);
1727 }
1728 
1729 #endif
1730 
1731 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(wait4,compat_pid_t,pid,compat_uint_t __user *,stat_addr,int,options,struct compat_rusage __user *,ru)1732 COMPAT_SYSCALL_DEFINE4(wait4,
1733 	compat_pid_t, pid,
1734 	compat_uint_t __user *, stat_addr,
1735 	int, options,
1736 	struct compat_rusage __user *, ru)
1737 {
1738 	struct rusage r;
1739 	long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1740 	if (err > 0) {
1741 		if (ru && put_compat_rusage(&r, ru))
1742 			return -EFAULT;
1743 	}
1744 	return err;
1745 }
1746 
COMPAT_SYSCALL_DEFINE5(waitid,int,which,compat_pid_t,pid,struct compat_siginfo __user *,infop,int,options,struct compat_rusage __user *,uru)1747 COMPAT_SYSCALL_DEFINE5(waitid,
1748 		int, which, compat_pid_t, pid,
1749 		struct compat_siginfo __user *, infop, int, options,
1750 		struct compat_rusage __user *, uru)
1751 {
1752 	struct rusage ru;
1753 	struct waitid_info info = {.status = 0};
1754 	long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1755 	int signo = 0;
1756 	if (err > 0) {
1757 		signo = SIGCHLD;
1758 		err = 0;
1759 		if (uru) {
1760 			/* kernel_waitid() overwrites everything in ru */
1761 			if (COMPAT_USE_64BIT_TIME)
1762 				err = copy_to_user(uru, &ru, sizeof(ru));
1763 			else
1764 				err = put_compat_rusage(&ru, uru);
1765 			if (err)
1766 				return -EFAULT;
1767 		}
1768 	}
1769 
1770 	if (!infop)
1771 		return err;
1772 
1773 	if (!user_write_access_begin(infop, sizeof(*infop)))
1774 		return -EFAULT;
1775 
1776 	unsafe_put_user(signo, &infop->si_signo, Efault);
1777 	unsafe_put_user(0, &infop->si_errno, Efault);
1778 	unsafe_put_user(info.cause, &infop->si_code, Efault);
1779 	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1780 	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1781 	unsafe_put_user(info.status, &infop->si_status, Efault);
1782 	user_write_access_end();
1783 	return err;
1784 Efault:
1785 	user_write_access_end();
1786 	return -EFAULT;
1787 }
1788 #endif
1789 
1790 /**
1791  * thread_group_exited - check that a thread group has exited
1792  * @pid: tgid of thread group to be checked.
1793  *
1794  * Test if the thread group represented by tgid has exited (all
1795  * threads are zombies, dead or completely gone).
1796  *
1797  * Return: true if the thread group has exited. false otherwise.
1798  */
thread_group_exited(struct pid * pid)1799 bool thread_group_exited(struct pid *pid)
1800 {
1801 	struct task_struct *task;
1802 	bool exited;
1803 
1804 	rcu_read_lock();
1805 	task = pid_task(pid, PIDTYPE_PID);
1806 	exited = !task ||
1807 		(READ_ONCE(task->exit_state) && thread_group_empty(task));
1808 	rcu_read_unlock();
1809 
1810 	return exited;
1811 }
1812 EXPORT_SYMBOL(thread_group_exited);
1813 
abort(void)1814 __weak void abort(void)
1815 {
1816 	BUG();
1817 
1818 	/* if that doesn't kill us, halt */
1819 	panic("Oops failed to kill thread");
1820 }
1821 EXPORT_SYMBOL(abort);
1822