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