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/kmsan.h>
  64#include <linux/random.h>
  65#include <linux/rcuwait.h>
  66#include <linux/compat.h>
  67#include <linux/io_uring.h>
  68#include <linux/kprobes.h>
  69#include <linux/rethook.h>
  70#include <linux/sysfs.h>
  71
  72#include <linux/uaccess.h>
  73#include <asm/unistd.h>
  74#include <asm/mmu_context.h>
  75
  76/*
  77 * The default value should be high enough to not crash a system that randomly
  78 * crashes its kernel from time to time, but low enough to at least not permit
  79 * overflowing 32-bit refcounts or the ldsem writer count.
  80 */
  81static unsigned int oops_limit = 10000;
  82
  83#ifdef CONFIG_SYSCTL
  84static struct ctl_table kern_exit_table[] = {
  85	{
  86		.procname       = "oops_limit",
  87		.data           = &oops_limit,
  88		.maxlen         = sizeof(oops_limit),
  89		.mode           = 0644,
  90		.proc_handler   = proc_douintvec,
  91	},
  92	{ }
  93};
  94
  95static __init int kernel_exit_sysctls_init(void)
  96{
  97	register_sysctl_init("kernel", kern_exit_table);
  98	return 0;
  99}
 100late_initcall(kernel_exit_sysctls_init);
 101#endif
 102
 103static atomic_t oops_count = ATOMIC_INIT(0);
 104
 105#ifdef CONFIG_SYSFS
 106static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
 107			       char *page)
 108{
 109	return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
 110}
 111
 112static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
 113
 114static __init int kernel_exit_sysfs_init(void)
 115{
 116	sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
 117	return 0;
 118}
 119late_initcall(kernel_exit_sysfs_init);
 120#endif
 121
 122static void __unhash_process(struct task_struct *p, bool group_dead)
 123{
 124	nr_threads--;
 125	detach_pid(p, PIDTYPE_PID);
 126	if (group_dead) {
 127		detach_pid(p, PIDTYPE_TGID);
 128		detach_pid(p, PIDTYPE_PGID);
 129		detach_pid(p, PIDTYPE_SID);
 130
 131		list_del_rcu(&p->tasks);
 132		list_del_init(&p->sibling);
 133		__this_cpu_dec(process_counts);
 134	}
 135	list_del_rcu(&p->thread_group);
 136	list_del_rcu(&p->thread_node);
 137}
 138
 139/*
 140 * This function expects the tasklist_lock write-locked.
 141 */
 142static void __exit_signal(struct task_struct *tsk)
 143{
 144	struct signal_struct *sig = tsk->signal;
 145	bool group_dead = thread_group_leader(tsk);
 146	struct sighand_struct *sighand;
 147	struct tty_struct *tty;
 148	u64 utime, stime;
 149
 150	sighand = rcu_dereference_check(tsk->sighand,
 151					lockdep_tasklist_lock_is_held());
 152	spin_lock(&sighand->siglock);
 153
 154#ifdef CONFIG_POSIX_TIMERS
 155	posix_cpu_timers_exit(tsk);
 156	if (group_dead)
 157		posix_cpu_timers_exit_group(tsk);
 158#endif
 159
 160	if (group_dead) {
 161		tty = sig->tty;
 162		sig->tty = NULL;
 163	} else {
 164		/*
 165		 * If there is any task waiting for the group exit
 166		 * then notify it:
 167		 */
 168		if (sig->notify_count > 0 && !--sig->notify_count)
 169			wake_up_process(sig->group_exec_task);
 170
 171		if (tsk == sig->curr_target)
 172			sig->curr_target = next_thread(tsk);
 173	}
 174
 175	add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 176			      sizeof(unsigned long long));
 177
 178	/*
 179	 * Accumulate here the counters for all threads as they die. We could
 180	 * skip the group leader because it is the last user of signal_struct,
 181	 * but we want to avoid the race with thread_group_cputime() which can
 182	 * see the empty ->thread_head list.
 183	 */
 184	task_cputime(tsk, &utime, &stime);
 185	write_seqlock(&sig->stats_lock);
 186	sig->utime += utime;
 187	sig->stime += stime;
 188	sig->gtime += task_gtime(tsk);
 189	sig->min_flt += tsk->min_flt;
 190	sig->maj_flt += tsk->maj_flt;
 191	sig->nvcsw += tsk->nvcsw;
 192	sig->nivcsw += tsk->nivcsw;
 193	sig->inblock += task_io_get_inblock(tsk);
 194	sig->oublock += task_io_get_oublock(tsk);
 195	task_io_accounting_add(&sig->ioac, &tsk->ioac);
 196	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 197	sig->nr_threads--;
 198	__unhash_process(tsk, group_dead);
 199	write_sequnlock(&sig->stats_lock);
 200
 201	/*
 202	 * Do this under ->siglock, we can race with another thread
 203	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
 204	 */
 205	flush_sigqueue(&tsk->pending);
 206	tsk->sighand = NULL;
 207	spin_unlock(&sighand->siglock);
 208
 209	__cleanup_sighand(sighand);
 210	clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
 211	if (group_dead) {
 212		flush_sigqueue(&sig->shared_pending);
 213		tty_kref_put(tty);
 214	}
 215}
 216
 217static void delayed_put_task_struct(struct rcu_head *rhp)
 218{
 219	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
 220
 221	kprobe_flush_task(tsk);
 222	rethook_flush_task(tsk);
 223	perf_event_delayed_put(tsk);
 224	trace_sched_process_free(tsk);
 225	put_task_struct(tsk);
 226}
 227
 228void put_task_struct_rcu_user(struct task_struct *task)
 229{
 230	if (refcount_dec_and_test(&task->rcu_users))
 231		call_rcu(&task->rcu, delayed_put_task_struct);
 232}
 233
 234void __weak release_thread(struct task_struct *dead_task)
 235{
 236}
 237
 238void release_task(struct task_struct *p)
 239{
 240	struct task_struct *leader;
 241	struct pid *thread_pid;
 242	int zap_leader;
 243repeat:
 244	/* don't need to get the RCU readlock here - the process is dead and
 245	 * can't be modifying its own credentials. But shut RCU-lockdep up */
 246	rcu_read_lock();
 247	dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
 248	rcu_read_unlock();
 249
 250	cgroup_release(p);
 251
 252	write_lock_irq(&tasklist_lock);
 253	ptrace_release_task(p);
 254	thread_pid = get_pid(p->thread_pid);
 255	__exit_signal(p);
 256
 257	/*
 258	 * If we are the last non-leader member of the thread
 259	 * group, and the leader is zombie, then notify the
 260	 * group leader's parent process. (if it wants notification.)
 261	 */
 262	zap_leader = 0;
 263	leader = p->group_leader;
 264	if (leader != p && thread_group_empty(leader)
 265			&& leader->exit_state == EXIT_ZOMBIE) {
 266		/*
 267		 * If we were the last child thread and the leader has
 268		 * exited already, and the leader's parent ignores SIGCHLD,
 269		 * then we are the one who should release the leader.
 270		 */
 271		zap_leader = do_notify_parent(leader, leader->exit_signal);
 272		if (zap_leader)
 273			leader->exit_state = EXIT_DEAD;
 274	}
 275
 276	write_unlock_irq(&tasklist_lock);
 277	seccomp_filter_release(p);
 278	proc_flush_pid(thread_pid);
 279	put_pid(thread_pid);
 280	release_thread(p);
 281	put_task_struct_rcu_user(p);
 282
 283	p = leader;
 284	if (unlikely(zap_leader))
 285		goto repeat;
 286}
 287
 288int rcuwait_wake_up(struct rcuwait *w)
 289{
 290	int ret = 0;
 291	struct task_struct *task;
 292
 293	rcu_read_lock();
 294
 295	/*
 296	 * Order condition vs @task, such that everything prior to the load
 297	 * of @task is visible. This is the condition as to why the user called
 298	 * rcuwait_wake() in the first place. Pairs with set_current_state()
 299	 * barrier (A) in rcuwait_wait_event().
 300	 *
 301	 *    WAIT                WAKE
 302	 *    [S] tsk = current	  [S] cond = true
 303	 *        MB (A)	      MB (B)
 304	 *    [L] cond		  [L] tsk
 305	 */
 306	smp_mb(); /* (B) */
 307
 308	task = rcu_dereference(w->task);
 309	if (task)
 310		ret = wake_up_process(task);
 311	rcu_read_unlock();
 312
 313	return ret;
 314}
 315EXPORT_SYMBOL_GPL(rcuwait_wake_up);
 316
 317/*
 318 * Determine if a process group is "orphaned", according to the POSIX
 319 * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 320 * by terminal-generated stop signals.  Newly orphaned process groups are
 321 * to receive a SIGHUP and a SIGCONT.
 322 *
 323 * "I ask you, have you ever known what it is to be an orphan?"
 324 */
 325static int will_become_orphaned_pgrp(struct pid *pgrp,
 326					struct task_struct *ignored_task)
 327{
 328	struct task_struct *p;
 329
 330	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 331		if ((p == ignored_task) ||
 332		    (p->exit_state && thread_group_empty(p)) ||
 333		    is_global_init(p->real_parent))
 334			continue;
 335
 336		if (task_pgrp(p->real_parent) != pgrp &&
 337		    task_session(p->real_parent) == task_session(p))
 338			return 0;
 339	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 340
 341	return 1;
 342}
 343
 344int is_current_pgrp_orphaned(void)
 345{
 346	int retval;
 347
 348	read_lock(&tasklist_lock);
 349	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
 350	read_unlock(&tasklist_lock);
 351
 352	return retval;
 353}
 354
 355static bool has_stopped_jobs(struct pid *pgrp)
 356{
 357	struct task_struct *p;
 358
 359	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 360		if (p->signal->flags & SIGNAL_STOP_STOPPED)
 361			return true;
 362	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 363
 364	return false;
 365}
 366
 367/*
 368 * Check to see if any process groups have become orphaned as
 369 * a result of our exiting, and if they have any stopped jobs,
 370 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
 371 */
 372static void
 373kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
 374{
 375	struct pid *pgrp = task_pgrp(tsk);
 376	struct task_struct *ignored_task = tsk;
 377
 378	if (!parent)
 379		/* exit: our father is in a different pgrp than
 380		 * we are and we were the only connection outside.
 381		 */
 382		parent = tsk->real_parent;
 383	else
 384		/* reparent: our child is in a different pgrp than
 385		 * we are, and it was the only connection outside.
 386		 */
 387		ignored_task = NULL;
 388
 389	if (task_pgrp(parent) != pgrp &&
 390	    task_session(parent) == task_session(tsk) &&
 391	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
 392	    has_stopped_jobs(pgrp)) {
 393		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 394		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 395	}
 396}
 397
 398static void coredump_task_exit(struct task_struct *tsk)
 399{
 400	struct core_state *core_state;
 401
 402	/*
 403	 * Serialize with any possible pending coredump.
 404	 * We must hold siglock around checking core_state
 405	 * and setting PF_POSTCOREDUMP.  The core-inducing thread
 406	 * will increment ->nr_threads for each thread in the
 407	 * group without PF_POSTCOREDUMP set.
 408	 */
 409	spin_lock_irq(&tsk->sighand->siglock);
 410	tsk->flags |= PF_POSTCOREDUMP;
 411	core_state = tsk->signal->core_state;
 412	spin_unlock_irq(&tsk->sighand->siglock);
 413	if (core_state) {
 414		struct core_thread self;
 415
 416		self.task = current;
 417		if (self.task->flags & PF_SIGNALED)
 418			self.next = xchg(&core_state->dumper.next, &self);
 419		else
 420			self.task = NULL;
 421		/*
 422		 * Implies mb(), the result of xchg() must be visible
 423		 * to core_state->dumper.
 424		 */
 425		if (atomic_dec_and_test(&core_state->nr_threads))
 426			complete(&core_state->startup);
 427
 428		for (;;) {
 429			set_current_state(TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
 430			if (!self.task) /* see coredump_finish() */
 431				break;
 432			schedule();
 433		}
 434		__set_current_state(TASK_RUNNING);
 435	}
 436}
 437
 438#ifdef CONFIG_MEMCG
 439/*
 440 * A task is exiting.   If it owned this mm, find a new owner for the mm.
 441 */
 442void mm_update_next_owner(struct mm_struct *mm)
 443{
 444	struct task_struct *c, *g, *p = current;
 445
 446retry:
 447	/*
 448	 * If the exiting or execing task is not the owner, it's
 449	 * someone else's problem.
 450	 */
 451	if (mm->owner != p)
 452		return;
 453	/*
 454	 * The current owner is exiting/execing and there are no other
 455	 * candidates.  Do not leave the mm pointing to a possibly
 456	 * freed task structure.
 457	 */
 458	if (atomic_read(&mm->mm_users) <= 1) {
 459		WRITE_ONCE(mm->owner, NULL);
 460		return;
 461	}
 462
 463	read_lock(&tasklist_lock);
 464	/*
 465	 * Search in the children
 466	 */
 467	list_for_each_entry(c, &p->children, sibling) {
 468		if (c->mm == mm)
 469			goto assign_new_owner;
 470	}
 471
 472	/*
 473	 * Search in the siblings
 474	 */
 475	list_for_each_entry(c, &p->real_parent->children, sibling) {
 476		if (c->mm == mm)
 477			goto assign_new_owner;
 478	}
 479
 480	/*
 481	 * Search through everything else, we should not get here often.
 482	 */
 483	for_each_process(g) {
 484		if (g->flags & PF_KTHREAD)
 485			continue;
 486		for_each_thread(g, c) {
 487			if (c->mm == mm)
 488				goto assign_new_owner;
 489			if (c->mm)
 490				break;
 491		}
 492	}
 493	read_unlock(&tasklist_lock);
 494	/*
 495	 * We found no owner yet mm_users > 1: this implies that we are
 496	 * most likely racing with swapoff (try_to_unuse()) or /proc or
 497	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
 498	 */
 499	WRITE_ONCE(mm->owner, NULL);
 500	return;
 501
 502assign_new_owner:
 503	BUG_ON(c == p);
 504	get_task_struct(c);
 505	/*
 506	 * The task_lock protects c->mm from changing.
 507	 * We always want mm->owner->mm == mm
 508	 */
 509	task_lock(c);
 510	/*
 511	 * Delay read_unlock() till we have the task_lock()
 512	 * to ensure that c does not slip away underneath us
 513	 */
 514	read_unlock(&tasklist_lock);
 515	if (c->mm != mm) {
 516		task_unlock(c);
 517		put_task_struct(c);
 518		goto retry;
 519	}
 520	WRITE_ONCE(mm->owner, c);
 521	lru_gen_migrate_mm(mm);
 522	task_unlock(c);
 523	put_task_struct(c);
 524}
 525#endif /* CONFIG_MEMCG */
 526
 527/*
 528 * Turn us into a lazy TLB process if we
 529 * aren't already..
 530 */
 531static void exit_mm(void)
 532{
 533	struct mm_struct *mm = current->mm;
 
 534
 535	exit_mm_release(current, mm);
 536	if (!mm)
 537		return;
 538	sync_mm_rss(mm);
 
 
 
 
 
 
 
 539	mmap_read_lock(mm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 540	mmgrab(mm);
 541	BUG_ON(mm != current->active_mm);
 542	/* more a memory barrier than a real lock */
 543	task_lock(current);
 544	/*
 545	 * When a thread stops operating on an address space, the loop
 546	 * in membarrier_private_expedited() may not observe that
 547	 * tsk->mm, and the loop in membarrier_global_expedited() may
 548	 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
 549	 * rq->membarrier_state, so those would not issue an IPI.
 550	 * Membarrier requires a memory barrier after accessing
 551	 * user-space memory, before clearing tsk->mm or the
 552	 * rq->membarrier_state.
 553	 */
 554	smp_mb__after_spinlock();
 555	local_irq_disable();
 556	current->mm = NULL;
 557	membarrier_update_current_mm(NULL);
 558	enter_lazy_tlb(mm, current);
 559	local_irq_enable();
 560	task_unlock(current);
 561	mmap_read_unlock(mm);
 562	mm_update_next_owner(mm);
 563	mmput(mm);
 564	if (test_thread_flag(TIF_MEMDIE))
 565		exit_oom_victim();
 566}
 567
 568static struct task_struct *find_alive_thread(struct task_struct *p)
 569{
 570	struct task_struct *t;
 571
 572	for_each_thread(p, t) {
 573		if (!(t->flags & PF_EXITING))
 574			return t;
 575	}
 576	return NULL;
 577}
 578
 579static struct task_struct *find_child_reaper(struct task_struct *father,
 580						struct list_head *dead)
 581	__releases(&tasklist_lock)
 582	__acquires(&tasklist_lock)
 583{
 584	struct pid_namespace *pid_ns = task_active_pid_ns(father);
 585	struct task_struct *reaper = pid_ns->child_reaper;
 586	struct task_struct *p, *n;
 587
 588	if (likely(reaper != father))
 589		return reaper;
 590
 591	reaper = find_alive_thread(father);
 592	if (reaper) {
 593		pid_ns->child_reaper = reaper;
 594		return reaper;
 595	}
 596
 597	write_unlock_irq(&tasklist_lock);
 598
 599	list_for_each_entry_safe(p, n, dead, ptrace_entry) {
 600		list_del_init(&p->ptrace_entry);
 601		release_task(p);
 602	}
 603
 604	zap_pid_ns_processes(pid_ns);
 605	write_lock_irq(&tasklist_lock);
 606
 607	return father;
 608}
 609
 610/*
 611 * When we die, we re-parent all our children, and try to:
 612 * 1. give them to another thread in our thread group, if such a member exists
 613 * 2. give it to the first ancestor process which prctl'd itself as a
 614 *    child_subreaper for its children (like a service manager)
 615 * 3. give it to the init process (PID 1) in our pid namespace
 616 */
 617static struct task_struct *find_new_reaper(struct task_struct *father,
 618					   struct task_struct *child_reaper)
 619{
 620	struct task_struct *thread, *reaper;
 621
 622	thread = find_alive_thread(father);
 623	if (thread)
 624		return thread;
 625
 626	if (father->signal->has_child_subreaper) {
 627		unsigned int ns_level = task_pid(father)->level;
 628		/*
 629		 * Find the first ->is_child_subreaper ancestor in our pid_ns.
 630		 * We can't check reaper != child_reaper to ensure we do not
 631		 * cross the namespaces, the exiting parent could be injected
 632		 * by setns() + fork().
 633		 * We check pid->level, this is slightly more efficient than
 634		 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
 635		 */
 636		for (reaper = father->real_parent;
 637		     task_pid(reaper)->level == ns_level;
 638		     reaper = reaper->real_parent) {
 639			if (reaper == &init_task)
 640				break;
 641			if (!reaper->signal->is_child_subreaper)
 642				continue;
 643			thread = find_alive_thread(reaper);
 644			if (thread)
 645				return thread;
 646		}
 647	}
 648
 649	return child_reaper;
 650}
 651
 652/*
 653* Any that need to be release_task'd are put on the @dead list.
 654 */
 655static void reparent_leader(struct task_struct *father, struct task_struct *p,
 656				struct list_head *dead)
 657{
 658	if (unlikely(p->exit_state == EXIT_DEAD))
 659		return;
 660
 661	/* We don't want people slaying init. */
 662	p->exit_signal = SIGCHLD;
 663
 664	/* If it has exited notify the new parent about this child's death. */
 665	if (!p->ptrace &&
 666	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
 667		if (do_notify_parent(p, p->exit_signal)) {
 668			p->exit_state = EXIT_DEAD;
 669			list_add(&p->ptrace_entry, dead);
 670		}
 671	}
 672
 673	kill_orphaned_pgrp(p, father);
 674}
 675
 676/*
 677 * This does two things:
 678 *
 679 * A.  Make init inherit all the child processes
 680 * B.  Check to see if any process groups have become orphaned
 681 *	as a result of our exiting, and if they have any stopped
 682 *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 683 */
 684static void forget_original_parent(struct task_struct *father,
 685					struct list_head *dead)
 686{
 687	struct task_struct *p, *t, *reaper;
 688
 689	if (unlikely(!list_empty(&father->ptraced)))
 690		exit_ptrace(father, dead);
 691
 692	/* Can drop and reacquire tasklist_lock */
 693	reaper = find_child_reaper(father, dead);
 694	if (list_empty(&father->children))
 695		return;
 696
 697	reaper = find_new_reaper(father, reaper);
 698	list_for_each_entry(p, &father->children, sibling) {
 699		for_each_thread(p, t) {
 700			RCU_INIT_POINTER(t->real_parent, reaper);
 701			BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
 702			if (likely(!t->ptrace))
 703				t->parent = t->real_parent;
 704			if (t->pdeath_signal)
 705				group_send_sig_info(t->pdeath_signal,
 706						    SEND_SIG_NOINFO, t,
 707						    PIDTYPE_TGID);
 708		}
 709		/*
 710		 * If this is a threaded reparent there is no need to
 711		 * notify anyone anything has happened.
 712		 */
 713		if (!same_thread_group(reaper, father))
 714			reparent_leader(father, p, dead);
 715	}
 716	list_splice_tail_init(&father->children, &reaper->children);
 717}
 718
 719/*
 720 * Send signals to all our closest relatives so that they know
 721 * to properly mourn us..
 722 */
 723static void exit_notify(struct task_struct *tsk, int group_dead)
 724{
 725	bool autoreap;
 726	struct task_struct *p, *n;
 727	LIST_HEAD(dead);
 728
 729	write_lock_irq(&tasklist_lock);
 730	forget_original_parent(tsk, &dead);
 731
 732	if (group_dead)
 733		kill_orphaned_pgrp(tsk->group_leader, NULL);
 734
 735	tsk->exit_state = EXIT_ZOMBIE;
 736	if (unlikely(tsk->ptrace)) {
 737		int sig = thread_group_leader(tsk) &&
 738				thread_group_empty(tsk) &&
 739				!ptrace_reparented(tsk) ?
 740			tsk->exit_signal : SIGCHLD;
 741		autoreap = do_notify_parent(tsk, sig);
 742	} else if (thread_group_leader(tsk)) {
 743		autoreap = thread_group_empty(tsk) &&
 744			do_notify_parent(tsk, tsk->exit_signal);
 745	} else {
 746		autoreap = true;
 747	}
 748
 749	if (autoreap) {
 750		tsk->exit_state = EXIT_DEAD;
 751		list_add(&tsk->ptrace_entry, &dead);
 752	}
 753
 754	/* mt-exec, de_thread() is waiting for group leader */
 755	if (unlikely(tsk->signal->notify_count < 0))
 756		wake_up_process(tsk->signal->group_exec_task);
 757	write_unlock_irq(&tasklist_lock);
 758
 759	list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
 760		list_del_init(&p->ptrace_entry);
 761		release_task(p);
 762	}
 763}
 764
 765#ifdef CONFIG_DEBUG_STACK_USAGE
 766static void check_stack_usage(void)
 767{
 768	static DEFINE_SPINLOCK(low_water_lock);
 769	static int lowest_to_date = THREAD_SIZE;
 770	unsigned long free;
 771
 772	free = stack_not_used(current);
 773
 774	if (free >= lowest_to_date)
 775		return;
 776
 777	spin_lock(&low_water_lock);
 778	if (free < lowest_to_date) {
 779		pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
 780			current->comm, task_pid_nr(current), free);
 781		lowest_to_date = free;
 782	}
 783	spin_unlock(&low_water_lock);
 784}
 785#else
 786static inline void check_stack_usage(void) {}
 787#endif
 788
 789static void synchronize_group_exit(struct task_struct *tsk, long code)
 790{
 791	struct sighand_struct *sighand = tsk->sighand;
 792	struct signal_struct *signal = tsk->signal;
 793
 794	spin_lock_irq(&sighand->siglock);
 795	signal->quick_threads--;
 796	if ((signal->quick_threads == 0) &&
 797	    !(signal->flags & SIGNAL_GROUP_EXIT)) {
 798		signal->flags = SIGNAL_GROUP_EXIT;
 799		signal->group_exit_code = code;
 800		signal->group_stop_count = 0;
 801	}
 802	spin_unlock_irq(&sighand->siglock);
 803}
 804
 805void __noreturn do_exit(long code)
 806{
 807	struct task_struct *tsk = current;
 808	int group_dead;
 809
 810	synchronize_group_exit(tsk, code);
 
 
 
 
 
 811
 812	WARN_ON(tsk->plug);
 813
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 814	kcov_task_exit(tsk);
 815	kmsan_task_exit(tsk);
 816
 817	coredump_task_exit(tsk);
 818	ptrace_event(PTRACE_EVENT_EXIT, code);
 819
 820	validate_creds_for_do_exit(tsk);
 821
 822	io_uring_files_cancel();
 
 
 
 
 
 
 
 
 
 
 
 823	exit_signals(tsk);  /* sets PF_EXITING */
 824
 825	/* sync mm's RSS info before statistics gathering */
 826	if (tsk->mm)
 827		sync_mm_rss(tsk->mm);
 828	acct_update_integrals(tsk);
 829	group_dead = atomic_dec_and_test(&tsk->signal->live);
 830	if (group_dead) {
 831		/*
 832		 * If the last thread of global init has exited, panic
 833		 * immediately to get a useable coredump.
 834		 */
 835		if (unlikely(is_global_init(tsk)))
 836			panic("Attempted to kill init! exitcode=0x%08x\n",
 837				tsk->signal->group_exit_code ?: (int)code);
 838
 839#ifdef CONFIG_POSIX_TIMERS
 840		hrtimer_cancel(&tsk->signal->real_timer);
 841		exit_itimers(tsk);
 842#endif
 843		if (tsk->mm)
 844			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
 845	}
 846	acct_collect(code, group_dead);
 847	if (group_dead)
 848		tty_audit_exit();
 849	audit_free(tsk);
 850
 851	tsk->exit_code = code;
 852	taskstats_exit(tsk, group_dead);
 853
 854	exit_mm();
 855
 856	if (group_dead)
 857		acct_process();
 858	trace_sched_process_exit(tsk);
 859
 860	exit_sem(tsk);
 861	exit_shm(tsk);
 862	exit_files(tsk);
 863	exit_fs(tsk);
 864	if (group_dead)
 865		disassociate_ctty(1);
 866	exit_task_namespaces(tsk);
 867	exit_task_work(tsk);
 868	exit_thread(tsk);
 869
 870	/*
 871	 * Flush inherited counters to the parent - before the parent
 872	 * gets woken up by child-exit notifications.
 873	 *
 874	 * because of cgroup mode, must be called before cgroup_exit()
 875	 */
 876	perf_event_exit_task(tsk);
 877
 878	sched_autogroup_exit_task(tsk);
 879	cgroup_exit(tsk);
 880
 881	/*
 882	 * FIXME: do that only when needed, using sched_exit tracepoint
 883	 */
 884	flush_ptrace_hw_breakpoint(tsk);
 885
 886	exit_tasks_rcu_start();
 887	exit_notify(tsk, group_dead);
 888	proc_exit_connector(tsk);
 889	mpol_put_task_policy(tsk);
 890#ifdef CONFIG_FUTEX
 891	if (unlikely(current->pi_state_cache))
 892		kfree(current->pi_state_cache);
 893#endif
 894	/*
 895	 * Make sure we are holding no locks:
 896	 */
 897	debug_check_no_locks_held();
 898
 899	if (tsk->io_context)
 900		exit_io_context(tsk);
 901
 902	if (tsk->splice_pipe)
 903		free_pipe_info(tsk->splice_pipe);
 904
 905	if (tsk->task_frag.page)
 906		put_page(tsk->task_frag.page);
 907
 908	validate_creds_for_do_exit(tsk);
 909	exit_task_stack_account(tsk);
 910
 911	check_stack_usage();
 912	preempt_disable();
 913	if (tsk->nr_dirtied)
 914		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
 915	exit_rcu();
 916	exit_tasks_rcu_finish();
 917
 918	lockdep_free_task(tsk);
 919	do_task_dead();
 920}
 
 921
 922void __noreturn make_task_dead(int signr)
 923{
 924	/*
 925	 * Take the task off the cpu after something catastrophic has
 926	 * happened.
 927	 *
 928	 * We can get here from a kernel oops, sometimes with preemption off.
 929	 * Start by checking for critical errors.
 930	 * Then fix up important state like USER_DS and preemption.
 931	 * Then do everything else.
 932	 */
 933	struct task_struct *tsk = current;
 934	unsigned int limit;
 935
 936	if (unlikely(in_interrupt()))
 937		panic("Aiee, killing interrupt handler!");
 938	if (unlikely(!tsk->pid))
 939		panic("Attempted to kill the idle task!");
 940
 941	if (unlikely(in_atomic())) {
 942		pr_info("note: %s[%d] exited with preempt_count %d\n",
 943			current->comm, task_pid_nr(current),
 944			preempt_count());
 945		preempt_count_set(PREEMPT_ENABLED);
 946	}
 947
 948	/*
 949	 * Every time the system oopses, if the oops happens while a reference
 950	 * to an object was held, the reference leaks.
 951	 * If the oops doesn't also leak memory, repeated oopsing can cause
 952	 * reference counters to wrap around (if they're not using refcount_t).
 953	 * This means that repeated oopsing can make unexploitable-looking bugs
 954	 * exploitable through repeated oopsing.
 955	 * To make sure this can't happen, place an upper bound on how often the
 956	 * kernel may oops without panic().
 957	 */
 958	limit = READ_ONCE(oops_limit);
 959	if (atomic_inc_return(&oops_count) >= limit && limit)
 960		panic("Oopsed too often (kernel.oops_limit is %d)", limit);
 961
 962	/*
 963	 * We're taking recursive faults here in make_task_dead. Safest is to just
 964	 * leave this task alone and wait for reboot.
 965	 */
 966	if (unlikely(tsk->flags & PF_EXITING)) {
 967		pr_alert("Fixing recursive fault but reboot is needed!\n");
 968		futex_exit_recursive(tsk);
 969		tsk->exit_state = EXIT_DEAD;
 970		refcount_inc(&tsk->rcu_users);
 971		do_task_dead();
 972	}
 973
 974	do_exit(signr);
 975}
 
 976
 977SYSCALL_DEFINE1(exit, int, error_code)
 978{
 979	do_exit((error_code&0xff)<<8);
 980}
 981
 982/*
 983 * Take down every thread in the group.  This is called by fatal signals
 984 * as well as by sys_exit_group (below).
 985 */
 986void __noreturn
 987do_group_exit(int exit_code)
 988{
 989	struct signal_struct *sig = current->signal;
 990
 991	if (sig->flags & SIGNAL_GROUP_EXIT)
 
 
 992		exit_code = sig->group_exit_code;
 993	else if (sig->group_exec_task)
 994		exit_code = 0;
 995	else {
 996		struct sighand_struct *const sighand = current->sighand;
 997
 998		spin_lock_irq(&sighand->siglock);
 999		if (sig->flags & SIGNAL_GROUP_EXIT)
1000			/* Another thread got here before we took the lock.  */
1001			exit_code = sig->group_exit_code;
1002		else if (sig->group_exec_task)
1003			exit_code = 0;
1004		else {
1005			sig->group_exit_code = exit_code;
1006			sig->flags = SIGNAL_GROUP_EXIT;
1007			zap_other_threads(current);
1008		}
1009		spin_unlock_irq(&sighand->siglock);
1010	}
1011
1012	do_exit(exit_code);
1013	/* NOTREACHED */
1014}
1015
1016/*
1017 * this kills every thread in the thread group. Note that any externally
1018 * wait4()-ing process will get the correct exit code - even if this
1019 * thread is not the thread group leader.
1020 */
1021SYSCALL_DEFINE1(exit_group, int, error_code)
1022{
1023	do_group_exit((error_code & 0xff) << 8);
1024	/* NOTREACHED */
1025	return 0;
1026}
1027
1028struct waitid_info {
1029	pid_t pid;
1030	uid_t uid;
1031	int status;
1032	int cause;
1033};
1034
1035struct wait_opts {
1036	enum pid_type		wo_type;
1037	int			wo_flags;
1038	struct pid		*wo_pid;
1039
1040	struct waitid_info	*wo_info;
1041	int			wo_stat;
1042	struct rusage		*wo_rusage;
1043
1044	wait_queue_entry_t		child_wait;
1045	int			notask_error;
1046};
1047
1048static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1049{
1050	return	wo->wo_type == PIDTYPE_MAX ||
1051		task_pid_type(p, wo->wo_type) == wo->wo_pid;
1052}
1053
1054static int
1055eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1056{
1057	if (!eligible_pid(wo, p))
1058		return 0;
1059
1060	/*
1061	 * Wait for all children (clone and not) if __WALL is set or
1062	 * if it is traced by us.
1063	 */
1064	if (ptrace || (wo->wo_flags & __WALL))
1065		return 1;
1066
1067	/*
1068	 * Otherwise, wait for clone children *only* if __WCLONE is set;
1069	 * otherwise, wait for non-clone children *only*.
1070	 *
1071	 * Note: a "clone" child here is one that reports to its parent
1072	 * using a signal other than SIGCHLD, or a non-leader thread which
1073	 * we can only see if it is traced by us.
1074	 */
1075	if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1076		return 0;
1077
1078	return 1;
1079}
1080
1081/*
1082 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1083 * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1084 * the lock and this task is uninteresting.  If we return nonzero, we have
1085 * released the lock and the system call should return.
1086 */
1087static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1088{
1089	int state, status;
1090	pid_t pid = task_pid_vnr(p);
1091	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1092	struct waitid_info *infop;
1093
1094	if (!likely(wo->wo_flags & WEXITED))
1095		return 0;
1096
1097	if (unlikely(wo->wo_flags & WNOWAIT)) {
1098		status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1099			? p->signal->group_exit_code : p->exit_code;
1100		get_task_struct(p);
1101		read_unlock(&tasklist_lock);
1102		sched_annotate_sleep();
1103		if (wo->wo_rusage)
1104			getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1105		put_task_struct(p);
1106		goto out_info;
1107	}
1108	/*
1109	 * Move the task's state to DEAD/TRACE, only one thread can do this.
1110	 */
1111	state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1112		EXIT_TRACE : EXIT_DEAD;
1113	if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1114		return 0;
1115	/*
1116	 * We own this thread, nobody else can reap it.
1117	 */
1118	read_unlock(&tasklist_lock);
1119	sched_annotate_sleep();
1120
1121	/*
1122	 * Check thread_group_leader() to exclude the traced sub-threads.
1123	 */
1124	if (state == EXIT_DEAD && thread_group_leader(p)) {
1125		struct signal_struct *sig = p->signal;
1126		struct signal_struct *psig = current->signal;
1127		unsigned long maxrss;
1128		u64 tgutime, tgstime;
1129
1130		/*
1131		 * The resource counters for the group leader are in its
1132		 * own task_struct.  Those for dead threads in the group
1133		 * are in its signal_struct, as are those for the child
1134		 * processes it has previously reaped.  All these
1135		 * accumulate in the parent's signal_struct c* fields.
1136		 *
1137		 * We don't bother to take a lock here to protect these
1138		 * p->signal fields because the whole thread group is dead
1139		 * and nobody can change them.
1140		 *
1141		 * psig->stats_lock also protects us from our sub-threads
1142		 * which can reap other children at the same time. Until
1143		 * we change k_getrusage()-like users to rely on this lock
1144		 * we have to take ->siglock as well.
1145		 *
1146		 * We use thread_group_cputime_adjusted() to get times for
1147		 * the thread group, which consolidates times for all threads
1148		 * in the group including the group leader.
1149		 */
1150		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1151		spin_lock_irq(&current->sighand->siglock);
1152		write_seqlock(&psig->stats_lock);
1153		psig->cutime += tgutime + sig->cutime;
1154		psig->cstime += tgstime + sig->cstime;
1155		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1156		psig->cmin_flt +=
1157			p->min_flt + sig->min_flt + sig->cmin_flt;
1158		psig->cmaj_flt +=
1159			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1160		psig->cnvcsw +=
1161			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1162		psig->cnivcsw +=
1163			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1164		psig->cinblock +=
1165			task_io_get_inblock(p) +
1166			sig->inblock + sig->cinblock;
1167		psig->coublock +=
1168			task_io_get_oublock(p) +
1169			sig->oublock + sig->coublock;
1170		maxrss = max(sig->maxrss, sig->cmaxrss);
1171		if (psig->cmaxrss < maxrss)
1172			psig->cmaxrss = maxrss;
1173		task_io_accounting_add(&psig->ioac, &p->ioac);
1174		task_io_accounting_add(&psig->ioac, &sig->ioac);
1175		write_sequnlock(&psig->stats_lock);
1176		spin_unlock_irq(&current->sighand->siglock);
1177	}
1178
1179	if (wo->wo_rusage)
1180		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1181	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1182		? p->signal->group_exit_code : p->exit_code;
1183	wo->wo_stat = status;
1184
1185	if (state == EXIT_TRACE) {
1186		write_lock_irq(&tasklist_lock);
1187		/* We dropped tasklist, ptracer could die and untrace */
1188		ptrace_unlink(p);
1189
1190		/* If parent wants a zombie, don't release it now */
1191		state = EXIT_ZOMBIE;
1192		if (do_notify_parent(p, p->exit_signal))
1193			state = EXIT_DEAD;
1194		p->exit_state = state;
1195		write_unlock_irq(&tasklist_lock);
1196	}
1197	if (state == EXIT_DEAD)
1198		release_task(p);
1199
1200out_info:
1201	infop = wo->wo_info;
1202	if (infop) {
1203		if ((status & 0x7f) == 0) {
1204			infop->cause = CLD_EXITED;
1205			infop->status = status >> 8;
1206		} else {
1207			infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1208			infop->status = status & 0x7f;
1209		}
1210		infop->pid = pid;
1211		infop->uid = uid;
1212	}
1213
1214	return pid;
1215}
1216
1217static int *task_stopped_code(struct task_struct *p, bool ptrace)
1218{
1219	if (ptrace) {
1220		if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1221			return &p->exit_code;
1222	} else {
1223		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1224			return &p->signal->group_exit_code;
1225	}
1226	return NULL;
1227}
1228
1229/**
1230 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1231 * @wo: wait options
1232 * @ptrace: is the wait for ptrace
1233 * @p: task to wait for
1234 *
1235 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1236 *
1237 * CONTEXT:
1238 * read_lock(&tasklist_lock), which is released if return value is
1239 * non-zero.  Also, grabs and releases @p->sighand->siglock.
1240 *
1241 * RETURNS:
1242 * 0 if wait condition didn't exist and search for other wait conditions
1243 * should continue.  Non-zero return, -errno on failure and @p's pid on
1244 * success, implies that tasklist_lock is released and wait condition
1245 * search should terminate.
1246 */
1247static int wait_task_stopped(struct wait_opts *wo,
1248				int ptrace, struct task_struct *p)
1249{
1250	struct waitid_info *infop;
1251	int exit_code, *p_code, why;
1252	uid_t uid = 0; /* unneeded, required by compiler */
1253	pid_t pid;
1254
1255	/*
1256	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1257	 */
1258	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1259		return 0;
1260
1261	if (!task_stopped_code(p, ptrace))
1262		return 0;
1263
1264	exit_code = 0;
1265	spin_lock_irq(&p->sighand->siglock);
1266
1267	p_code = task_stopped_code(p, ptrace);
1268	if (unlikely(!p_code))
1269		goto unlock_sig;
1270
1271	exit_code = *p_code;
1272	if (!exit_code)
1273		goto unlock_sig;
1274
1275	if (!unlikely(wo->wo_flags & WNOWAIT))
1276		*p_code = 0;
1277
1278	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1279unlock_sig:
1280	spin_unlock_irq(&p->sighand->siglock);
1281	if (!exit_code)
1282		return 0;
1283
1284	/*
1285	 * Now we are pretty sure this task is interesting.
1286	 * Make sure it doesn't get reaped out from under us while we
1287	 * give up the lock and then examine it below.  We don't want to
1288	 * keep holding onto the tasklist_lock while we call getrusage and
1289	 * possibly take page faults for user memory.
1290	 */
1291	get_task_struct(p);
1292	pid = task_pid_vnr(p);
1293	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1294	read_unlock(&tasklist_lock);
1295	sched_annotate_sleep();
1296	if (wo->wo_rusage)
1297		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1298	put_task_struct(p);
1299
1300	if (likely(!(wo->wo_flags & WNOWAIT)))
1301		wo->wo_stat = (exit_code << 8) | 0x7f;
1302
1303	infop = wo->wo_info;
1304	if (infop) {
1305		infop->cause = why;
1306		infop->status = exit_code;
1307		infop->pid = pid;
1308		infop->uid = uid;
1309	}
1310	return pid;
1311}
1312
1313/*
1314 * Handle do_wait work for one task in a live, non-stopped state.
1315 * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1316 * the lock and this task is uninteresting.  If we return nonzero, we have
1317 * released the lock and the system call should return.
1318 */
1319static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1320{
1321	struct waitid_info *infop;
1322	pid_t pid;
1323	uid_t uid;
1324
1325	if (!unlikely(wo->wo_flags & WCONTINUED))
1326		return 0;
1327
1328	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1329		return 0;
1330
1331	spin_lock_irq(&p->sighand->siglock);
1332	/* Re-check with the lock held.  */
1333	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1334		spin_unlock_irq(&p->sighand->siglock);
1335		return 0;
1336	}
1337	if (!unlikely(wo->wo_flags & WNOWAIT))
1338		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1339	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1340	spin_unlock_irq(&p->sighand->siglock);
1341
1342	pid = task_pid_vnr(p);
1343	get_task_struct(p);
1344	read_unlock(&tasklist_lock);
1345	sched_annotate_sleep();
1346	if (wo->wo_rusage)
1347		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1348	put_task_struct(p);
1349
1350	infop = wo->wo_info;
1351	if (!infop) {
1352		wo->wo_stat = 0xffff;
1353	} else {
1354		infop->cause = CLD_CONTINUED;
1355		infop->pid = pid;
1356		infop->uid = uid;
1357		infop->status = SIGCONT;
1358	}
1359	return pid;
1360}
1361
1362/*
1363 * Consider @p for a wait by @parent.
1364 *
1365 * -ECHILD should be in ->notask_error before the first call.
1366 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1367 * Returns zero if the search for a child should continue;
1368 * then ->notask_error is 0 if @p is an eligible child,
1369 * or still -ECHILD.
1370 */
1371static int wait_consider_task(struct wait_opts *wo, int ptrace,
1372				struct task_struct *p)
1373{
1374	/*
1375	 * We can race with wait_task_zombie() from another thread.
1376	 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1377	 * can't confuse the checks below.
1378	 */
1379	int exit_state = READ_ONCE(p->exit_state);
1380	int ret;
1381
1382	if (unlikely(exit_state == EXIT_DEAD))
1383		return 0;
1384
1385	ret = eligible_child(wo, ptrace, p);
1386	if (!ret)
1387		return ret;
1388
1389	if (unlikely(exit_state == EXIT_TRACE)) {
1390		/*
1391		 * ptrace == 0 means we are the natural parent. In this case
1392		 * we should clear notask_error, debugger will notify us.
1393		 */
1394		if (likely(!ptrace))
1395			wo->notask_error = 0;
1396		return 0;
1397	}
1398
1399	if (likely(!ptrace) && unlikely(p->ptrace)) {
1400		/*
1401		 * If it is traced by its real parent's group, just pretend
1402		 * the caller is ptrace_do_wait() and reap this child if it
1403		 * is zombie.
1404		 *
1405		 * This also hides group stop state from real parent; otherwise
1406		 * a single stop can be reported twice as group and ptrace stop.
1407		 * If a ptracer wants to distinguish these two events for its
1408		 * own children it should create a separate process which takes
1409		 * the role of real parent.
1410		 */
1411		if (!ptrace_reparented(p))
1412			ptrace = 1;
1413	}
1414
1415	/* slay zombie? */
1416	if (exit_state == EXIT_ZOMBIE) {
1417		/* we don't reap group leaders with subthreads */
1418		if (!delay_group_leader(p)) {
1419			/*
1420			 * A zombie ptracee is only visible to its ptracer.
1421			 * Notification and reaping will be cascaded to the
1422			 * real parent when the ptracer detaches.
1423			 */
1424			if (unlikely(ptrace) || likely(!p->ptrace))
1425				return wait_task_zombie(wo, p);
1426		}
1427
1428		/*
1429		 * Allow access to stopped/continued state via zombie by
1430		 * falling through.  Clearing of notask_error is complex.
1431		 *
1432		 * When !@ptrace:
1433		 *
1434		 * If WEXITED is set, notask_error should naturally be
1435		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1436		 * so, if there are live subthreads, there are events to
1437		 * wait for.  If all subthreads are dead, it's still safe
1438		 * to clear - this function will be called again in finite
1439		 * amount time once all the subthreads are released and
1440		 * will then return without clearing.
1441		 *
1442		 * When @ptrace:
1443		 *
1444		 * Stopped state is per-task and thus can't change once the
1445		 * target task dies.  Only continued and exited can happen.
1446		 * Clear notask_error if WCONTINUED | WEXITED.
1447		 */
1448		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1449			wo->notask_error = 0;
1450	} else {
1451		/*
1452		 * @p is alive and it's gonna stop, continue or exit, so
1453		 * there always is something to wait for.
1454		 */
1455		wo->notask_error = 0;
1456	}
1457
1458	/*
1459	 * Wait for stopped.  Depending on @ptrace, different stopped state
1460	 * is used and the two don't interact with each other.
1461	 */
1462	ret = wait_task_stopped(wo, ptrace, p);
1463	if (ret)
1464		return ret;
1465
1466	/*
1467	 * Wait for continued.  There's only one continued state and the
1468	 * ptracer can consume it which can confuse the real parent.  Don't
1469	 * use WCONTINUED from ptracer.  You don't need or want it.
1470	 */
1471	return wait_task_continued(wo, p);
1472}
1473
1474/*
1475 * Do the work of do_wait() for one thread in the group, @tsk.
1476 *
1477 * -ECHILD should be in ->notask_error before the first call.
1478 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1479 * Returns zero if the search for a child should continue; then
1480 * ->notask_error is 0 if there were any eligible children,
1481 * or still -ECHILD.
1482 */
1483static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1484{
1485	struct task_struct *p;
1486
1487	list_for_each_entry(p, &tsk->children, sibling) {
1488		int ret = wait_consider_task(wo, 0, p);
1489
1490		if (ret)
1491			return ret;
1492	}
1493
1494	return 0;
1495}
1496
1497static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1498{
1499	struct task_struct *p;
1500
1501	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1502		int ret = wait_consider_task(wo, 1, p);
1503
1504		if (ret)
1505			return ret;
1506	}
1507
1508	return 0;
1509}
1510
1511static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1512				int sync, void *key)
1513{
1514	struct wait_opts *wo = container_of(wait, struct wait_opts,
1515						child_wait);
1516	struct task_struct *p = key;
1517
1518	if (!eligible_pid(wo, p))
1519		return 0;
1520
1521	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1522		return 0;
1523
1524	return default_wake_function(wait, mode, sync, key);
1525}
1526
1527void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1528{
1529	__wake_up_sync_key(&parent->signal->wait_chldexit,
1530			   TASK_INTERRUPTIBLE, p);
1531}
1532
1533static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1534				 struct task_struct *target)
1535{
1536	struct task_struct *parent =
1537		!ptrace ? target->real_parent : target->parent;
1538
1539	return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1540				     same_thread_group(current, parent));
1541}
1542
1543/*
1544 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1545 * and tracee lists to find the target task.
1546 */
1547static int do_wait_pid(struct wait_opts *wo)
1548{
1549	bool ptrace;
1550	struct task_struct *target;
1551	int retval;
1552
1553	ptrace = false;
1554	target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1555	if (target && is_effectively_child(wo, ptrace, target)) {
1556		retval = wait_consider_task(wo, ptrace, target);
1557		if (retval)
1558			return retval;
1559	}
1560
1561	ptrace = true;
1562	target = pid_task(wo->wo_pid, PIDTYPE_PID);
1563	if (target && target->ptrace &&
1564	    is_effectively_child(wo, ptrace, target)) {
1565		retval = wait_consider_task(wo, ptrace, target);
1566		if (retval)
1567			return retval;
1568	}
1569
1570	return 0;
1571}
1572
1573static long do_wait(struct wait_opts *wo)
1574{
1575	int retval;
1576
1577	trace_sched_process_wait(wo->wo_pid);
1578
1579	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1580	wo->child_wait.private = current;
1581	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1582repeat:
1583	/*
1584	 * If there is nothing that can match our criteria, just get out.
1585	 * We will clear ->notask_error to zero if we see any child that
1586	 * might later match our criteria, even if we are not able to reap
1587	 * it yet.
1588	 */
1589	wo->notask_error = -ECHILD;
1590	if ((wo->wo_type < PIDTYPE_MAX) &&
1591	   (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1592		goto notask;
1593
1594	set_current_state(TASK_INTERRUPTIBLE);
1595	read_lock(&tasklist_lock);
1596
1597	if (wo->wo_type == PIDTYPE_PID) {
1598		retval = do_wait_pid(wo);
1599		if (retval)
1600			goto end;
1601	} else {
1602		struct task_struct *tsk = current;
1603
1604		do {
1605			retval = do_wait_thread(wo, tsk);
1606			if (retval)
1607				goto end;
1608
1609			retval = ptrace_do_wait(wo, tsk);
1610			if (retval)
1611				goto end;
1612
1613			if (wo->wo_flags & __WNOTHREAD)
1614				break;
1615		} while_each_thread(current, tsk);
1616	}
1617	read_unlock(&tasklist_lock);
1618
1619notask:
1620	retval = wo->notask_error;
1621	if (!retval && !(wo->wo_flags & WNOHANG)) {
1622		retval = -ERESTARTSYS;
1623		if (!signal_pending(current)) {
1624			schedule();
1625			goto repeat;
1626		}
1627	}
1628end:
1629	__set_current_state(TASK_RUNNING);
1630	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1631	return retval;
1632}
1633
1634static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1635			  int options, struct rusage *ru)
1636{
1637	struct wait_opts wo;
1638	struct pid *pid = NULL;
1639	enum pid_type type;
1640	long ret;
1641	unsigned int f_flags = 0;
1642
1643	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1644			__WNOTHREAD|__WCLONE|__WALL))
1645		return -EINVAL;
1646	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1647		return -EINVAL;
1648
1649	switch (which) {
1650	case P_ALL:
1651		type = PIDTYPE_MAX;
1652		break;
1653	case P_PID:
1654		type = PIDTYPE_PID;
1655		if (upid <= 0)
1656			return -EINVAL;
1657
1658		pid = find_get_pid(upid);
1659		break;
1660	case P_PGID:
1661		type = PIDTYPE_PGID;
1662		if (upid < 0)
1663			return -EINVAL;
1664
1665		if (upid)
1666			pid = find_get_pid(upid);
1667		else
1668			pid = get_task_pid(current, PIDTYPE_PGID);
1669		break;
1670	case P_PIDFD:
1671		type = PIDTYPE_PID;
1672		if (upid < 0)
1673			return -EINVAL;
1674
1675		pid = pidfd_get_pid(upid, &f_flags);
1676		if (IS_ERR(pid))
1677			return PTR_ERR(pid);
1678
1679		break;
1680	default:
1681		return -EINVAL;
1682	}
1683
1684	wo.wo_type	= type;
1685	wo.wo_pid	= pid;
1686	wo.wo_flags	= options;
1687	wo.wo_info	= infop;
1688	wo.wo_rusage	= ru;
1689	if (f_flags & O_NONBLOCK)
1690		wo.wo_flags |= WNOHANG;
1691
1692	ret = do_wait(&wo);
1693	if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1694		ret = -EAGAIN;
1695
1696	put_pid(pid);
1697	return ret;
1698}
1699
1700SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1701		infop, int, options, struct rusage __user *, ru)
1702{
1703	struct rusage r;
1704	struct waitid_info info = {.status = 0};
1705	long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1706	int signo = 0;
1707
1708	if (err > 0) {
1709		signo = SIGCHLD;
1710		err = 0;
1711		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1712			return -EFAULT;
1713	}
1714	if (!infop)
1715		return err;
1716
1717	if (!user_write_access_begin(infop, sizeof(*infop)))
1718		return -EFAULT;
1719
1720	unsafe_put_user(signo, &infop->si_signo, Efault);
1721	unsafe_put_user(0, &infop->si_errno, Efault);
1722	unsafe_put_user(info.cause, &infop->si_code, Efault);
1723	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1724	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1725	unsafe_put_user(info.status, &infop->si_status, Efault);
1726	user_write_access_end();
1727	return err;
1728Efault:
1729	user_write_access_end();
1730	return -EFAULT;
1731}
1732
1733long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1734		  struct rusage *ru)
1735{
1736	struct wait_opts wo;
1737	struct pid *pid = NULL;
1738	enum pid_type type;
1739	long ret;
1740
1741	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1742			__WNOTHREAD|__WCLONE|__WALL))
1743		return -EINVAL;
1744
1745	/* -INT_MIN is not defined */
1746	if (upid == INT_MIN)
1747		return -ESRCH;
1748
1749	if (upid == -1)
1750		type = PIDTYPE_MAX;
1751	else if (upid < 0) {
1752		type = PIDTYPE_PGID;
1753		pid = find_get_pid(-upid);
1754	} else if (upid == 0) {
1755		type = PIDTYPE_PGID;
1756		pid = get_task_pid(current, PIDTYPE_PGID);
1757	} else /* upid > 0 */ {
1758		type = PIDTYPE_PID;
1759		pid = find_get_pid(upid);
1760	}
1761
1762	wo.wo_type	= type;
1763	wo.wo_pid	= pid;
1764	wo.wo_flags	= options | WEXITED;
1765	wo.wo_info	= NULL;
1766	wo.wo_stat	= 0;
1767	wo.wo_rusage	= ru;
1768	ret = do_wait(&wo);
1769	put_pid(pid);
1770	if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1771		ret = -EFAULT;
1772
1773	return ret;
1774}
1775
1776int kernel_wait(pid_t pid, int *stat)
1777{
1778	struct wait_opts wo = {
1779		.wo_type	= PIDTYPE_PID,
1780		.wo_pid		= find_get_pid(pid),
1781		.wo_flags	= WEXITED,
1782	};
1783	int ret;
1784
1785	ret = do_wait(&wo);
1786	if (ret > 0 && wo.wo_stat)
1787		*stat = wo.wo_stat;
1788	put_pid(wo.wo_pid);
1789	return ret;
1790}
1791
1792SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1793		int, options, struct rusage __user *, ru)
1794{
1795	struct rusage r;
1796	long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1797
1798	if (err > 0) {
1799		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1800			return -EFAULT;
1801	}
1802	return err;
1803}
1804
1805#ifdef __ARCH_WANT_SYS_WAITPID
1806
1807/*
1808 * sys_waitpid() remains for compatibility. waitpid() should be
1809 * implemented by calling sys_wait4() from libc.a.
1810 */
1811SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1812{
1813	return kernel_wait4(pid, stat_addr, options, NULL);
1814}
1815
1816#endif
1817
1818#ifdef CONFIG_COMPAT
1819COMPAT_SYSCALL_DEFINE4(wait4,
1820	compat_pid_t, pid,
1821	compat_uint_t __user *, stat_addr,
1822	int, options,
1823	struct compat_rusage __user *, ru)
1824{
1825	struct rusage r;
1826	long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1827	if (err > 0) {
1828		if (ru && put_compat_rusage(&r, ru))
1829			return -EFAULT;
1830	}
1831	return err;
1832}
1833
1834COMPAT_SYSCALL_DEFINE5(waitid,
1835		int, which, compat_pid_t, pid,
1836		struct compat_siginfo __user *, infop, int, options,
1837		struct compat_rusage __user *, uru)
1838{
1839	struct rusage ru;
1840	struct waitid_info info = {.status = 0};
1841	long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1842	int signo = 0;
1843	if (err > 0) {
1844		signo = SIGCHLD;
1845		err = 0;
1846		if (uru) {
1847			/* kernel_waitid() overwrites everything in ru */
1848			if (COMPAT_USE_64BIT_TIME)
1849				err = copy_to_user(uru, &ru, sizeof(ru));
1850			else
1851				err = put_compat_rusage(&ru, uru);
1852			if (err)
1853				return -EFAULT;
1854		}
1855	}
1856
1857	if (!infop)
1858		return err;
1859
1860	if (!user_write_access_begin(infop, sizeof(*infop)))
1861		return -EFAULT;
1862
1863	unsafe_put_user(signo, &infop->si_signo, Efault);
1864	unsafe_put_user(0, &infop->si_errno, Efault);
1865	unsafe_put_user(info.cause, &infop->si_code, Efault);
1866	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1867	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1868	unsafe_put_user(info.status, &infop->si_status, Efault);
1869	user_write_access_end();
1870	return err;
1871Efault:
1872	user_write_access_end();
1873	return -EFAULT;
1874}
1875#endif
1876
1877/**
1878 * thread_group_exited - check that a thread group has exited
1879 * @pid: tgid of thread group to be checked.
1880 *
1881 * Test if the thread group represented by tgid has exited (all
1882 * threads are zombies, dead or completely gone).
1883 *
1884 * Return: true if the thread group has exited. false otherwise.
1885 */
1886bool thread_group_exited(struct pid *pid)
1887{
1888	struct task_struct *task;
1889	bool exited;
1890
1891	rcu_read_lock();
1892	task = pid_task(pid, PIDTYPE_PID);
1893	exited = !task ||
1894		(READ_ONCE(task->exit_state) && thread_group_empty(task));
1895	rcu_read_unlock();
1896
1897	return exited;
1898}
1899EXPORT_SYMBOL(thread_group_exited);
1900
1901__weak void abort(void)
1902{
1903	BUG();
1904
1905	/* if that doesn't kill us, halt */
1906	panic("Oops failed to kill thread");
1907}
1908EXPORT_SYMBOL(abort);